Methods and systems for predicting treatment outcomes, patient selection and personalized denervation therapy

Abstract

A method performed by one or more computers may include receiving patient specific data; producing a set of predictive features that characterize the patient specific data; performing one or more simulations of a body of the patient based on the patient specific data to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient; processing the set of predictive features using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

Description

METHODS AND SYSTEMS FOR PREDICTING TREATMENT OUTCOMES, PATIENT SELECTION AND PERSONALIZED DENERVATION THERAPYCROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Application Nos. 63 / 553,592, filed February 14, 2024, and 63 / 701,350, filed September 30, 2024, the contents of each are incorporated by reference herein.BACKGROUND

[0002] This specification relates to processing data using machine learning models.

[0003] Machine learning models receive an input and generate an output, e.g., a predicted output, based on the received input. Some machine learning models are parametric models and generate the output based on the received input and on values of the parameters of the model.

[0004] Some machine learning models are deep models that employ multiple layers of models to generate an output for a received input. For example, a deep neural network is a deep machine learning model that includes an output layer and one or more hidden layers that each apply a non-linear transformation to a received input to generate an output.SUMMARY

[0005] This specification describes systems and methods implemented as computer programs on one or more computers in one or more locations that may predict effects of denervation procedures for a patient based on patient specific data.

[0006] According to one aspect, there is provided a method performed by one or more computers that includes: receiving patient specific data; producing a set of predictive features that characterize the patient specific data; performing one or more simulations of a body of the patient based on the patient specific data to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient; processing the set of predictive features and the one or more simulation results using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

[0007] According to one aspect, there is provided a method performed by one or more computers, comprising: receiving patient specific data, wherein the patient specific datacomprises continuously-monitored patient physiological data measured over a continuous time period; producing a set of predictive features that characterize the patient specific data, wherein the set of predictive features comprises one or more aggregation measures computed from the continuously-monitored patient physiological data measured over the continuous time period; processing at least the set of predictive features using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on a body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

[0008] According to one aspect, there is provided a method performed by one or more computers, comprising: acquiring one or more two-dimensional (2D) medical images of an interior of a patient; for each of the 2D medical images, processing an input generated from the 2D medical image using a neural network to generate a respective segmentation of the 2D medical image that identifies pixels of the 2D medical image that depict one or more blood vessels; determining features of the one or more blood vessels from the respective segmentations; and generating treatment information using the features of the one or more blood vessels; and providing the treatment information for display to a clinician.

[0009] According to one aspect, there is provided a method performed by one or more computers, comprising: acquiring a respective 2D medical image of an interior of a patient at each of a plurality of angles; generating a respective segmentation of the 2D medical image that identifies pixels of the 2D medical image that depict one or more blood vessels; generating, from the respective segmentations of the 2D medical images, a 3D model of a vasculature of the patient at a volume of interest; generating information about recommended treatment based on the 3D model, and displaying the information about recommended treatment.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] FIG. 1 is a schematic drawing of an environment in which denervation treatment predictions, based on patient specific medical data, are made for a particular patient and provided to a clinician who may treat that patient.

[0011] FIG. 2 is a block diagram of an exemplary denervation prediction system.

[0012] FIG. 3 is a flow diagram of an exemplary process for predicting denervation procedure effects.

[0013] FIG. 4 is a flow diagram of an exemplary process for simulating a denervation procedure.

[0014] FIG. 5 is a flow diagram of an exemplary process for producing denervation treatment recommendations .

[0015] FIG. 6A is an exemplary tissue treatment system.

[0016] FIG. 6B is an exemplary embodiment of a catheter of the tissue treatment system of FIG. 6A.

[0017] FIG. 7 is a block diagram of an exemplary controller that may be included in the tissue treatment system of FIG. 6A.

[0018] FIG. 8 is a flow diagram of an exemplary process for producing treatment recommendations before, during, and after a denervation treatment.

[0019] FIG. 9A is a flow diagram of an exemplary process for training a denervation procedure prediction model.

[0020] FIG. 9B shows an example of dynamic training of the denervation procedure prediction model.

[0021] FIG. 10 is a first plot illustrating the performance of an exemplary implementation of the described techniques.

[0022] FIG. 11 is a second plot further illustrating the performance of the exemplary implementation of the described techniques.

[0023] FIG. 12 is a flow diagram of an exemplary process for generating treatment information from 2D medical images.

[0024] FIG. 13 shows an example of treatment information being presented.

[0025] FIG. 14 is a flow diagram of an exemplary process for training a neural network to perform segmentation of blood vessels.

[0026] FIG. 15 shows an exemplary process for generating a 3D model of the vasculature from 2D medical images.

[0027] FIG. 16 illustrates how different 2D images of a volume, taken at different angles, may be combined to provide a 3D image of a blood vessel or vessel within that volume.

[0028] Like reference numbers and designations in the various drawings indicate like elements.DETAILED DESCRIPTION

[0029] Particular embodiments of the subject matter described in this specification may be implemented so as to realize one or more of the following advantages.

[0030] The described systems may use physical simulations of denervation procedures as part of predicting an effectiveness for the denervation procedures.

[0031] In some examples, the system may perform the simulations based on patient specific data. For example, the system may simulate an ablation of tissue within or associated with an organ of the patient as part of a simulated denervation procedure and may simulate changes to blood flow through an artery of the patient and neurological responses in the patient resulting from the simulated denervation procedure based on medical images of one or more organs, blood vessels, nerves, and / or other anatomical structures of the patient and on physiological data measured for the patient. By using these simulations, the described systems may produce more accurate predictions of denervation procedure effects based on the specific anatomy and physiology of the patient.

[0032] Performing multiple denervation procedures to different organs of the patient may increase the efficacy of one or more denervation procedures. The described systems may determine the effects of multiple denervation procedures for the same denervation treatment (e.g., to different target organs for the procedures and / or to different vascular locations associated with the target organs). The described systems may use the simulated procedures to determine denervation procedures that optimize the predicted effectiveness. The described systems may also determine predicted one or more physiological effects (e.g., changes in blood pressure (systolic, diastolic, average), heart rate, blood and / or hepatic glucose level, glucagon level, insulin level, blood and / or tissue norepinephrine level, markers of inflammation, estrogen, cytokines, vascular muscle tone, triglyceride level, blood C peptide level, ejection fraction, ventricular wall dimensions, cholesterol level, including high-density lipoprotein (HDL) levels, low-density lipoprotein (LDL) levels, very-low-density lipoprotein (VLDL) levels, changes in sympathetic drive) based on the simulated procedures. The described systems may therefore provide numerous treatment recommendations and procedure suggestions to assist a clinician of the patient determine a most effective denervation treatment for the patient.

[0033] In some other examples, even if the system does not make use of simulated procedures, by making predictions using features determined based on the slope of continuously-monitored physiological data obtained for a given patient, the described systems may more effectively make predictions relating to denervation procedure(s) for the given patient. In particular, providing such features as input to a machine learning model provides the model with additional information that the model may incorporate to improve the accuracy of the predictions generated by the model.

[0034] The described systems may provide predictions at different times relative to a denervation treatment for the patient. For example, the described system may predict effects of denervation procedures for a planned denervation treatment based on patient specific data received before the planned treatment. As another example, the described system may predict effects of denervation procedures for an ongoing denervation treatment based on patient specific data received during the treatment. As another example, the described system may receive patient specific data received after a completed denervation treatment and may provide follow-up recommendations to improve the efficacy of the treatment. The described systems may therefore provide procedure predictions and recommendations for every phase of a denervation treatment of a patient.

[0035] The described systems may therefore provide treatment predictions and suggestions to better assist clinicians in deciding a denervation treatment for the patient. The described systems may more accurately determine whether denervation treatment is required and may therefore help avoid performing unnecessary procedures. Additionally, the described systems may determine suggested procedures to assist the clinician select the most effective denervation treatment course for the patient.

[0036] This specification also describes techniques for generating treatment information, 3D blood vessel geometry, or both from one or more 2D images of the body of the patient. By providing this information to a clinician, the clinician may more effectively plan and perform a treatment, e.g., one or more denervation procedures. Moreover, as will be described below, providing this information to the clinician may result in a reduced need for contrast injection into the body of the patient, reducing the risk to the patient in performing any given procedure.

[0037] The details of one or more embodiments of the subject matter of this specification are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.

[0038] These features and other features are described in more detail below.

[0039] There is no substitute for the training and experience of a trained medical professional. The education and experience of a clinician enables them to evaluate a candidate for denervation, make a determination whether that patient would benefit from denervation, and consider how they would perform that denervation. However, even the most experienced physician may not make a perfect decision every time. Consequently, there is a need to improve the selection of patients for treatment with denervation, and to improve the likelihood of successful treatment of those patients with a denervation procedure.

[0040] Referring to FIG. 1 , a system may include a denervation treatment prediction system 100, which may process patient specific data 102 to generate denervation treatment predictions 106 for a patient 104. The denervation treatment prediction system 100 may process patient specific data 102 to provide a clinician 108 or other user with the overall health status of a patient and / or potential for response to any denervation. The denervation treatment predictions 106 may assist a clinician 108 in determining the effectiveness of one or more denervation procedures 110 for the patient 104. The denervation treatment predictions 106 may include, for example, predicted physiological changes, such as, but not limited to, changes in blood pressure, heart rate, tissue damage, and the like; predicted effectiveness, such as, but not limited to, likelihoods of success, categorical classifications, and the like; and / or suggested procedure locations, treatment recommendations, and the like, that may be used by the clinician 108 before, during, or after a denervation treatment to determine the effectiveness of the denervation procedures 110 for the patient 104 and / or provide a post-denervation treatment plan.

[0041] In some implementations, the denervation treatment prediction system 100 may make use of data from a corpus of patient population data 111 to generate the treatment predictions 106. The corpus of patient population data 111 stores data relating to an overall patient population that is made up of multiple patients. For example, the corpus of patient population data 111 can store statistics relating to blood flow across the patient population, data identifying known ablation locations that have been successful across the patient population, and so on. Making use of the corpus of patient population data 111 to generate the treatment predictions 106 will be described in more detail below.

[0042] The patient specific data 102 may include any of a variety of data that describes aspects of the health or body of the patient 104. For example, the patient specific data 102 may include anatomical data characterizing an internal structure of the body of the patient 104, e.g., medical images that illustrate the locations of one or more various tissues, organs, bones, nerves, blood vessels, structures, and so on within the body of the patient 104. As a particular example, anatomical data for the patient 104 may include any of a variety of types of medical images for the patient 104, e.g., data from a CT scan, data from an MRI scan, data from an X-Ray, data from an angiogram, data from a PET scan, data from an ultrasound, and so on. The patient specific data 102 may include data collected from any of a variety of appropriate medical imaging systems 114, e.g., a CT scan machine, an MRI scan machine, an X-ray machine, a PET scan machine, an ultrasound imager, and so on.

[0043] The patient specific data 102 may include physiological data characterizing processes within the body of the patient 104, such as but not limited to, blood pressure (systolic and / or diastolic), isolated systolic blood pressure, isolated diastolic blood pressure, daytime ambulatory blood pressure, twenty-four hour ambulatory blood pressure, home blood pressure, office blood pressure, nocturnal blood pressure, pulse pressure, onset date and / or duration of hypertension diagnosis, variability of nocturnal blood pressure, orthostatic blood pressure, arterial wall stiffness / calcification, ambulatory arterial stiffness index, central aortic blood pressure, pulse wave reflection, aortic calcification, sleep duration, sleep quality, time that sleep began, time that sleep ended, heart rate, nocturnal heart rate, daytime ambulatory heart rate, nervous system activity, sleep apnea, skin sodium content, elevated plasma or cerebrospinal fluid NaCl concentration, metabolic activity, heart rate variability, organ function, body mass index (BMI), age, sex, race, abdominal obesity, abdominal circumference, weight, height, blood glucose level, blood alcohol level, blood hormone level, and / or any other useful measurable physiological data. As an example, the physiological data may include the results of any of a variety of lab tests for the patient 104, such as but not limited to blood work (i.e., one or more blood tests useful in determining one or more physiological and / or biochemical states of the patient 104), plasma renin activity, urinalysis, biopsy analysis, cytological analysis, eGFR, results indicating whether the patient 104 is perimenopausal or menopausal, and so on.

[0044] As another example, the patient specific data 102 may include one or more classes of one or more hypertension medications taken by the patient 104. Hypertension medications are typically divided into different classes based on their mechanisms of action; such classes include diuretics, beta-blockers, angiotensin-converting enzyme (ACE) inhibitors, angiotensin II receptor blockers (ARBs), calcium-channel blockers, alpha blockers, alpha-2 receptor agonists, and vasodilators. Diuretics, beta-blockers, angiotensin-converting enzyme (ACE) inhibitors, angiotensin II receptor blockers (ARBs), and calcium-channel blockers act on the central sympathetic nervous system. If a patient’s hypertension appears to be affected by a centrally acting hypertensive medication, this may be a factor that weighs in favor of a denervation procedure. The patient’s hypertension may be affected by the medication but the patient may have side effects that make the patient intolerant to such usage. Therefore, such historical patient data may be advantageously used to select the patient.

[0045] The patient specific data 102 may be collected in one or more ways: via at least one medical imaging system 114, via at least one physiological monitoring system 116, or directly from the patient 104 such as by obtaining the patient’s medical history, by discussion, and / orby direct measurement by a clinician 108 using a stethoscope, blood pressure cuff, and / or other suitable device(s). The medical history of the patient 104 may include one or more of: duration of hypertension, daytime systolic / diastolic blood pressure, nighttime systolic / diastolic blood pressure, baseline systolic / diastolic blood pressure, isolated systolic hypertension, isolated diastolic hypertension, presence of diabetes, heart rate, nocturnal heart rate, daytime ambulatory heart rate, presence of coronary artery disease, presence and severity of renal artery disease, age, sex, BMI, height, heart rate, electrocardiogram signal (ECG), arterial stiffness (measured by pulse wave velocity or otherwise), baroreflex sensitivity, skin sodium level, elevated plasma or cerebrospinal fluid NaCl concentration, serum concentrations of endothelial adhesion molecules (intercellular cell adhesion molecule- 1 (ICAM-1) and vascular cell adhesion molecule- 1 (VCAM-1), vascular endothelial growth factor and its soluble receptor fms-like tyrosine kinase-1 (sFLT-1), plaque composition (calcification, lipid content, fibrous plaque, etc.) in the vicinity of potential target ablation sites, image of intima-media of the blood vessel, estimated glomerular filtration rate (eGFR), relevant comorbidities, and type, dosage, duration of use, and number of hypertension medications taken.

[0046] The patient specific data 102 may be collected from one or more of a variety of physiological monitoring systems 116, such as but not limited to data from a blood pressure monitor, data from a heart rate monitor, data from an ECG machine, data from an EEG machine, data from an oximeter, data from a thermometer, data from a blood glucose monitor, data from one or more hormonal testing systems, and / or data from a respiration rate monitor. Physiological monitoring systems 116 may include a smartwatch, a smart ring, a smart phone, a wearable fitness monitor, exercise equipment at home or at a gym, a scale, a personal EKG monitor such as the Kardia Mobile® ECG monitor of AliveCor, Inc. in Mountain View, California and the ZIO® ECG monitor of iRhythm Technologies, Inc. in San Francisco, California, a pulse oximeter, one or more devices such as a gyroscope and / or accelerometer that monitor the patient’s orientation and movement, a continuous positive airway pressure (CPAP) machine, one or more devices previously implanted in a patient 104 such as a pacemaker, cardioverter, or cardioverter-defibrillator, and / or any other suitable device or devices.

[0047] The patient specific data 102 may include biomarkers, such as galectin-3, intercellular cell adhesion molecule-1 (ICAM-1), vascular cell adhesion molecule-1 (VCAM-1), and soluble receptor fms-like tyrosine kinase-1 (sFLT-1).

[0048] The patient specific data 102 may include data relating to hormone levels of the patient 104, e.g., one or more of estrogen, progesterone, or testosterone levels.

[0049] The patient specific data 102 may include cardiac baroreflex sensitivity, calculated by progressive elevation of systolic blood pressure during >3 heart beats where R-R intervals simultaneously prolong.

[0050] The patient specific data 102 may include parameters derived from data directly measured from or otherwise collected from the patient 104, such as but not limited to maxima of particular measurements, minima of particular measurements, and mean values of particular measurements. As another example, patient specific data 102 may include sleep time of a patient 104 that is calculated from the difference between a measured or reported time the patient 104 went to sleep and a measured or reported time the patient 104 awoke.

[0051] The patient specific data 102 may include additional data that may be relevant to predicting the outcome of a denervation treatment. For example, the patient specific data 102 may specify a height, a weight, an age, a sex, a race, and / or other data associated with the patient 104. As another example, the patient specific data 102 may include listings of medications prescribed to or taken by the patient 104. As another example, the patient specific data 102 may include number and / or class of medications prescribed to or taken by the patient 104. As another example, the patient specific data 102 may include comorbidities, such as but not limited to diabetes, atrial fibrillation or other arrhythmia, heart disease, anxiety, depression, obesity, hyperthyroidism, sleep apnea, and / or kidney disease. As another example, the patient specific data 102 may include prior hospitalizations for hypertensive crisis, prior myocardial infarction or cerebrovascular events, history of heart failure, systemic inflammatory disease, and / or history of chronic kidney disease. As another example, the patient specific data 102 may specify information reported by the patient 104, such as a pain index, a family history, a description of symptoms, dietary habits, smoking habits, recreational drug use, alcohol use, exercise habits, and so on. The patient specific data 102 may specify information reported by the clinician 108, such as an evaluation of a risk for the patient 104, treatment recommendations, the results of an examination of the patient 104, and so on. The patient specific data 102 may be acquired directly from the patient 104 or may be acquired from a database of stored patient information, depending on the characteristics of that patient specific data 102.

[0052] Each of the denervation procedures 110 is a medical procedure that results in the targeted destruction of certain nerves within the body of the patient 104. For example, each denervation procedure 110 may be an ultrasonic (e.g., unfocused ultrasonic), radio frequency, microwave, laser, chemical and / or other ablation of one or more certain target nerves within the body of the patient 104. Each denervation procedure 110 may be applied to at least onetarget organ of the patient 104 (e.g., a kidney, liver, intestine, spleen, stomach, duodenum, pancreas) and / or nerve tissue associated with at least one target organ of the patient 104, may be applied to at least one particular location for the procedure (e.g. , a particular location within and / or associated with the target organ), and may be applied with a particular intensity, power, and duration (which may be referred to as the dose) for the procedure. A denervation treatment may include multiple denervation procedures 110 for multiple locations associated with an organ, or for two or more different target organs. A denervation treatment may be performed by an energy emitter inserted intravascularly into the body of a patient 104, and the denervation procedure or procedures may be performed by emitting ultrasound, radio frequency, microwave, and / or laser energy outward from the energy emitter through a wall of a blood vessel into one or more nerves in and / or in proximity to the wall of that blood vessel.

[0053] The denervation treatment prediction system 100 may process the patient specific data 102 to produce the denervation treatment predictions 106 at any point in time relative to when a denervation treatment may be performed for the patient 104. For example, the patient specific data 102 may include patient specific data 102 obtained before a denervation treatment for the patient 104 and the denervation treatment prediction system 100 may produce denervation treatment predictions 106 for a planned denervation treatment of the patient 104. As another example, the patient specific data 102 may include patient specific data 102 obtained during a denervation treatment for the patient 104 and the denervation treatment prediction system 100 may produce denervation treatment predictions 106 to guide and assist an ongoing denervation treatment of the patient 104. As yet another example, the patient specific data 102 may include follow-up patient specific data 102 obtained after a denervation treatment for the patient 104 and the denervation treatment prediction system 100 may produce denervation treatment predictions 106 that provide evaluations of and / or follow-up recommendations for a completed denervation treatment of the patient 104.

[0054] The denervation treatment predictions 106 include data characterizing the predicted effects of the one or more denervation procedures 110 for display on a user device e.g., a monitor, display, computer, and / or smartphone accessible to the clinician 108 and / or other user). When the patient specific data 102 includes medical image data, the data characterizing the predicted effects of the one or more denervation procedures 110 may include overlays for the medical images that illustrate the predicted effects of the one or more denervation procedures 110.

[0055] The data characterizing the predicted effects of the one or more denervation procedures 110 may characterize the predicted effects of the one or more denervation procedures 110 byany of a variety of means to assist the clinician 108. For example, the data characterizing the predicted effects may characterize a distribution of the predicted effects of each of the denervation procedures 110. As another example, the data characterizing the predicted effects may characterize predicted physiological changes in the body of the patient 104 resulting from the denervation procedures 110. As another example, the data characterizing the predicted effects may characterize a predicted effectiveness of each of the denervation procedures 110. As particular examples, the data characterizing the predicted effects may include a predicted effectiveness score or a categorical effectiveness classification for each of the denervation procedures 110 based on predicted physiological changes in the body of the patient 104.

[0056] The data characterizing the predicted effects of the one or more denervation procedures 110 may characterize predicted effects of denervation procedures at different locations in the body of the patient 104. The denervation treatment prediction system 100 may determine one or more suggested denervation procedures for the patient 104 and may include data characterizing the suggested denervation procedures within the denervation treatment predictions 106. In particular, the denervation treatment prediction system 100 may determine one or more suggested target organs and / or one or more suggested locations for the suggested denervation procedures. In some implementations, the denervation treatment prediction system 100 may determine the suggested denervation procedures by optimizing a predicted effectiveness of the denervation procedures 110. In some implementations, the denervation treatment prediction system 100 may determine an indication of whether denervation treatment is recommended and may include an indication of whether denervation treatment is recommended within the denervation treatment predictions 106.

[0057] The denervation treatment prediction system 100 may determine various recommendations for assisting the clinician 108 at different points in time relative to when a denervation treatment may be performed for the patient 104. For example, before a planned denervation treatment, the denervation treatment prediction system 100 may determine one or more suggested denervation procedures for the planned denervation treatment. In some implementations, the denervation treatment prediction system 100 may determine an indication whether denervation treatment is recommended and may include an indication whether denervation treatment is recommended within the denervation treatment predictions 106.

[0058] As another example, during a denervation treatment, the denervation treatment prediction system 100 may include data characterizing suggested denervation procedures for the ongoing denervation treatment within the denervation treatment predictions 106. During a denervation treatment, the denervation treatment prediction system 100 may process patientspecific data 102 obtained during the denervation treatment (e.g., medical images and physiological data obtained while monitoring the patient 104) and may include within the denervation treatment predictions 106 data characterizing a predicted effectiveness of the denervation procedures 110 performed during the treatment. In some embodiments, the denervation treatment predictions 106 include a particular location or locations for performing a particular denervation treatment, and / or the characteristics of one or more doses for delivery into tissue of the patient 104, which a physician may modify in their professional judgment.

[0059] As another example, after a denervation treatment, the denervation treatment prediction system 100 may include data characterizing completed denervation procedures 110 within the denervation treatment predictions 106. For example, the denervation treatment prediction system 100 may determine an effectiveness of the completed denervation procedures 110. As another example, the denervation treatment prediction system 100 may determine suggested follow-up denervation procedures and include data characterizing the suggested follow-up procedures within the treatment predictions 106. As another, the denervation treatment prediction system 100 may determine an indication whether follow-up treatment is recommended and may include the indication whether follow-up treatment is recommended within the treatment predictions 106.

[0060] In some implementations, the denervation treatment prediction system 100 may determine recommendations for the patient 104 based on the predicted effects of the denervation procedures 110. For example, the denervation treatment prediction system 100 may include recommendations for the patient 104 to improve the effectiveness of a planned or completed denervation treatment. As a particular example, the denervation treatment prediction system 100 may include medication recommendations (e.g., types of medication and / or doses of particular medications) for the patient 104 to improve the effectiveness of the denervation treatment. As another example, after a particular denervation treatment has been performed on a specific patient 104, the denervation treatment prediction system 100 may automatically complete an electronic order for medication for the patient 104 for transmission to a pharmacy after that electronic order is reviewed and approved by a physician. As another particular example, the denervation treatment prediction system 100 may include lifestyle change recommendations for the patient 104 to improve the effectiveness of the denervation treatment. As another particular example, the denervation treatment prediction system 100 may include recommendations for the patient 104 may include medication recommendations (e.g., types of medication and / or doses of particular medications) and / or lifestyle change for the patient 104 in liu of, or prior to, a denervation treatment.

[0061] FIG. 2 shows an example denervation treatment prediction system 100. The denervation treatment prediction system 100 is an example of a system implemented as computer programs on one or more computers in one or more locations in which the systems, components, and techniques described below are implemented.

[0062] The denervation treatment prediction system 100 may include a simulation system 202 and a denervation procedure prediction model 204. The simulation system 202 may perform simulations of the body of the patient 104 based on the patient specific data 102 to produce one or more simulation results 206. The simulation system 202 is configured to produce simulation results 206 that are predictive of effects of denervation procedures being performed to the body of the patient 104. The denervation procedure prediction model 204 is configured to process the patient specific data 102 and the simulation results 206 to produce the treatment predictions 106. The operation of the denervation treatment prediction system 100 is explained in more detail with reference to FIG. 3 below.

[0063] The simulation system 202 may perform simulations of one or more denervation procedures being performed to the patient 104. The simulation system 202 may produce simulation results that include simulated physiological changes in the body of the patient 104 as a result of the simulated denervation procedures. In particular, the simulation system 202 may produce simulation results that include simulated blood pressure changes in the body of the particular patient 104 as a result of the simulated denervation procedures. The simulated denervation procedures may include a procedure that utilizes ultrasound energy for denervation, a procedure that utilizes RF energy for denervation, a procedure that utilizes microwave energy for denervation, a procedure that utilizes laser energy for denervation, and / or a procedure that utilizes a chemical for denervation.

[0064] The simulation system 202 may perform a variety of simulations that are predictive of effects of denervation procedures being performed to the body of the patient 104. As a particular example, the simulation system 202 may directly simulate a transfer of energy to tissue of one or more organs of the patient 104 for one or more simulated denervation procedures. As another example, the simulation system 202 may directly simulate blood flow through arteries of the patient 104 before and after one or more simulated denervation procedures. As another example, the simulation system 202 may simulate nervous system responses for the patient 104 before and after one or more simulated denervation procedures. The operation of the simulation system 202 is explained in more detail with reference to FIG. 4 below. In some implementations, denervation treatment prediction system 100 does not include the simulation system 202.

[0065] The denervation procedure prediction model 204 may be any of a variety of machine learning models appropriate for processing the patient specific data 102 and the simulation results 206 and producing the treatment predictions 106. For example, the denervation procedure prediction model 204 may be a machine learning model based on decision trees. As a particular example, the denervation procedure prediction model 204 may be a random forest model. As another particular example, the denervation procedure prediction model 204 may be an XGBoost model.

[0066] In some implementations, the denervation procedure prediction model 204 is a neural network. When the denervation procedure prediction model 204 is a neural network, the denervation procedure prediction model 204 may have any architecture suited to processing the patient specific data 102 and the simulation results 206. For example, when the patient specific data 102 or the simulation results 206 include medical images, the denervation procedure prediction model 204 may include sub-networks suitable for processing image data, such as but not limited to convolutional neural networks (CNNs), vision transformers (ViTs), capsule networks (CapsNets), and / or generative adversarial networks (GANs). As another example, when the patient specific data 102 or the simulation results 206 include time-series data, the denervation procedure prediction model 204 may include sub-networks suitable for processing time-series data, such as but not limited to recurrent neural networks (RNNs), transformers, and / or long short-term memory (LSTM) networks.

[0067] As part of processing the patient specific data 102, the denervation procedure prediction model 204 produces a set of predictive features that characterizes the patient specific data 102. The denervation procedure prediction model 204 may include sub-networks of a variety of appropriate architectures for producing features characterizing the patient specific data 102. For example, the denervation procedure prediction model 204 may include sub-networks appropriate for producing features that characterize images, such as but not limited to CNNs, ViTs, CapsNets, and / or GANs. As another example, the denervation procedure prediction model 204 may include sub-networks appropriate for producing features that characterize timeseries data, such as but not limited to RNNs and / or transformers. As a further example, when the patient specific data 102 includes medical images, the denervation procedure prediction model 204 may extract patient specific geometric and / or pathologic features from the medical image data.

[0068] In general, the denervation procedure prediction model 204 may employ any of a variety of techniques for producing features that characterize the patient specific data 102. For example, the denervation procedure prediction model 204 may perform dimensionalityreduction to produce features for some or all of the patient specific data 102. As another example, the denervation procedure prediction model 204 may apply an embedding to produce features for some or all of the patient specific data 102. As another example, the denervation procedure prediction model 204 may compress some or all of the patient specific data 102 to produce features for the patient specific data 102. When the denervation procedure prediction model 204 produces features characterizing the patient specific data 102 using sub-networks, these sub-networks may be trained to perform feature extraction as part of training the denervation procedure prediction model 204.

[0069] The denervation procedure prediction model 204 may include a variety of sub-networks for generating particular outputs included within the treatment predictions 106. The types of outputs included within the treatment predictions 106 are described in more detail with reference to FIG. 5 below.

[0070] Generally, prior to using the denervation procedure prediction model 204, the denervation treatment prediction system 100 trains the denervation procedure prediction model 204 on a training data set such as but not limited to the corpus of patient population data 111.

[0071] The training data set includes data for a set of example patients 104. For example, the data for the example patients 104 can be determined using logged data that stores the outcome of denervation procedures previously performed on the example patients 104.

[0072] For example, for each of the example patients 104, the training data includes: (i) example patient specific data 102 (such as but not limited to physiological data and / or medical images) for the example patient 104, (ii) data identifying one or more example denervation procedures and characteristics thereof (such as but not limited to, target organ(s), location(s), intensity, and / or dose,) for the example patient 104, e.g., one or more denervation procedures that were actually performed on the example patient 104, and (iii) one or more target predictions for the example patient 104. The target predictions may include observed procedure effects, such as but not limited to physiological changes and / or procedure success. That is, the target predictions may have been generated from outcomes of the denervation procedures performed on the example patient 104 or measurements taken after the denervation procedures were performed or both.

[0073] After training the denervation procedure prediction model 204 on the training data set, the denervation treatment prediction system 100 deploys the denervation procedure prediction model 204 for use in making predictions for future patients 104, as described above. Optionally, the denervation treatment prediction system 100 can further train (“fine-tune”) thedenervation procedure prediction model 204 as additional data becomes available after predictions are made by the denervation prediction model 204.

[0074] Training the denervation procedure prediction model 204 is described in more detail below with reference to FIG. 9 A and 9B.

[0075] FIG. 3 is a flow diagram of an example process 300 for predicting denervation procedure effects. For convenience, the process 300 will be described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1, appropriately programmed in accordance with this specification, may perform the process 300.

[0076] At block 302, the denervation treatment prediction system 100 receives patient specific data 102 for a patient 104. In some implementations, the patient specific data 102 includes anatomical data characterizing at least one internal structure of the body of the patient 104, such as but not limited to a location and / or shape of one or more tissues, organs, bones, nerves, blood vessels, and / or structures. The anatomical data may include medical images of the body of the patient 104. In some implementations, the patient specific data 102 includes physiological data characterizing processes within the body of the patient 104, such as but not limited to heart rate, blood pressure, orthostatic blood pressure, metabolic processes, and / or nervous system activity. The patient specific data 102 may include any item of data described above with regard to FIG. 1.

[0077] The patient specific data 102 may include additional information for the patient 104. For example, the patient specific data 102 may specify information such as a height, a weight, an age, a sex, a race, and / or other data associated with the patient 104. As another example, the patient specific data 102 may include listings of medications prescribed to or taken by the patient 104. As another example, the patient specific data 102 may specify information reported by the patient 104, such as a pain index, a family history, a description of symptoms, dietary habits, exercise habits, and so on. The patient specific data 102 may specify information reported by a clinician 108 of the patient 104, such as an evaluation of a risk for the patient 104, treatment recommendations, the results of an examination of the patient 104, and so on.

[0078] The denervation treatment prediction system 100 may receive the patient specific data 102 from any of a variety of sources. For example, the denervation treatment prediction system 100 may receive medical images from a medical imaging system such as but not limited to a CT scan machine, an MRI scan machine, an X-ray machine, a PET scan machine, and / or an ultrasound imager. As another example, the denervation treatment prediction system 100 may receive data from equipment monitoring physiological data for the patient 104, such as but notlimited to a blood pressure monitor, a heart rate monitor, an ECG machine, an EEG machine, an oximeter, a thermometer, and / or a respiration rate monitor. As an example, the equipment monitoring physiological data for the patient 104 may be wearable equipment or other equipment that is able to monitor physiological data for the patient 104 over a continuous period of time. The continuous period of time may be six hours, twelve hours, twenty hours, forty-eight hours, or one week. Examples of data received from equipment that continuously monitors physiological data for the patient 104 include ambulatory systolic blood pressure, ambulatory heart rate, and ambulatory pulse pressure. The denervation treatment prediction system 100 may also receive patient specific data 102 entered by a clinician 108 for the patient 104. The denervation treatment prediction system 100 may also receive prior patient data 112 stored as part of the patient’s medical records.

[0079] The denervation treatment prediction system 100 may receive the patient specific data obtained at any point in time relative to when a denervation treatment may be performed for the patient 104. For example, the patient specific data 102 may include patient specific data 102 obtained before a planned denervation treatment for the patient 104. As another example, the patient specific data 102 may include patient specific data 102 obtained during an ongoing denervation treatment for the patient 104. As yet another example, the patient specific data 102 may include follow-up patient specific data 102 obtained after a denervation treatment for the patient 104. The process of using the denervation treatment prediction system 100 before, during, and after a denervation treatment for the patient 104 is explained in more detail with reference to FIG. 6 below.

[0080] At block 304, the denervation treatment prediction system 100 produces a set of predictive features based on the patient specific data 102. The denervation treatment prediction system 100 may produce predictive features that characterize some or all of the received patient specific data 102. For example, when the denervation treatment prediction system 100 receives medical image data, the denervation treatment prediction system 100 may extract patient specific geometric and pathologic features from the medical image data.

[0081] As another example, the denervation treatment prediction system 100 may use continuously monitored patient specific data 102 that includes physiological data from the physiological monitoring system 116 to calculate predictive features. For example, the denervation treatment prediction system 100 may calculate the mean, maximum, minimum, standard deviation (i.e., variability), or other aggregation measure of one or more of the ambulatory blood pressure, heart rate, or pulse pressure over a continuous period of time. By way of example and not limitation, that continuous period of time may be six hours, twelvehours, twenty hours, twenty-four hours, forty-eight hours, or one week. As another example, the denervation treatment prediction system 100 may determine the first derivative (i.e., slope) and second derivative (i.e., rate of change of the slope) of continuously monitored physiological data, and additionally calculate the mean, maximum, minimum, and / or standard deviation (i.e., variability) of the first and second derivatives over the continuous period of time. The physiological monitoring system 116 may continuously monitor and collect patient specific data 102 that includes physiological data and may transmit that data to the denervation treatment prediction system 100 in any suitable manner. As one example, the physiological monitoring system may transmit data to the denervation treatment prediction system 100 continuously. As another example, the physiological monitoring system 116 may store the continuously monitored patient specific data 102 in memory in one or more locations and transmit that patient specific data 102 to the denervation treatment prediction system 100 in one or more batch transmissions. A batch transmission may be performed at scheduled times, when data storage thresholds are met, when parameters of patient specific data 102 are met, when requested by a clinician 108, and / or at any other suitable time.

[0082] The denervation treatment prediction system 100 may use features of the physiological data to identify specific times within the continuous period of time related to activity of the patient 104. For example, the denervation treatment prediction system 100 may identify patient specific wake-up and / or sleep times based on the slopes of the systolic blood pressure and heart rate continuously monitored over a continuous period of time, such as a twenty-four hour period. In particular, the denervation treatment prediction system 100 may identify a patient specific wake-up time as the time at which the continuously monitored systolic blood pressure has the steepest positive slope during the continuous period of time. The denervation treatment prediction system 100 may identify a patient specific sleep time as the time at which the continuously monitored heart rate has the steepest negative slope during the period of time.

[0083] The denervation treatment prediction system 100 may further identify smaller periods of time within the continuous period of time that are designated by the specific times related to the activity of the patient 104. For example, the denervation treatment prediction system 100 may identify a “morning” as the period of time starting at the identified patient specific wakeup time and ending two hours after the identified patient specific wake-up time. The denervation treatment prediction system 100 may identify a “daytime” as the period of time starting two hours after the identified patient specific wake-up time and ending at the patient specific sleep time. The denervation treatment prediction system 100 may identify a“nighttime” as the period of time starting at the patient specific sleep time and ending at the patient specific wake-up time.

[0084] The denervation treatment prediction system 100 may calculate predictive features based on patient physiological data monitored continuously over any of the identified smaller periods of time. For example, the denervation treatment prediction system 100 may calculate the mean, maximum, minimum, and / or standard deviation (i.e., variability) of one or more of the ambulatory blood pressure, heart rate, or pulse pressure over any of the identified morning, daytime, or nighttime periods. As another example, the denervation treatment prediction system 100 may additionally or alternatively calculate the mean, maximum, minimum, and / or standard deviation (i.e., variability) of the first derivative (i.e., slope) and second derivative (i.e. , rate of change of slope) of the continuously monitored patient physiological data over any of the morning, daytime, or nighttime periods.

[0085] As another example, the patient specific data 102 and, accordingly, the predictive features, may include data indicative of the hormone levels of the patient 104. For example, this data may include one or more of: measurements of the estrogen, progesterone or testosterone levels of the patient 104, a feature indicating whether the patient 104 is perimenopausal or post-menopausal, and / or a number of years after the beginning of menopause. Including this information, either alone or in combination with the other features described above, may increase the effectiveness of the predictions generated by the denervation treatment prediction system 100. For example, it has been found that estrogen levels are highly correlated to cardiovascular risk over the female life span. As a result, conditions such as hypertension and cardiovascular disease in women may be in part driven by changes in hormone levels with time and that hormones, in particular estrogen, are closely regulating the autonomic nervous system. Based on this, autonomic modulation, including renal denervation, may be particularly effective in the treatment of both the underlying disease ensuring state and also symptoms and side effects associated with changes in hormones, particularly in perimenopause and menopause. Thus, including information characterizing hormone levels within the patient 104 may inform the predictions generated by the denervation treatment prediction system 100 regarding appropriate treatment for a given patient 104.

[0086] In some implementations, at block 306, the denervation prediction system 100 performs one or more simulations based on the patient specific data 102 to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient 104. The denervation treatment prediction system 100 may determine parameters for each of the one or more simulations based on the patient specific data 102. Forexample, when the denervation treatment prediction system 100 receives anatomical data for the patient 104, the system may perform simulations of processes within one or more simulated blood vessels, simulated body lumens, and / or simulated organ systems, with the structures of the simulated one or more organ systems determined based on the anatomical data for the patient 104. As another example, when the denervation treatment prediction system 100 receives physiological data for the patient 104, the system may perform simulations of processes within a simulated body, where parameters for the simulated processes are determined based on the physiological data for the patient 104. As another example, the denervation treatment prediction system 100 may perform simulations of one or more denervation procedures. In some implementations, the one or more simulation results may characterize simulated physiological changes in the body of the patient 104, such as but not limited to simulated changes in blood pressure and / or neurological activity, resulting from one or more simulated denervation procedures. The process of performing simulations based on the patient specific data 102 is explained in more detail with reference to FIG. 4 below. In some implementations, the process 300 may omit block 306.

[0087] At block 308, the denervation treatment prediction system 100 processes the predictive features and, optionally, the simulation results to predict the effects of one or more denervation procedures. The denervation treatment prediction system 100 may predict physiological changes in the body of the patient 104 resulting from the one or more denervation procedures. In particular, the denervation treatment prediction system 100 may predict a change in blood pressure in the body of the patient 104 resulting from the one or more denervation procedures.

[0088] In some implementations, the denervation treatment prediction system 100 may predict an effectiveness for the one or more denervation procedures. The predicted effectiveness may be based on predicted physiological changes in the body of the patient 104 (e.g., changes in blood pressure, nerve loss, tissue damage, changes in blood flow, and / or neurological activity) resulting from the one or more denervation procedures performed on the patient 104.

[0089] At block 310, the denervation treatment prediction system 100 provides treatment predictions for review by the clinician 108. In particular, the denervation treatment prediction system 100 provides data characterizing the predicted effects of the one or more denervation procedures for display on a user device, such as but not limited to a monitor, display, computer, tablet, and / or smartphone accessible by the clinician 108.

[0090] In some implementations, the denervation treatment prediction system 100 may determine and provide treatment recommendations for review by the clinician 108. For example, the denervation treatment prediction system 100 may provide data characterizing oneor more suggested denervation procedures. As a further example, the denervation treatment prediction system 100 may determine and provide one or more suggested target organs, suggested locations, and suggested intensities for the suggested denervation procedures. As a further example, the denervation treatment prediction system 100 may determine one or more optimum characteristics of a suggested denervation procedure, such as dose strength and / or dose duration, and provide recommendations in the form of instructions to control a tissue treatment denervation treatment prediction system 100 that are transmitted to that tissue treatment denervation treatment prediction system 100. Those instructions may be implemented as defaults or presets for that particular procedure on that particular patient 104 and may be overridden by a clinician 108 based on their professional judgment. The denervation treatment prediction system 100 may provide data characterizing a predicted effectiveness for the suggested denervation procedures.

[0091] When the denervation prediction system 100 receives medical images as part of the patient specific data 102, the denervation treatment prediction system 100 may produce overlays for the medical images for the patient 104 based on the data characterizing the predicted effects of the one or more denervation procedures. For example, when the denervation treatment prediction system 100 produces suggested denervation procedures, the system may produce overlays that illustrate locations of the suggested denervation procedure(s). As another example, the denervation treatment prediction system 100 may produce overlays that illustrate predicted effects or the predicted effectiveness of suggested denervation procedures. The denervation treatment prediction system 100 may provide data characterizing the overlays for display on the user device.

[0092] In some implementations, the denervation treatment prediction system 100 may generate an indication of whether treatment is recommended and provide the indication whether treatment is recommended for clinician review. In some implementations, the denervation treatment prediction system 100 may determine various recommendations for the patient 104 and provide the recommendations for display on the user device. For example, the denervation treatment prediction system 100 may determine recommended lifestyle changes, medications, and medication dosages for the patient 104 to improve the effectiveness of the denervation treatment.

[0093] With regard to the suggestions for denervation treatment locations, the denervation treatment prediction system 100 may utilize one or more different factors taken from patient specific data 102 and / or from the corpus of patient population data 111 relating to an overall patient population. Relevant factors to consider in determining one or more denervationlocations include the anatomical identity of at least one blood vessel, the diameter of at least one blood vessel at different locations along its length, the locations where at least one blood vessel branches, the presence of calcification at one or more locations on an inner surface of at least one blood vessel and / or within the wall of the at least one blood vessel, the presence of stenosis at one or more locations within at least one blood vessel, the presence of a stent at one or more locations within at least one blood vessel, the presence of an aneurysm in at least one blood vessel, the presence of tumors, fibromuscular disease, and / or other disease conditions in or in proximity to at least one blood vessel, the proximity of nerves or other targets to a lumen of one or more blood vessel, and / or the proximity of at least one blood vessel to other blood vessels and / or organs. The presence of calcification at one or more locations in at least one blood vessel to be treated may be determined by any suitable method, including angiography, intravascular ultrasound, and / or blood flow measurement. Because calcification decreases the diameter of a blood vessel, that decrease in diameter reduces the blood flow in that blood vessel compared to a blood vessel without calcification, and an amount of calcification can be inferred from that blood flow rate. The denervation treatment prediction system 100 may compare the patient specific data 102 to the corpus of patient population data 111 to determine whether the blood flow rate through a particular blood vessel is consistent with a blood vessel that is relatively free of calcification at a location, or a blood vessel that is calcified at that location.

[0094] The denervation treatment locations may include two or more denervation treatment locations within a single main renal artery. Where ultrasound is delivered intravascularly for denervation treatment, a larger number of ablations may increase the probability of better outcomes for the patient 104. Conventionally, two to three ablations may be performed in a single main renal artery. However, by utilizing the corpus of patient population data 111 in addition to the patient specific data 102, the denervation treatment prediction system 100 considers whether a number of ablations greater than two or three is contraindicated for a specific patient 104. If the patient 104 is not contraindicated for more than two to three ablations, the denervation treatment prediction system 100 may suggest four, five, six, or more denervation treatment locations inside a single main renal artery and / or branch and / or accessory renal artery.

[0095] In other embodiments, the denervation treatment prediction system 100 utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to consider whether a number of ablations less than two is indicated for a specific patient 104.

[0096] In one example, an ultrasound transducer 628, which will be described in more detail below, may be substantially 6 mm in length and is used to deliver unfocused ultrasound energyoutwardly from an intravascular location, allowing it to generate lesions of substantially 5 mm in length, as measured along the longitudinal centerline of the renal artery in which the ultrasound transducer 628 is placed, during each actuation thereof. In other embodiments, the transducer can be shorter or longer than 6 mm in length. The ultrasound transducer 628 may be moved a short distance longitudinally along the renal artery after each actuation, such that two to three ablations performed by the ultrasound transducer 628 may result in lesions that extend substantially 1 to 1.5 cm along the main renal artery. Each additional ablation thus adds substantially 5 mm to the total length of the lesions along the main renal artery. More generally, for an ultrasound transducer 628 of length Lt, where the ultrasound transducer 628 is moved adjacent to its previous position after each ablation, the total length of the lesion is approximately Lt* N, where N is the number of individual ablations. The denervation treatment prediction system 100 may suggest that number N be increased or decreased in order to increase the probability of better outcomes (i.e., effective and / or safe) for the patient 104. Even more generally, for an energy emitter having length Lt, where the energy emitter is moved adjacent to its previous position after each ablation, the total length of the lesion is approximately Lt* N, where N is the number of individual ablations. Such an energy emitter may additionally or alternatively include one or more RF electrodes placed in contact with an interior of a vessel wall, a microwave emitter, a laser or other optical emission source, and / or an outlet or outlets for cryoablation fluid or chemical agents.

[0097] The denervation treatment locations may include locations in different arteries, veins, and / or other locations utilized to treat nerves associated with different organs and / or arteries. Such locations may vary based on the type of therapy that is needed by the patient 104. As one example, for a patient 104 with hypertension, the denervation treatment prediction system 100 may suggest one or more denervation treatment locations for denervation of one or more renal nerves (each associated with a renal artery), as well as one or more denervation treatment locations for denervation of at least a portion of the celiac plexus and / or at least a portion of the aorticorenal ganglia and / or ganglionated plexuses and / or other nerves in communication with the central sympathetic nervous system, for example one or more hepatic and / or splanchnic nerves. By utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to make suggestions, the denervation treatment prediction system 100 is capable of making suggestions of denervation treatment locations, along with making predictions of success for the procedure, that the clinician 108 might not consider on their own.

[0098] With regard to the suggestions for denervation treatment locations, the denervation treatment prediction system 100 may utilize one or more different factors taken from patientspecific data 102 and / or from the corpus of patient population data 111 to suggest an optimized intensity, power, and duration (referred to above as the dose) of energy to be applied at each suggested denervation treatment location. The optimal dose may be different at one or more of the suggested denervation treatment locations. By utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to make suggestions, the denervation treatment prediction system 100 is capable of making suggestions of a denervation treatment dose at each location, along with making predictions of success for the procedure, than the clinician 108 may be able to determine on their own.

[0099] Without wishing to be bound to a particular theory, in some patients 104, the renal nerve(s) may be easier to treat within one or more distal branches at the distal end of the renal artery, particularly where RF electrodes are used for ablation that have an average ablation depth of 2 to 4 mm. In some patients 104, the renal nerve(s) may be easier to treat immediately proximal to the distal branch(es) at the distal end of the renal artery, particularly where an ultrasound transducer, which can have longer ablation depths, is used for ablation. The denervation treatment prediction system 100 may utilize patient specific data 102 that includes imaging data of the renal artery, nerves, and / or nearby organs and structures, such as the kidneys and large intestine, in combination with the corpus of patient population data 111, to determine whether at least one denervation treatment location should be within a distal branch of the renal artery, and / or within the renal artery immediately proximal to the distal branch(es). Further, the denervation treatment prediction system 100 may utilize patient specific data 102 that includes imaging data of the renal artery, nerves and nearby organs and structures, such as the kidneys and large intestine, in combination with the corpus of patient population data 111, to determine the outward extent of the treatment volume from the energy emitter utilized. For example, the denervation treatment prediction system 100 may suggest a dose that results in a treatment area extending outward from the lumen of a vessel a distance from 1 mm to 10 mm.

[0100] The renal artery, as well as other blood vessels, may change its tissue composition along at least a portion of its length. The patient specific data 102 may include properties of that tissue measured along at least a portion of that renal artery or other blood vessel, such as but not limited to the presence and / or amount of fatty tissue and its location(s), the presence and / or amount of collagen and its location(s), the presence and / or amount of muscular fiber and its location(s), and / or the thickness of the intima, media, and / or adventitia along the length of the renal artery and / or other blood vessel and / or the stiffness along the length of the blood vessel.

[0101] The denervation treatment prediction system 100 may suggest one or more denervation locations away from locations in the renal artery or other blood vessel where the media is thin in order to reduce the risk of stenosis at the denervation location(s).

[0102] The denervation treatment prediction system 100 may suggest one or more denervation locations away from locations in the renal artery or other blood vessel where the media is thick, when the energy emitter is an ultrasound transducer, in order to reduce the risk of stenosis at the denervation location(s). For example, in applications using ultrasound, the denervation treatment prediction system 100 may suggest increasing the flow rate in order to protect a greater depth in the near field where the media is thicker.

[0103] In general, the denervation treatment prediction system 100 may suggest one or more denervation locations that are not at locations where treatment may increase the risk of stenosis above a particular level.

[0104] In making suggestions for one or more denervation treatment locations, the denervation treatment prediction system 100 may take into account the presence or absence of one or more heat sinks in proximity to a treatment site. Heat sinks include veins and lymph nodes, which include liquid that can absorb energy from an energy emitter more quickly than, or in addition to, tissue in the treatment volume, which could result in less effective treatment. If a heat sink is present in proximity to a treatment site, and ablation is to be performed using an ultrasound transducer 628 to deliver ultrasound energy to the treatment site, the denervation treatment prediction system 100 may increase the power delivered to the ultrasound transducer 628 in order to increase the power emitted from the ultrasound transducer 628 during treatment, as compared to the power emitted from the ultrasound transducer 628 during treatment where a heat sink is not present in proximity to the treatment site. If a heat sink is present in proximity to a treatment site, and ablation is to be performed using RF electrodes, the denervation treatment prediction system 100 may utilize that information to select a different location for the application of denervation treatment, because increasing power to RF electrodes does not overcome the loss of energy to a heat sink, and the increase in power to RF electrodes may cause damage to the inner wall of the blood vessel in contact with the RF electrodes.

[0105] In some cases, the denervation treatment prediction system 100 may recommend against denervation altogether based on the condition of the blood vessel(s) that would be treated with denervation, because proceeding with denervation would not provide a benefit to the patient 104 and / or the risk to the patient would be too great.

[0106] The various types of predictions, recommendations, and outputs from the denervation prediction system are explained in more detail with reference to FIG. 5 below.

[0107] FIG. 4 is a flow diagram of an example process for simulating a denervation procedure. For convenience, the process 400 will be described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1, appropriately programmed in accordance with this specification, may perform the process 400.

[0108] At block 402, the denervation treatment prediction system 100 receives patient specific data 102, the contents of which are described above with regard to FIG. 1. The patient specific data 102 may include physiological data that characterizes processes within the body of the patient 104, such as but not limited to blood pressures, blood pressure variability, heart rates, metabolic activity, neurological activity, and / or other physiological data associated with the patient 104. The patient specific data 102 may include anatomical data that characterizes an internal structure of the body of the patient 104, e.g., the structure and locations of one or more various tissues, organs, bones, nerves, blood vessels, whether the patient 104 has solitary renal arteries, branching pattern of renal arteries, length / diameter of renal arteries, length / diameter of accessory renal arteries, length / diameter of distal and / or proximal branches and / or other structures within the body of the patient 104. In some embodiments, the patient specific data 102 may include medical images that illustrate the internal structure of the body of the patient 104, such as obtained from a CT scan, data from an MRI scan, data from an X-Ray, data from an angiogram, data from a PET scan, data from an ultrasound, and so on.

[0109] At block 404, the denervation treatment prediction system 100 performs one or more simulations of a body of the patient 104 based on the patient specific data 102. The denervation treatment prediction system 100 may determine parameters for each of the simulations based on the patient specific data 102. As an example, the simulations may include simulations of processes within at least one simulated organ system, and the denervation treatment prediction system 100 may determine the structure of the at least one simulated organ system based on the anatomical data received for the patient 104. As a particular example, the denervation treatment prediction system 100 may determine the structure of the at least one simulated organ system based on medical images of the body of the patient 104. As another example, the simulations may include simulations processes within a simulated body, and the system may determine parameters for the simulated process based on the received physiological data for the patient 104. As a particular example, the denervation treatment prediction system 100 may use data related to blood circulation within the body of the patient 104 e.g., blood pressures, heart rates, and / or blood flow rate to one or more specific organs such as the kidneys) to perform simulations of blood flow within a simulated body of the patient 104. As anotherparticular example, the denervation treatment prediction system 100 may use data specifying neurological activities measured for the body of the patient 104 to perform simulations of neurological responses within a simulated body of the patient 104.

[0110] The simulations may include simulated denervation procedures. The denervation treatment prediction system 100 may perform simulations for multiple denervation procedures in different locations within the body of the patient 104. For example, the simulations may include multiple simulated denervation procedures at different locations within a particular organ of the patient. As another example, the simulations may include simulations of denervation procedures to multiple different organs of the patient. The simulations may include simulated denervation procedures to any one or more of a variety of organs within the body of the patient 104, such as but not limited to a kidney, heart, a liver, a duodenum, a stomach, a spleen, and / or a pancreas. In particular, the simulations may be physical simulations (e.g., of blood flow, ablation processes, nervous system responses) of the denervation procedures using various computational methods (e.g., computational fluid dynamics and / or finite element method simulations) based on patient specific anatomical and physiological data. Blood flow characteristics may be determined by using computational fluid dynamics, and the velocity field of blood flow in an artery may be determined by solving the Navier-Stokes equation numerically.

[0111] At block 406, the denervation treatment prediction system 100 may perform simulations of blood flow through arteries of the patient 104. In particular, the denervation treatment prediction system 100 may simulate blood flow through arteries of the patient 104 before, during and after simulated denervation procedures. As an example, the denervation treatment prediction system 100 may simulate and determine changes in blood pressure resulting from a simulated denervation procedure and / or a post-denervation treatment plan. The denervation treatment prediction system 100 may simulate blood flow through arteries of the patient 104 using computational fluid dynamics (CFD) simulations. In particular, the denervation treatment prediction system 100 may perform CFD simulations of blood flow through simulated arteries based on medical images of the patient’s arteries.

[0112] At block 408, in some implementations, the denervation treatment prediction system 100 may simulate arterial wall responses to the simulated blood flows. The denervation treatment prediction system 100 may determine the simulated arterial wall responses based on material properties of simulated arteries, such as but not limited to their elasticity and / or rigidity. The denervation treatment prediction system 100 may determine simulated arterial wall responses as part of performing CFD simulation of blood flow through simulated arteries.

[0113] In some implementations, at block 410, the denervation treatment prediction system 100 may simulate tissue ablation for simulated denervation procedures. In particular, the denervation treatment prediction system 100 may simulate transfers of energy from an energy source (such as but not limited to an ultrasonic transducer that emits ultrasonic energy, one or more electrodes that emit radio frequency (RF) or electrical energy, a microwave antenna that emits microwave energy, and / or a laser that emits laser energy) to tissues of the patient 104. The denervation treatment prediction system 100 may simulate the transfers of energy from the energy source into tissue using, e.g., a finite element method that may directly simulate energy transfer to the tissue based on one or more material properties of simulated tissue, such as but not limited to thermal conductivity, elasticity, and / or rigidity.

[0114] At block 412, the denervation treatment prediction system 100 may simulate nervous system responses, e.g., sympathetic, parasympathetic, and autonomic nervous system responses, of the patient 104. In particular, the denervation treatment prediction system 100 may simulate nervous system responses before, during, and after simulated denervation procedures.

[0115] At block 414, the denervation treatment prediction system 100 returns simulation results that are predictive of effects of the denervation procedures being performed to the body of the patient 104. In particular, the simulation results may be predictive of physiological changes e.g., changes in blood pressure, heart rate, and / or neurological activity) in the body of the patient 104 resulting from the denervation procedures and / or post-denervation treatment plan. When the denervation treatment prediction system 100 simulates denervation procedures, the simulation results may characterize predicted physiological changes in the body of the patient 104 as a result of the simulated denervation procedures and / or post-denervation treatment plan. For example, the simulation results may characterize predicted changes in blood pressure, nerve loss, tissue damage, changes in blood flow, neurological activity, and / or other changes in the body of the patient 104 as a result of the simulated denervation procedure(s) and / or post-denervation treatment plan.

[0116] FIG. 5 is a flow diagram of an example process for producing denervation treatment recommendations and / or post-denervation treatment plan. For convenience, the process 500 will be described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1, appropriately programmed in accordance with this specification, may perform the process 500.

[0117] At block 502, the denervation treatment prediction system 100 determines predicted effects of one or more denervation procedures to a body of a patient 104. As described above,the denervation treatment prediction system 100 determines predicted effects of the denervation procedures using the denervation procedure prediction model 204. The denervation treatment prediction system 100 processes patient specific data 102 for the patient to produce predictive features based on the patient specific data 102. The denervation treatment prediction system 100 may also perform simulations to produce simulation results that are predictive of the effects of the denervation procedures to the body of the patient 104. The denervation procedure prediction model 204 processes the predictive features and, optionally, the simulation results to predict the effects of the denervation procedures and / or postdenervation treatment plan. The denervation procedure prediction model 204 may output various data characterizing the predicted effects of the denervation procedures and / or postdenervation treatment plan.

[0118] The predicted effects of the denervation procedures may include predicted physiological changes in the body of the patient 104 (e.g., changes in blood pressure, nerve loss, tissue damage, changes in blood flow, and / or neurological activity) resulting from the denervation procedures being performed on the body of the patient 104 and / or post-denervation treatment plan.

[0119] The data characterizing the predicted effects of the denervation procedures may include data that characterizes distributions of the predicted effects. For example, the data characterizing the predicted effects may include probabilities or likelihoods for the predicted effects. As another example, the data characterizing the predicted effects may include various statistical properties, such as but not limited to means, medians, deviations, and / or variances) for the predicted effects. As another example, the data characterizing the predicted effects may include parameters specifying distributions of the predicted effects.

[0120] The data characterizing the predicted effects of the denervation procedures may include a predicted effectiveness of the denervation procedures, as determined by the denervation procedure prediction model 204. The predicted effectiveness may be determined based on the predicted physiological changes in the body of the patient 104 resulting from the denervation procedures and / or post-denervation treatment plan.

[0121] The predicted effectiveness for a denervation procedure may be a predicted effectiveness score. For example, the predicted effectiveness score may characterize a probability of success for the denervation procedure. As another example, the predicted effectiveness score may characterize a probability that particular physiological changes occur within the body of the patient 104 as a result of the denervation procedure, such as but not limited to a probability that the patient 104 experiences a blood pressure change past aparticular threshold value. As another example, the predicted effectiveness score may characterize predicted magnitudes of physiological changes in the body of the patient 104 as a result of the denervation procedure.

[0122] The predicted effectiveness for a denervation procedure may be a categorical effectiveness classification of the denervation procedure. For example, the categorical effectiveness classification may specify whether the denervation procedure is predicted to be successful. As another example, the categorical effectiveness classification may specify a predicted category of physiological changes in the body of the patient 104 resulting from the denervation procedure. For example, such categories may include categories of blood pressure change of “low” (e.g., less than 5 mmHg), “moderate” (e.g., between 5 mmHg and 10 mmHg), and “high” (e.g., greater than 10 mmHg) resulting blood pressure drops). As another example, the categorical effectiveness classification may specify a predicted category of physiological changes in the body of the patient 104 resulting from the denervation procedure, e.g., whether the patient 104 is likely to be a super responder, an average responder, or a nonresponder. As another example, the categorical effectiveness classification may specify a predicted category of physiological changes in the body of the patient 104 resulting from the denervation procedure, e.g., whether the patient 104 is likely to have a blood pressure decrease in stages of hypertension (e.g., from stage 2 hypertension (a blood pressure greater or equal to 140 / 90 mmHg) to stage 1 hypertension (a blood pressure less than 140 / 90 mmHg but greater or equal to 130 / 80 mmHg), or to elevated (systolic less than 130 mmHg but systolic / diastolic greater or equal to 120 / 80), or to normal (less than 120 / 80). As another example, the categorical effectiveness classification may specify a predicted risk category of the denervation procedure for the patient 104 e.g., high, moderate, and low risk).

[0123] In some implementations, the denervation treatment prediction system 100 simulates denervation procedures and may predict the effects of the simulated denervation procedures on the body of the patient 104. The data characterizing the effects of the denervation procedures may include data characterizing the predicted effects of the simulated denervation procedures.

[0124] At block 504, the denervation treatment prediction system 100 may determine, using the denervation procedure prediction model 204, one or more suggested denervation procedures. In particular, the denervation treatment prediction system 100 may determine suggested denervation procedures based on the predicted effects of the denervation procedures. As an example, the denervation treatment prediction system 100 may determine one or more target organs for the suggested denervation procedures. As another example, the denervation treatment prediction system 100 may determine suggested locations for the suggesteddenervation procedures. In some implementations, the denervation treatment prediction system 100 may determine the suggested denervation procedures by optimizing a predicted effectiveness for the denervation procedures. For example, the denervation treatment prediction system 100 may determine the suggested procedures by determining suggested locations for the denervation procedures that optimizes the predicted effectiveness. As a particular example, the denervation treatment prediction system 100 may simulate multiple denervation procedures and may select the simulated procedures maximizing predicted effectiveness as the suggested denervation procedures. As another particular example, the denervation treatment prediction system 100 may, over a sequence of iterations, optimize locations for the suggested denervation procedures to maximize predicted effectiveness. The denervation treatment prediction system 100 may produce data characterizing the suggested denervation procedures for display on a user device, such as but not limited to a monitor, a display, a computer, and / or a smartphone accessible by the clinician 108 and / or user.

[0125] At block 506, the denervation treatment prediction system 100 may determine, using the denervation procedure prediction model 204, denervation treatment recommendations. For example, the denervation treatment prediction system 100 may determine, based on the predicted effects of the denervation procedures, whether denervation treatment is recommended for the patient 104. The denervation treatment prediction system 100 may provide the denervation treatment recommendations for clinician and / or user review.

[0126] At block 508, the denervation treatment prediction system 100 may produce, using the denervation procedure prediction model 204, various medical image overlays. When the denervation treatment prediction system 100 receives medical images, the medical image overlays may be overlays for the received medical images. The medical image overlays may illustrate predicted effects of the denervation procedures. For example, the medical image overlays may illustrate predicted physiological changes resulting from the denervation procedures, e.g., changes in blood pressure, changes in heart rate, blood and / or hepatic glucose level, glucagon level, insulin level, leptin levels, blood and / or tissue norepinephrine level, markers of inflammation, estrogen, progesterone, testosterone, cytokines, vascular muscle tone, triglyceride level, blood C peptide level, ejection fraction, ventricular wall dimensions, cholesterol level, including high-density lipoprotein (HDL) levels, low-density lipoprotein (LDL) levels, very-low-density lipoprotein (VLDL) levels, changes in sympathetic drive), nerve loss, tissue damage, changes in blood flow, and / or neurological activity. As another example, the medical image overlays may illustrate a predicted effectiveness e.g., an effectiveness score or a categorical effectiveness classification) for each of the denervationprocedures. When the denervation treatment prediction system 100 simulates denervation procedures, the medical image overlays may include overlays that illustrate the simulated denervation procedures.

[0127] When the denervation treatment prediction system 100 produces suggested denervation procedures, the medical image overlays may include overlays that illustrate the suggested denervation procedures. For example, the medical image overlays may illustrate predicted effects of the suggested denervation procedures. As another example, the medical image overlays may illustrate a predicted effectiveness (e.g., an effectiveness score or a categorical effectiveness classification) for each of the suggested denervation procedures. As another example, the medical image overlays may illustrate suggested locations for the suggested denervation procedures, as described in greater detail below.

[0128] The denervation treatment prediction system 100 may provide medical image overlays for display on a user device, such as but not limited to a monitor, a display, a computer, and / or a smartphone accessible by the clinician 108.

[0129] At block 510, the denervation treatment prediction system 100 may determine, using the denervation procedure prediction model 204, various recommendations for the patient 104 based on the patient specific data 102 and the predicted effects of the denervation procedures, and output those recommendations. The denervation treatment prediction system 100 may, for example, determine recommendations for the patient 104 based on optimizing the predicted effectiveness for the denervation procedures. As an example, the recommendations for the patient 104 may include recommended lifestyle changes for the patient 104. For example, the recommendations may tailor an exercise program (such as but not limited to an aerobic, resistance training, intensity, and / or duration program), smoking reduction or cessation, alcohol reduction or cessation, a sleep schedule (such as but not limited to a change in the number of hours of sleep per 24-hour period, the times of such sleep, and / or the intervals between such sleep), suggestions for sleep quality improvement (such as treatment for sleep apnea) or other recommendations for the patient 104 based on the patient specific data 102. Such an exercise program may include a recommendation for aerobic exercise to decrease diastolic blood pressure, and / or isometric exercise to decrease systolic blood pressure. As another example, the recommendations for the patient 104 may include medication recommendations for the patient 104. As a further example, the recommendations for the patient 104 may include recommended medication dosages for the patient 104. The denervation treatment prediction system 100 may provide data characterizing therecommendations for the patient 104 for display on a user device, such as but not limited to a monitor, a display, a computer, and / or a smartphone accessible by the clinician 108.

[0130] Also at block 510, the recommendations may be output to the clinician 108. Recommendations may be output to the clinician 108 on any suitable device. As one example, recommendations may be output to the clinician 108 and / or user on a tablet, smartphone, computer, or display in an examination room or in the office of the clinician 108. As another example, recommendations may be output to the clinician 108 and / or user on a display of the controller 604, which is described below.

[0131] To produce the data characterizing the predicted effects of the denervation procedures, the denervation treatment prediction system 100 may perform any combination of blocks 504 through 510. For example, the denervation treatment prediction system 100 may perform none, a subset, or all of blocks 504 through 510.

[0132] At block 512, the denervation treatment prediction system 100 may return the data characterizing the predicted effects of the denervation procedures for clinician review. In particular, the denervation treatment prediction system 100 provides the data characterizing the predicted effects of the denervation procedures for display on a user device, such as but not limited to a monitor, a display, a computer, and / or a smartphone accessible by the clinician 108.

[0133] FIG. 6A is an example tissue treatment system 600. The tissue treatment system 600 may include a catheter 602 that may be delivered intraluminally, e.g. , intravascularly, to a target anatomical region of a subject. When so placed, an energy emitter of the denervation treatment prediction system 100 may be positioned within a target anatomy, such as but not limited to within a body lumen such as a blood vessel. The tissue treatment system 600 also may include a controller 604, and a connection cable 606. The tissue treatment system 600 may also include a balloon 608, a reservoir 610, a cartridge 612, and a control mechanism, such as a handheld remote control. In certain embodiments, the controller 604 may be connected to the catheter 602 through the cartridge 612 and the connection cable 606. In certain embodiments, the controller 604 may interface with the cartridge 612 to provide cooling fluid to the catheter 602 for inflating and deflating the balloon 608. The controller 604 may include a display. The controller 604 may also be referred to as the control unit.

[0134] FIG. 6B is an example of a particular embodiment of the catheter 602 of the tissue treatment system 600. In this embodiment of the catheter 602, the energy emitter may be an ultrasound transducer 628 located distal 614 to a catheter shaft 624, within an interior of the balloon 608. The ultrasound transducer 628 may be an ultrasound transducer used to emitenergy toward the vessel wall. For example, the ultrasound transducer 628 may be activated to deliver unfocused ultrasonic energy radially outwardly so as to suitably heat, and thus treat, tissue within the target anatomical region. The ultrasound transducer 628 may be activated at a frequency, time, and energy level suitable for treating the targeted tissue. One or more electrical couplings 618 may be provided on the catheter 602, for connection to the controller 604 or other power source, and such electrical coupling(s) 618 are electrically connected to the energy emitter, such as the ultrasound transducer 628, to provide energy to the energy emitter. One or more fluidic input ports 620 may be provided on the catheter 602, to accept input fluid, such as cooling fluid, inflation fluid, or other fluid. The fluidic input port(s) 620 may be connected to a fluid lumen or lumens that conduct that input fluid to an interior of the balloon 608. One or more fluidic outlet ports 622 may be provided on the catheter 602, to accept outlet fluid, such as spent cooling fluid, spent inflation fluid, or other fluid. The fluidic outlet port(s) 622 may be connected to a fluid lumen or lumens that conduct that outlet fluid from an interior of the balloon 608.

[0135] FIG. 7 is a block diagram of the controller 604 that may be included in the tissue treatment system 600. The controller 604 may include one or more processors 712, a memory 714, a display 716, and an ultrasound excitation source 718, but may include additional and / or alternative components. Each processor 712 may communicate with the memory 714, which may be a non-transitory computer-readable medium storing instructions. A user interface displayed on the display 716 interacts with the processor 712 to cause transmission of electrical signals at selected actuation frequencies to the ultrasound transducer 628 via wires of the connection cable 606 and the cabling 282 that extends through the catheter shaft 624. These wires electrically couple the controller 604 to the ultrasound transducer 628 so that the controller 604 may control the excitation source 718 to control one or more ultrasound treatment parameters that are used to perform sonication. That is, the controller 604 controls the excitation source 718, which transmits energy to the ultrasound transducer 628, such as through the connection cable 606, which is then converted to ultrasound energy by the ultrasound transducer 628. While the ultrasound excitation source 718 in FIG. 7 is shown as being part of the controller 604, it is also possible that the ultrasound excitation source 718 is external to the controller 604 while still being controlled by the controller 604, and more specifically, by the processor 712 of the controller 604.

[0136] The display 716 may include a touch screen, buttons, and / or switches, to display the user interface to and receive input from an operator (user) to enter patient data, select treatment parameters, view records stored on a storage / retrieval unit (not shown), and / or otherwisecommunicate with the processor 712. According to some embodiments, the display 716 may be part of a tablet, smartphone, computer, or similar device.

[0137] In some embodiments, at least one recommendation determined at a block of determining recommendations for a patient 104 based on patient specific data 102 and predicted effects of denervation procedures in a process for producing denervation treatment recommendations, such as block 510 in the process 500 of FIG. 5, includes transmitting instructions to a tissue treatment system to control at least one characteristic of a renal denervation procedure to be performed on that specific patient 104. Those instructions may include, and are not limited to, dose strength, dose duration, balloon pressure, and / or flow rate. Those instructions to control at least one characteristic of a renal denervation procedure may be implemented as defaults for that particular procedure and may be overridden by a physician based on their professional judgment. In some embodiments, at least one recommendation determined at block 510 may include preparing an electronic order for medication to be transmitted to a pharmacy, such as one associated with the patient 104. The clinician 108 may review that electronic order before placement and may change that order based on their professional judgment.

[0138] FIG. 8 is a flow diagram of an example process 800 for producing treatment recommendations before, during, and after a denervation treatment. For convenience, the process 800 will be described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1 , appropriately programmed in accordance with this specification, may perform the process 800.

[0139] At block 802, the denervation treatment prediction system 100 may receive patient specific data 102, as described above with regard to FIG. 1, obtained before a planned denervation treatment for a patient 104. The patient specific data 102 may include physiological data that characterizes processes within the body of the patient 104, such as but not limited to blood pressures, heart rates, metabolic activity, and / or neurological activity. The patient specific data 102 may include anatomical data that characterizes an internal structure of the body of the patient 104, such as but not limited to the structure and location of one or more tissues, organs, bones, nerves, blood vessels, and / or other structures within the body of the patient 104. In particular, the patient specific data 102 may include medical images that illustrate the internal structure of the body of the patient 104, such as but not limited to images obtained from a CT scan, from an MRI scan, from an X-Ray, from an angiogram, from a PETscan, from an external ultrasound scan, and / or from an intravascular ultrasound scan such as by an IVUS device.

[0140] The patient specific data 102 may include additional information for the patient 104. For example, the patient specific data 102 may specify information such as a height, a weight, an age, a sex, a race, and / or other data associated with the patient 104. As another example, the patient specific data 102 may include listings of medications prescribed to or taken by the patient 104. As another example, the patient specific data 102 may specify information reported by the patient 104, such as but not limited to a pain index, a family history, a description of symptoms, dietary habits, exercise habits, and so on. The patient specific data 102 may specify information reported by a clinician 108 of the patient 104, such as but not limited to an evaluation of a risk for the patient 104, treatment recommendations, the results of an examination of the patient 104, and so on.

[0141] At block 804, the denervation treatment prediction system 100 may provide denervation treatment recommendations for the planned denervation treatment to a clinician 108 of the patient 104 and / or user. As described above, the denervation treatment prediction system 100 may predict the effectiveness of denervation procedures for the denervation treatment. The denervation treatment prediction system 100 may predict physiological changes (e.g., changes in blood pressure and / or neurological activity, etc.) in the body of the patient 104 resulting from the denervation procedures and may determine the predicted effectiveness of each procedure based on the predicted physiological changes. For example, the denervation treatment prediction system 100 may predict effectiveness scores or categorical effectiveness classifications that characterize probabilities of success, physiological outcomes, risk factors, and / or other outcomes.

[0142] The denervation treatment prediction system 100 may utilize factors such as the locations of denervation treatment, patient specific data 102 such as but not limited to BMI, AB PM, and gender of the patient 104, and compare those to the corpus of patient population data 111 as part of predicting a treatment outcome. The denervation treatment prediction system 100 may predict a treatment outcome in terms of a predicted drop in baseline blood pressure. That drop may be in systolic blood pressure, diastolic blood pressure, or both. The denervation treatment prediction system 100 may predict an immediate treatment outcome, and / or a treatment outcome three months, six months, and / or other interval after treatment. As described below, an app may be provided that the patient 104 may download to their smartphone, tablet, computer, or other device. Communication between the denervation treatment prediction system 100 and the app may be useful in tracking the physiologicalcharacteristics of the patient 104 and may be utilized by the denervation treatment prediction system 100 to update its prediction of a treatment outcome for a patient 104 over time. Predictions made further out in time may be less accurate due to the introduction of confounding factors such as a change in medication administered to a patient 104, one or more additional unexpected disease states of the patient 104, and / or other factors. By communicating these facts to the denervation treatment prediction system 100 via the app or otherwise, the denervation treatment prediction system 100 is able to update its predictions of future outcomes accordingly.

[0143] The denervation treatment prediction system 100 may determine suggested denervation procedures for the planned denervation treatment. The denervation treatment prediction system 100 may determine the suggested denervation procedures by simulating the procedures based on the patient specific data 102. In some implementations, the denervation treatment prediction system 100 may determine the suggested denervation procedures by optimizing the predicted effectiveness of the procedures. As examples, the denervation treatment prediction system 100 may determine one or more target organs and one or more suggested locations of the suggested denervation procedure or procedures that optimize the predicted effectiveness.

[0144] The denervation treatment prediction system 100 may produce additional recommendations for the planned denervation treatment. For example, the denervation treatment prediction system 100 may indicate to the clinician whether denervation treatment is recommended. As another example, the denervation treatment prediction system 100 may provide recommendations for actions the patient 104 may perform to improve the effectiveness of the planned denervation treatment. For example, the denervation treatment prediction system 100 may recommend lifestyle changes for the patient 104 to improve treatment outcomes.

[0145] When the denervation treatment prediction system 100 receives medical images as part of the patient specific data 102, the denervation treatment prediction system 100 may provide medical image overlays that illustrate the treatment predictions for the clinician 108. For example, the denervation treatment prediction system 100 may produce overlays that illustrate simulated denervation effects, such as but not limited to predicted effectiveness, predicted changes in blood flow, predicted neurological activity, predicted changes in blood pressure, and / or predicted tissue damage. As another example, the denervation treatment prediction system 100 may produce overlays that illustrate suggested one or more denervation procedures, such as but not limited to target organs for the suggested procedure(s) and / or suggested locations for the suggested procedure(s).

[0146] At block 806, the denervation treatment prediction system 100 may receive patient specific data 102 obtained during an ongoing denervation treatment for the patient 104. For example, the denervation treatment prediction system 100 may receive data from physiological monitoring of the patient 104 during the treatment. As another example, the denervation treatment prediction system 100 may receive updated medical images obtained for the patient 104 during the treatment. In particular, the denervation treatment prediction system 100 may receive patient specific data 102 that characterizes denervation procedures that have been completed by the clinician 108 during the treatment.

[0147] At block 808, the denervation treatment prediction system 100 may provide denervation procedure predictions for the ongoing denervation treatment to the clinician. In particular, the denervation treatment prediction system 100 may provide updated suggestions and recommendations to the clinician 108 during the ongoing treatment. For example, the denervation treatment prediction system 100 may determine suggested denervation procedures based on procedures the clinician 108 has performed during the ongoing treatment. As another example, the denervation treatment prediction system 100 may provide medical image overlays that illustrate predicted effects of the performed treatments and illustrate additional suggested procedures for the clinician 108 during the ongoing treatment.

[0148] At block 810, while the denervation treatment continues, the denervation treatment prediction system 100 may continue obtaining updated patient specific data 102 and providing updated predictions for the clinician 108. After the treatment is complete, as determined by the clinician 108, the denervation treatment prediction system 100 may provide predictions regarding the long-term effects of the completed treatment. For example, the denervation treatment prediction system 100 may provide predictions, based on the denervation procedures completed by the clinician 108, of long-term physiological changes for the patient 104 as a result of the treatment.

[0149] At block 812, the denervation treatment prediction system 100 may receive patient specific data 102 obtained after a completed denervation treatment for the patient 104. For example, the denervation treatment prediction system 100 may receive data from follow-up physiological monitoring and medical imaging of the patient 104. As another example, the denervation treatment prediction system 100 may receive follow-up data reported by the patient 104 during follow-up for the treatment, such as but not limited to blood pressure, body temperature, body mass index, abdominal obesity, abdominal circumference, cardiac output, exercise tolerance, sleep duration, sleep quality, blood glucose level, blood alcohol level, blood hormone level, pulse rate, respiration rate, pulse oxygenation, sleep apnea, skin sodiumcontent, elevated plasma or cerebrospinal fluid NaCl concentration, metabolic activity, heart rate variability, tobacco use data, stress level, diet information, a pain index and / or a description of symptoms.

[0150] The denervation treatment prediction system 100 may be configured to communicate with a device of the patient 104 after the denervation procedure. The device of the patient 104 may be a smartphone, a computer, a tablet, or other suitable device. An app may be provided that the patient 104 may download to their smartphone. Communication between the denervation treatment prediction system 100 and a device of the patient 104 may be useful in obtaining short-term and long-term results of the denervation procedure on the patient 104. These results may benefit the patient 104, and additionally or instead may be used to train the denervation treatment prediction system 100 to better predict the results of a denervation procedure or procedures for other patients 104. In some embodiments, the patient 104 may input follow-up data, as described in the previous paragraph, into the app on their smartphone, into a website, into an email or text, or into other communication means, which is then transmitted to the denervation treatment prediction system 100. The denervation treatment prediction system 100 may receive information reported by the clinician 108 of the patient 104 during follow-up for the treatment, such as an evaluation of a risk for the patient 104, treatment recommendations, the results of an examination of the patient 104, and so on.

[0151] At block 814, the denervation treatment prediction system 100 may provide follow-up recommendations and predictions. For example, the denervation treatment prediction system 100 may determine whether further denervation treatment is recommended for the patient 104. As another example, the denervation treatment prediction system 100 may evaluate long-term physiological effects of the treatment for the patient 104, such as but not limited to changes in blood pressure and / or neurological activity. As another example, the denervation treatment prediction system 100 may recommend medication and / or lifestyle changes for the patient 104 based on the follow-up data. As another example, the denervation treatment prediction system 100 may recommend medications and medication doses for the patient 104 based on the followup data. Additionally or alternatively, the denervation treatment prediction system 100 may recommend an exercise regimen, smoking cessation program, meeting with a dietician, etc.

[0152] By utilizing the corpus of patient population data 111 in addition to the patient specific data 102, the denervation treatment prediction system 100 may recommend medications and medication dosages for the patient 104 to improve treatment outcomes, and / or may predict the results of one or more changes in medications and / or medication doses after treatment. As one example, the denervation treatment prediction system 100 may recommend one or morechanges in dosage of the existing hypertension medication taken by the patient 104 after a denervation procedure. Such changes may be suggested as a single step function, or as suggestions for multiple changes over weeks or months following a denervation procedure. As another example, the denervation treatment prediction system 100 may recommend that the patient 104 cease taking one or more specific medications at certain times after a denervation procedure. As another example, the denervation treatment prediction system 100 may recommend that the patient 104 titrate down the dosage of one or more of the hypertension medications. As another example, the denervation treatment prediction system 100 may recommend that the patient 104 change the class of one or more hypertension medications previously taken by the patient 104 after the denervation treatment. Hypertension medications are typically divided into different classes based on their mechanisms of action; such classes include diuretics, beta-blockers, angiotensin-converting enzyme (ACE) inhibitors, angiotensin II receptor blockers (ARBs), calcium-channel blockers, alpha blockers, alpha-2 receptor agonists, and vasodilators. After denervation for treatment of hypertension, the denervation treatment prediction system 100 may determine that the patient 104 would benefit from changing at least one class of their previous hypertension medication, from eliminating at least one class of their previous hypertension medication, and / or from adding at least one class to their previous hypertension medication.

[0153] As described above, estrogen and other hormone levels are highly correlated to cardiovascular risk over the female life span. As a result, the denervation treatment prediction system 100 may recommend that a female patient 104 begin, or continue, treatment with estrogen or other hormone after a denervation treatment. Such hormone therapy may provide a benefit for a female patient 104 who has had a denervation treatment for hypertension or other cardiovascular condition. The denervation treatment prediction system 100 may recommend a particular dosage of estrogen or other hormone, or a particular set of dosages of estrogen or other hormone across a period of time after denervation.

[0154] As described below, an app may be provided that the patient 104 may download to their smartphone, tablet, computer, or other device. Communication between the denervation treatment prediction system 100 and the app may be useful in tracking and / or updating the medication suggestions described above. The app may provide additional patient specific data 102 to the denervation treatment prediction system 100, such as blood pressure. The blood pressure may be automatically transmitted to the denervation treatment prediction system 100 such as by a smart watch or a smart ring, and / or may be input into the app by the patient 104 after obtaining their blood pressure from a different device. The denervation treatmentprediction system 100 may utilize such post-treatment patient specific data 102 in order to fine tune its suggestions to that patient 104, and / or to change those suggestions altogether. Such post-treatment suggestions to the patient 104 may be mediated by the clinician 108, so that the clinician 108 can utilize their professional expertise to ensure that the denervation treatment prediction system 100 is making reasonable suggestions to the patient 104. The app may allow the denervation treatment prediction system 100 to monitor compliance by the patient 104 with post-treatment medication and other suggestions. Additionally, the patient specific data 102 may be anonymized and added to the corpus of patient population data 111, allowing the denervation treatment prediction system 100 to make even more accurate predictions and suggestions in the future. FIG. 9A is a flow diagram of an example process for training the denervation procedure prediction model 204 of FIG. 2. For convenience, the process 900 is described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1, appropriately programmed in accordance with this specification, may perform the process 900.

[0155] At block 902, the denervation treatment prediction system 100 receives a training data set for a set of example patients 104. For each of the example patients 104, the training data includes: (i) example patient specific data 102 (such as but not limited to physiological data and / or medical images as described above with regard to FIG. 1) for the example patient 104,(ii) one or more example denervation procedures and characteristics thereof (such as but not limited to, target organ(s), location(s), intensity, and / or dose,) for the example patient 104, and(iii) one or more target predictions for the example patient 104. The target predictions may include observed procedure effects, such as but not limited to physiological changes and / or procedure success.

[0156] As a specific example, for any given patient 104, the denervation treatment prediction system 100 may obtain one or more anatomical, physiological, or pathological characteristics, including medication history, for the given patient 104. The system can also obtain 24-hour ambulatory blood pressure (systolic and / or diastolic) and heart rate data for both baseline and follow-up time points after the denervation procedure. In this example, the target outcome to represent the response effectiveness of the denervation treatment for the given patient 104 may include, for example, the differences between the pre and post-treatment (or follow-up) measurements of one or more of i) the average daytime systolic ambulatory blood pressure, ii) the average 24-hour systolic ambulatory blood pressure, iii) the average daytime diastolic ambulatory blood pressure, iv) the average 24-hour diastolic ambulatory blood pressure, or v) the number of medication after denervation treatment. Optionally, the denervation treatmentprediction system 100 may generate a classification of the patient 104 as a responder or a nonresponder could be determined using one or more combinations of factors listed above with a certain threshold. That is, the denervation treatment prediction system 100 may classify the patient 104 as a responder if one or more of the above differences exceed a respective threshold.

[0157] At block 904, the denervation treatment prediction system 100 trains the denervation procedure prediction model 204 on the received training data. In particular, the denervation treatment prediction system 100 trains the denervation procedure prediction model 204 on a training objective, e.g., a training loss, that causes the denervation procedure prediction model 204 to replicate the target predictions for the example patients 104 when processing the corresponding example patient specific data 102 and example denervation procedures. As an example, the denervation treatment prediction system 100 may train the denervation procedure prediction model 204 using a reconstruction objective e.g., mean squared error and / or ridge regression loss) for the target predictions. As another example, the denervation treatment prediction system 100 may train the denervation procedure prediction model 204 to maximize a likelihood of generating the target predictions when processing the example patient specific data 102. The denervation treatment prediction system 100 may train the denervation procedure prediction model 204 on the training objective using any appropriate technique, e.g., stochastic gradient descent with backpropagation when the denervation procedure prediction model 204 is a neural network or through gradient boosting when the denervation procedure prediction model 204 is a decision tree-based model.

[0158] In some implementations, the denervation treatment prediction system 100 may train the denervation procedure prediction model 204 to replicate the example denervation procedures from the training data when suggesting denervation procedures based on the example patient specific data 102.

[0159] For example, depending on the types of predictions generated by the denervation procedure prediction model 204, the denervation treatment prediction system 100 may train the denervation procedure prediction model 204 as a regression model or a classification model with the collected features according to the prediction target outcome, including but not limited to the effectiveness of denervation procedure in reducing blood pressure, the reduction in the number of medication, and so on.

[0160] In some implementations, the denervation treatment prediction system 100 may train the denervation procedure prediction model using reinforcement learning techniques, such as but not limited to Q learning, policy gradients, and / or proximal policy optimization.

[0161] After training, at block 906, the denervation treatment prediction system 100 receives patient specific medical data 102 for a given patient 104, as described earlier with regard to FIG. 1.

[0162] At block 908, the denervation treatment prediction system 100 produces denervation treatment predictions for the patient 104. Examples of generating the denervation treatment predictions are described above with reference to FIGS. 2-5.

[0163] In some implementations, at block 910, the denervation treatment prediction system 100 may receive target treatment predictions for the patient 104. As an example, the target treatment predictions may include observable physiological effects (e.g., changes in blood pressure, tissue damage, neurological activity, and / or changes in blood flow) resulting from one or more denervation procedures performed for the patient 104. As another example, the target treatment predictions may include measures of success of one or more denervation procedures performed for the patient 104.

[0164] At block 912, the denervation treatment prediction system 100 may update the denervation procedure prediction model 204 based on the target treatment predictions for the patient 104. Here, “updating” the denervation procedure prediction model 204 refers to further training the denervation procedure prediction model 204 using the target treatment predictions. For example, the denervation treatment prediction system 100 may update the denervation procedure prediction model 204 by performing gradient descent for an objective function based on a difference between the predictions for the patient 104 and the target treatment predictions for the patient 104. As another example, the denervation treatment prediction system 100 may update the denervation procedure prediction model 204 to increase a likelihood of producing the target treatment predictions for the patient 104.

[0165] In some implementations, at block 914, the denervation treatment prediction system 100 may update the training data set by adding some or all of the patient specific data 102 and the target treatment predictions for the patient 104 to the training data set. In some implementations, to help preserve patient privacy, the denervation treatment prediction system 100 uploads anonymized or censored versions of the patient specific data 102 and target treatment predictions for the patient 104 to the training data set.

[0166] Thus, in some implementations, the denervation treatment prediction system 100 performs “dynamic” training by performing an adaptive learning process that allows continuous parameter updates of the denervation procedure prediction model 204 as new data becomes available.

[0167] FIG. 9B shows an example 950 of the dynamic training of the denervation procedure prediction model 204.

[0168] In particular, FIG. 9B shows how the denervation treatment prediction system 100 can ensure an adaptive learning system by dynamically re-ranking input features, e.g., features in the patient specific data 102, and optimizing the training of the denervation procedure prediction model 204 as the available data evolves.

[0169] Generally, performing the dynamic training will help update the model efficiently, e.g. , allowing for real-time updates with a new data set, by incorporating a hierarchical structure of feature sets. In particular, in the example 950, the features include (but are not limited to) primary features (high weight) 960, secondary features (low weight) 970, and tertiary features (no weight) 980.

[0170] In particular, the denervation treatment prediction system 100 can perform an original training step 952 on an original training data set.

[0171] The original training may conventional traditional feature selection and validation steps, creating the primary, secondary and tertiary features.

[0172] When new training data 954 becomes available, the denervation treatment prediction system 100 may evaluate the performance using primary features and compare it with the performance from the secondary and tertiary features. Then, the denervation treatment prediction system 100 may re-rank the features and perform model training and hyperparameter tuning with the re-ranked features.

[0173] This dynamic training approach can ensure robust prediction performance of the trained model with respect to new data and provide a stable service in prediction.

[0174] For example, the denervation treatment prediction system 100 may select the optimal feature sets based on evaluation metrics such as, but not limited to, the area under the Receiver Operating Characteristic curve (AUC) or partial AUC over a specific range of interest for sensitivity and specificity e.g., sensitivity > threshold and specificity > threshold), F-score, overall accuracy, sensitivity and specificity, and so on.

[0175] In some implementations, the denervation treatment prediction system 100 omits performing simulations based on patient specific data 102 from the process for predicting denervation procedure effects for a patient 104. In such implementations, the denervation treatment prediction system 100 may predict denervation procedure effects for a patient 104 based on specific predictive features that are produced based on patient specific data 102. These specific predictive features may include the variability of the slope of systolic blood pressure,the maximum rate of change of the slope of systolic blood pressure, the minimum rate of change of the slope of systolic blood pressure, the variability of the rate of change of the slope of pulse pressure, and the mean rate of change of the slope of heart rate.

[0176] FIG. 10 is a plot illustrating the performance of an example implementation of the described techniques. The example implementation does not include a simulation system. In particular, in the example of FIG. 10, the denervation procedure prediction model 204 of the denervation treatment prediction system 100 is a machine learning model, specifically an XGBoost regression model, that was developed to predict the response of patients 104 to a renal denervation (RDN) procedure. In this implementation, the denervation procedure prediction model 204 was trained using example patient specific data 102 consisting of the demographics, home and office-based measurements, and ambulatory blood pressure monitoring (AB PM) data of two hundred and fifty patients who underwent ultrasound RDN procedures. The AB PM data consisted of 802 physiological AB PM parameters, such as twenty- four-hour continuously-monitored systolic blood pressure, pulse pressure, and heart rate. The target predictions for the example patients 104 that were used to train the denervation procedure prediction model 204 were physiological blood pressure reactions, which were standardized and calculated using ABPM data. Responders were defined as those with a > 5 mmHg drop from baseline to two months. Five-fold cross-validation was repeated 100 times to evaluate the Area Under the Curve (AUC) value as an evaluation metric.

[0177] Having been thus trained, the denervation procedure prediction model 204 of this implementation uses two clinical and five physiological ABPM parameters as predictive features to predict effects of RDN procedures on patients 104. The two clinical parameters used are orthostatic hypertension and home pulse pressure; while the five physiological ABPM parameters used are the variability of the slope of systolic blood pressure, the maximum rate of change of the slope of systolic blood pressure, the minimum rate of change of the slope of systolic blood pressure, the variability of the rate of change of the slope of pulse pressure, and the mean rate of change of the slope of heart rate. The denervation procedure prediction model 204 demonstrated high accuracy in the validation cohort, as indicated by an AUC of 0.776 (95% confidence interval (CI): 0.734-0.810), compared with a model that used only the clinical parameters as predictive features (AUC: 0.650, 95% CI: 0.616-0.688).

[0178] In particular, FIG. 10 is a plot 1000 illustrating the performance of the denervation procedure prediction model 204 of this example implementation as compared to a model that uses only the two conventional parameters of orthostatic hypertension and home pulse pressure (i.e., it does not incorporate any ABPM parameters). Line 1002 represents the change in thesensitivity with increasing specificity of the denervation procedure prediction model 204 used in this implementation. Line 1004 represents the change in the sensitivity with increasing specificity of a model using only the two conventional parameters. The AUC of line 1002 is larger than that of line 1004, demonstrating the superior performance of the denervation procedure prediction model 204 of this implementation of predicting effects of RDN procedures on patients. In particular, the AUC of line 1002 is larger by 0.13 than the AUC of line 1004.

[0179] FIG. 11 is a plot further illustrating the performance of the example implementation of the described techniques of predicting effects of RDN procedures on patients. There are four distinct points along the x-axis of plot 1100, each representing the predicted probability that a given patient 104 would respond to a RDN procedure. For example, the left-most point on the x-axis represents the denervation procedure prediction model 204 predicting that a given patient 104 would have a 0-25% chance of responding to an RDN procedure. The next point on the x-axis (second from the left) represents the denervation procedure prediction model 204 predicting that a given patient 104 would have a 25-50% chance of responding to an RDN procedure. The next point on the x-axis (third from the left, second from the right) represents the denervation procedure prediction model 204 predicting that a given patient 104 would have a 50-75% chance of responding to an RDN procedure. Finally, the right-most point on the x- axis represents the denervation procedure prediction model 204 predicting that a given patient 104 would have a 75-100% chance of responding to a RDN procedure.

[0180] The y-axis of plot 1100 represents the blood pressure drop in mmHg for a given patient 104 from the baseline blood pressure of that patient 104 to two months later. Blocks 1102, 1104, 1106, and 1108 thus each represent the blood pressure drop experienced by patients (on average, subject to the indicated confidence intervals) with the given predicted probability of responding to a RDN procedure. For example, block 1102 indicates that patients who were predicted by the denervation procedure prediction model 204 to have a 0-25% chance of responding to a RDN procedure experienced a blood pressure drop of about -4 mmHg, subject to the indicated confidence interval 1101. Likewise, block 1104 indicates that patients who were predicted by the denervation procedure prediction model 204 to have a 25-50% chance of responding to a RDN procedure experienced a blood pressure drop of about 2.5 mmHg, subject to the indicated confidence interval 1103. Block 1106 indicates that patients who were predicted by the denervation procedure prediction model 204 to have a 50-75% chance of responding to a RDN procedure experienced a blood pressure drop of about 8 mmHg, subject to the indicated confidence interval 1105. Block 1108 indicates that patients who werepredicted by the denervation procedure prediction model 204 to have a 75-100% chance of responding to a RDN procedure experienced a blood pressure drop of about 12 mmHg, subject to the indicated confidence interval 1107.

[0181] Thus, plot 1100 illustrates a correlation between the prediction made by the denervation procedure prediction model 204 of the probability that a patient 104 would respond to a RDN procedure and the resulting blood pressure drop experienced by the patient 104. Specifically, patients 104 who were predicted to have larger probabilities of responding to a RDN procedure tended to experience larger drops in blood pressure. Since the magnitude of the drop in blood pressure generally correlates with the effectiveness of the procedure, the correlation illustrated in plot 1100 demonstrates the ability of the denervation procedure prediction model 204 to accurately predict the effectiveness of a RDN procedure for a patient 104 based on the relevant predictive features of that patient 104.

[0182] As will be described in more detail below, the denervation treatment prediction system 100 may additionally or instead generate treatment information from 2D medical images of the patient. This treatment information may then be presented to the clinician 108, e.g., prior to, during, or both the procedure. This may provide additional relevant information to the clinician 108, resulting in improved efficacy of treatment.

[0183] FIG. 12 is a flow diagram of an example process for generating treatment information from 2D medical images. For convenience, the process 1200 is described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1, appropriately programmed in accordance with this specification, may perform the process 1200.

[0184] At block 1202, the denervation treatment prediction system 100 acquires one or more 2D medical images of the interior of the body of the patient 104. Such images may be acquired from an X-ray machine, an MRI machine, a CT machine, an ultrasound machine, or any other suitable medical imaging machine. Such 2D medical images may be acquired at a point in time before a procedure is to be performed on the patient 104. For example, the patient 104 may be imaged at a cath lab, with or without contrast agent, days or weeks prior to a scheduled denervation procedure. For example, these 2D medical images may be included as part of the patient specific data 102 of FIG. 1. As will be described in more detail below, because of the use of the denervation treatment prediction system 100, there may be a reduced need for contrast agent when capturing these images.

[0185] In some implementations, the denervation treatment prediction system 100 acquires a single medical image for a given patient 104.

[0186] In some other implementations, the denervation treatment prediction system 100 acquires multiple 2D medical images for the same patient 104. For example, the multiple different medical images may be taken from different angles by the same imaging device, e.g., simultaneously or at different times. As another example, the multiple images may be taken from the same angle, but are collected at different times, e.g., over the course of some motion by the patient 104 e.g., breathing).

[0187] In some implementations, at block 1204, the denervation treatment prediction system 100 performs image pre-processing on the medical images acquired at block 1202. Image preprocessing is the process of manipulating raw image data into a usable and meaningful format. It provides for the reduction or elimination of unwanted distortions in the raw image data and enhances specific qualities useful for a particular machine learning / computer vision application such as the denervation treatment prediction system 100. One or more preprocessing processes may be performed at block 1204. One example preprocessing process is resizing each 2D medical image to a uniform size. Another example preprocessing process is gray scaling, which converts color 2D medical images to grayscale. Gray scaling simplifies the 2D medical image data and additionally may reduce the computational needs for machine learning / computer vision applications. Another example preprocessing process is noise reduction. Smoothing, blurring, and filtering techniques may be applied to the 2D medical images to remove unwanted noise from them. Another example preprocessing process is normalization, which adjusts the intensity values of pixels to a desired range, such as between 0 to 1. Normalization may improve the performance of machine learning / computer vision models. Another example preprocessing process is binarization, which is the process of converting grayscale 2D medical images to black and white images by thresholding, thereby reducing precision to 1 bit by representing integer values -1 and 1 with binary values 0 and 1 respectively. Binarization may improve the performance of machine learning / computer vision models by simplifying operations therein to binary operations, generally resulting in a reduction in complexity. Another example preprocessing process is contrast enhancement, in which the contrast within a 2D medical image is adjusted using histogram equalization. Contrast enhancement improves the visibility of details in an image by redistributing pixel intensities, stretching the contrast range of an image, making it useful for medical imaging so that subtle features may be more easily identified.

[0188] Optionally, at block 1206, the denervation treatment prediction system 100 applies adaptive Frangi filtering to the 2D medical images, e.g., that were preprocessed at block 1204 or that were obtained at block 1202. Frangi filtering is an effective technique to enhance andextract tubular structures in 2D medical images (Frangi, A.F., et. al. (1998); Multiscale vessel enhancement filtering. In: Wells, W.M., Colchester, A., Delp, S. (eds) Medical Image Computing and Computer-Assisted Intervention — MICCAI’98. MICCAI 1998. Lecture Notes in Computer Science, vol 1496. Springer, Berlin, Heidelberg. https: / / doi.org / 10.1007 / BFb0056195). The Frangi filtering process involves analyzing the local intensity and Hessian matrix of a 2D medical image to detect vessel-like structures, making it useful for vessel segmentation and analysis.

[0189] At block 1208, for each medical image, the denervation treatment prediction system 100 generates a segmentation of the medical image. For example, the denervation treatment prediction system 100 may process an input that includes the medical image, i.e., the output of block 1202, 1204, or 1206, using a neural network, e.g., a CNN or ViT, to generate a segmentation mask over the medical image.

[0190] The segmentation mask may designate each pixel of the medical image as either depicting a blood vessel or not depicting a blood vessel. That is, the denervation treatment prediction system 100 processes an input that includes (i) the medical image, i.e., the medical image obtained at block 1202 or a pre-processed version obtained at block 1204, and, (ii) optionally, a version of the medical image with Frangi filtering applied using a neural network to generate as output a segmentation map that segments the blood vessels depicted in the medical image. For example, the neural network may generate an output that assigns, to each pixel, a score, e.g., a probability, that indicates the likelihood that the pixel depicts a blood vessel. The denervation treatment prediction system 100 may then determine that any pixel with a score that exceeds a specified threshold, e.g., .5, .7, or .9, depicts a blood vessel and that any pixel with a score that does not exceed the specified threshold does not depict a blood vessel. In some implementations, including the version of the medical image with Frangi filtering applied as part of the input may improve the ability of the neural network to accurately segment the blood vessels, e.g., by providing a reference estimate of the target segmentation. Training such a neural network is described in more detail below with reference to FIG. 13.

[0191] As described above, in some implementations, the denervation treatment prediction system 100 obtains multiple different medical images. In these implementations, the denervation treatment prediction system 100 may combine the segmentations generated for the multiple different medical images to generate a final segmentation.

[0192] For example, when the multiple different medical images are taken from different angles, the denervation treatment prediction system 100 may combine the segmentations by registering each pixel in each image in a common coordinate system, e.g., the coordinatesystem of a designated one of the multiple images and averaging or otherwise combining the scores for each pixel that is registered to the same point in the common coordinate system before comparing the resulting combined scores to the threshold.

[0193] As another example, when the multiple images are taken from the same angle, but are collected at different times, e.g., over the course of some motion by the patient 104 (e.g., breathing), the denervation treatment prediction system 100 may combine the segmentation for the multiple frames in a way that corrects for the motion of the patient 104, e.g., by averaging the masks together in a way that reduces noise in the segmentation output caused by the motion of the patient 104. One or more registration methods may be used to combine the multiple images, including optical flow based on the intensity tracking of images through consecutive frames (e.g., fluoroscopy images), or point patchings based on key points using SIFT (Scale- Invariant Feature Transformation).

[0194] At block 1210, the denervation treatment prediction system 100 identifies features of the blood vessel(s) in the 2D medical image(s) from the segmentation output. For example, the denervation treatment prediction system 100 may identify the boundaries, e.g., the edges of the blood vessel walls, of the blood vessels in the image. Optionally, the features may also include, e.g. , the areas of the blood vessels in the image. The geometric features e.g. , length, diameter, curvature, torsion, etc.) can be computed from the extracted segmentation of the lumen boundary of the blood vessel(s). In some implementations, the denervation treatment prediction system 100 performs post-processing on the segmentation output prior to computing the features, post-processing on initial features computed from the segmentation output, or both. For example, post-processing may include centerline extraction and smoothing of centerline via Fourier smoothing or similar processes. Once the tangential direction is determined based on the centerline, a diameter of each blood vessel(s) can be calculated by detecting the intersection between the orthogonal line and the lumen boundary. For a robust measure in all vascular networks, including bifurcation or trifurcation of blood vessels, an inscribed circle can be employed to measure diameter.

[0195] At block 1212, the denervation treatment prediction system 100 generates information about recommended treatment from the features of the blood vessels in the medical image(s) and displays the information to the clinician 108. That generation is performed using the identification from block 1210 and, optionally, the output from the denervation treatment prediction system 100. In some examples, the generation also makes use of the corpus of patient population data 111.

[0196] In some implementations, the denervation treatment prediction system 100 may integrate additional information from one or more additional sources when generating the treatment information. For example, the additional information may include respective locations of one or more nerves within the body of the patient 104. As another example, the additional information may include the location of a stenosis area. As a particular example, the locations of a stenosis area may be relevant as an area to avoid for ablation, e.g., because calcification may interfere with ablation. The information may include any of a variety of information that relates to performing a treatment, e.g., an ablation, on the patient 104. For example, the information may include an indication of one or more suggested locations of where to perform a procedure, e.g., an ablation. For example, to generate the one or more suggested locations, the denervation treatment prediction system 100 may utilize one or more different factors taken from the features of the blood vessels and, in same cases, other patient specific data 102 and / or from the corpus of patient population data 111 relating to the overall patient population. Relevant factors to consider in determining one or more denervation locations include the anatomical identity of a blood vessel, the diameter of a blood vessel at different locations along its length, the locations where a blood vessel branches, the presence of calcification at one or more locations on an inner surface of a blood vessel and / or within the wall of the blood vessel, the presence of stenosis at one or more locations within a blood vessel, the presence of a stent at one or more locations within a blood vessel, the presence of an aneurysm in a blood vessel, the presence of tumors, fibromuscular disease, and / or other disease conditions in or in proximity to a blood vessel, and / or the proximity of the blood vessel to other blood vessels and / or organs. The presence of calcification at one or more locations in a blood vessel to be treated may be determined by any suitable method, including angiography, intravascular ultrasound, and / or blood flow measurement. Because calcification decreases the diameter of a blood vessel, that decrease in diameter reduces the blood flow in that blood vessel compared to a blood vessel without calcification, and an amount of calcification can be inferred from that blood flow rate. The denervation treatment prediction system 100 may compare the patient specific data 102 to the corpus of patient population data 111 to determine whether the blood flow rate through a particular blood vessel is consistent with a blood vessel that is relatively free of calcification at a location, or a blood vessel that is calcified at that location.

[0197] The denervation treatment locations may include two or more denervation treatment locations within a single main renal artery. Where ultrasound is delivered intravascularly for denervation treatment, a larger number of ablations may increase the probability of better outcomes for the patient 104. Conventionally, two to three ablations may be performed in asingle main renal artery. However, by utilizing the corpus of patient population data 111 in addition to the patient specific data 102, the denervation treatment prediction system 100 considers whether a number of ablations greater than two or three is contraindicated for a specific patient 104. If the patient 104 is not contraindicated for more than two to three ablations, the denervation treatment prediction system 100 may suggest four, five, six, or more denervation treatment locations inside a single main renal artery and / or branch and / or accessory renal artery.

[0198] In other embodiments, the denervation treatment prediction system 100 utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to consider whether a number of ablations less than two is indicated for a specific patient 104.

[0199] In one example, an ultrasound transducer 628, which will be described in more detail below, may be substantially 6 mm in length and is used to deliver unfocused ultrasound energy outwardly from an intravascular location, allowing it to generate lesions of substantially 5 mm in length, as measured along the longitudinal centerline of the renal artery in which the ultrasound transducer 628 is placed, during each actuation thereof. In other embodiments, the transducer can be shorter or longer than 6 mm in length. The ultrasound transducer 628 may be moved a short distance longitudinally along the renal artery after each actuation, such that two to three ablations performed by the ultrasound transducer 628 may result in lesions that extend substantially 1 to 1.5 cm along the main renal artery. Each additional ablation thus adds substantially 5 mm to the total length of the lesions along the main renal artery. More generally, for an ultrasound transducer 628 of length Lt, where the ultrasound transducer 628 is moved adjacent to its previous position after each ablation, the total length of the lesion is approximately Lt* N, where N is the number of individual ablations. The denervation treatment prediction system 100 may suggest that number N be increased or decreased in order to increase the probability of better outcomes (i.e., effective and / or safe) for the patient 104. Even more generally, for an energy emitter having length Lt, where the energy emitter is moved adjacent to its previous position after each ablation, the total length of the lesion is approximately Lt* N, where N is the number of individual ablations. Such an energy emitter may additionally or alternatively include one or more RF electrodes placed in contact with an interior of a vessel wall, a microwave emitter, a laser or other optical emission source, and / or an outlet or outlets for cryoablation fluid or chemical agents.

[0200] The denervation treatment locations may include locations in different arteries, veins, and / or other locations utilized to treat nerves associated with different organs and / or arteries. Such locations may vary based on the type of therapy that is needed by the patient 104. As oneexample, for a patient 104 with hypertension, the denervation treatment prediction system 100 may suggest one or more denervation treatment locations for denervation of one or more renal nerves (each associated with a renal artery), as well as one or more denervation treatment locations for denervation of at least a portion of the celiac plexus and / or at least a portion of the aorticorenal ganglia and / or ganglionated plexuses and / or other nerves in communication with the central sympathetic nervous system, for example one or more hepatic and / or splanchnic nerves. By utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to make suggestions, the denervation treatment prediction system 100 is capable of making suggestions of denervation treatment locations, along with making predictions of success for the procedure, that the clinician 108 might not consider on their own.

[0201] With regard to the suggestions for denervation treatment locations, the denervation treatment prediction system 100 may utilize one or more different factors taken from patient specific data 102 and / or from the corpus of patient population data 111 to suggest an optimized intensity, power, and duration (referred to above as the dose) of energy to be applied at each suggested denervation treatment location. The optimal dose may be different at one or more of the suggested denervation treatment locations. By utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to make suggestions, the denervation treatment prediction system 100 is capable of making suggestions of a denervation treatment dose at each location, along with making predictions of success for the procedure, than the clinician 108 may be able to determine on their own.

[0202] Similarly to the above, the information may include an identification of one or more treatment-relevant locations. For example, the information may include an optimal insertion location for a transducer, catheter or other device.

[0203] As another example, the information may include a recommendation of a prophylactic medicine or another medication to be administered to the patient 104, a dose of the medication to be given at each suggested ablation location, or both. As an example, by utilizing the corpus of patient population data 111 in addition to the patient specific data 102, the denervation treatment prediction system 100 may recommend medications and medication dosages for the patient 104 to improve treatment outcomes, and / or may predict the results of one or more changes in medications and / or medication doses after treatment. As one example, the denervation treatment prediction system 100 may recommend one or more changes in dosage of the existing hypertension medication taken by the patient 104 after a denervation procedure. Such changes may be suggested as a single step function, or as suggestions for multiple changes over weeks or months following a denervation procedure. As another example, the denervationtreatment prediction system 100 may recommend that the patient 104 cease taking one or more specific medications at certain times after a denervation procedure. As another example, the denervation treatment prediction system 100 may recommend that the patient 104 titrate down the dosage of one or more of the hypertension medications. As another example, the denervation treatment prediction system 100 may recommend that the patient 104 change the class of one or more hypertension medications previously taken by the patient 104 after the denervation treatment. Hypertension medications are typically divided into different classes based on their mechanisms of action; such classes include diuretics, beta-blockers, angiotensinconverting enzyme (ACE) inhibitors, angiotensin II receptor blockers (ARBs), calcium- channel blockers, alpha blockers, alpha-2 receptor agonists, and vasodilators. After denervation for treatment of hypertension, the denervation treatment prediction system 100 may determine that the patient 104 would benefit from changing at least one class of their previous hypertension medication, from eliminating at least one class of their previous hypertension medication, and / or from adding at least one class to their previous hypertension medication.

[0204] As described above, estrogen and other hormone levels are highly correlated to cardiovascular risk over the female life span. As a result, the denervation treatment prediction system 100 may recommend that a female patient 104 begin, or continue, treatment with estrogen or other hormone after a denervation treatment. Such hormone therapy may provide a benefit for a female patient 104 who has had a denervation treatment for hypertension or other cardiovascular condition. The denervation treatment prediction system 100 may recommend a particular dosage of estrogen, or a particular set of dosages of estrogen across a period of time after denervation.

[0205] As described below, an app may be provided that the patient 104 may download to their smartphone, tablet, computer, or other device. Communication between the denervation treatment prediction system 100 and the app may be useful in tracking and / or updating the medication suggestions described above. The app may provide additional patient specific data 102 to the denervation treatment prediction system 100, such as blood pressure. The blood pressure may be automatically transmitted to the denervation treatment prediction system 100 such as by a smart watch or a smart ring, and / or may be input into the app by the patient 104 after obtaining their blood pressure from a different device. The denervation treatment prediction system 100 may utilize such post-treatment patient specific data 102 in order to fine tune its suggestions to that patient 104, and / or to change those suggestions altogether. Such post-treatment suggestions to the patient 104 may be mediated by the clinician 108, so that theclinician 108 can utilize their professional expertise to ensure that the denervation treatment prediction system 100 is making reasonable suggestions to the patient 104. The app may allow the denervation treatment prediction system 100 to monitor compliance by the patient 104 with post-treatment medication and other suggestions. Additionally, the patient specific data 102 may be anonymized and added to the corpus of patient population data 111, allowing the denervation treatment prediction system 100 to make even more accurate predictions and suggestions in the future.

[0206] As another example, the information may include a recommendation of whether to proceed with a given treatment. For example, the denervation treatment prediction system 100 may recommend against denervation altogether based on the condition of the blood vessel(s) that would be treated with denervation, because proceeding with denervation would not provide a benefit to the patient 104 or would be unsafe at a required dosage. As another example, the denervation treatment prediction system 100 may include a recommendation of an adjustment between an ablation treatment and a bifurcation treatment, e.g., based on the condition of the blood vessel(s) as reflected by the features.

[0207] As another example, the information may include an estimate of the energy that will be required to perform a treatment. For example, this information may include optimized intensity, power, and duration (referred to above as the dose) of energy to be applied at each suggested denervation treatment location. The optimal dose may be different at one or more of the suggested denervation treatment locations. By utilizing the corpus of patient population data 111 in addition to the patient specific data 102 to make suggestions, the denervation treatment prediction system 100 is capable of making suggestions of a denervation treatment dose at each location, along with making predictions of success for the procedure, than the clinician 108 may be able to determine on their own.

[0208] Without wishing to be bound to a particular theory, in some patients 104, the renal nerve(s) may be easier to treat within one or more distal branches at the distal end of the renal artery, particularly where RF electrodes are used for ablation that have an average ablation depth of 2 to 4 mm. In some patients 104, the renal nerve(s) may be easier to treat immediately proximal to the distal branch(es) at the distal end of the renal artery, particularly where an ultrasound transducer, which can have longer ablation depths, is used for ablation. The denervation treatment prediction system 100 may utilize patient specific data 102 that includes imaging data of the renal artery, nerves, and / or nearby organs and structures, such as the kidneys and large intestine, in combination with the corpus of patient population data 111, to determine whether at least one denervation treatment location should be within a distal branchof the renal artery, and / or within the renal artery immediately proximal to the distal branch(es). Further, the denervation treatment prediction system 100 may utilize patient specific data 102 that includes imaging data of the renal artery, nerves and nearby organs and structures, such as the kidneys and large intestine, in combination with the corpus of patient population data 111, to determine the outward extent of the treatment volume from the energy emitter utilized. For example, the denervation treatment prediction system 100 may suggest a dose that results in a treatment area extending outward from the lumen of a vessel a distance from 1 mm to 10 mm.

[0209] The renal artery, as well as other blood vessels, may change its tissue composition along at least a portion of its length. The patient specific data 102 may include properties of that tissue measured along at least a portion of that renal artery or other blood vessel, such as but not limited to the presence and / or amount of fatty tissue and its location(s), the presence and / or amount of collagen and its location(s), the presence and / or amount of muscular fiber and its location(s), and / or the thickness of the intima, media, and / or adventitia along the length of the renal artery and / or other blood vessel and / or the stiffness along the length of the blood vessel.

[0210] The denervation treatment prediction system 100 may suggest one or more denervation locations away from locations in the renal artery or other blood vessel where the media is thin in order to reduce the risk of stenosis at the denervation location(s).

[0211] The denervation treatment prediction system 100 may suggest one or more denervation locations away from locations in the renal artery or other blood vessel where the media is thick, when the energy emitter is an ultrasound transducer, in order to reduce the risk of stenosis at the denervation location(s). For example, in applications using ultrasound, the denervation treatment prediction system 100 may suggest increasing the flow rate in order to protect a greater depth in the near field where the media is thicker.

[0212] In general, the denervation treatment prediction system 100 may suggest one or more denervation locations that are not at locations where treatment may increase the risk of stenosis above a particular level.

[0213] In making suggestions for one or more denervation treatment locations, the denervation treatment prediction system 100 may take into account the presence or absence of one or more heat sinks in proximity to a treatment site. Heat sinks include veins and lymph nodes, which include liquid that can absorb energy from an energy emitter more quickly than, or in addition to, tissue in the treatment volume, which could result in less effective treatment. If a heat sink is present in proximity to a treatment site, and ablation is to be performed using an ultrasound transducer 628 to deliver ultrasound energy to the treatment site, the denervation treatment prediction system 100 may increase the power delivered to the ultrasound transducer 628 inorder to increase the power emitted from the ultrasound transducer 628 during treatment, as compared to the power emitted from the ultrasound transducer 628 during treatment where a heat sink is not present in proximity to the treatment site. If a heat sink is present in proximity to a treatment site, and ablation is to be performed using RF electrodes, the denervation treatment prediction system 100 may utilize that information to select a different location for the application of denervation treatment, because increasing power to RF electrodes does not overcome the loss of energy to a heat sink, and the increase in power to RF electrodes may cause damage to the inner wall of the blood vessel in contact with the RF electrodes.

[0214] As another example, the information may include a guidance path for a catheter, transducer, or other device after being inserted at the optimal insertion location. An example of a guidance path is illustrated below with reference to FIG. 13.

[0215] As another example, the information may include about the health of a vessel. For example, the information about the health of the vessel can be determined by the flow of contrast that has been injected.

[0216] As another example, the information may include a prediction about the effectiveness or impact of a given treatment. The denervation treatment prediction system 100 may utilize factors such as the locations of denervation treatment, patient specific data 102 such as but not limited to BMI, AB PM, and gender of the patient 104, and compare those to the corpus of patient population data 111 as part of predicting a treatment outcome. The denervation treatment prediction system 100 may predict a treatment outcome in terms of a predicted drop in baseline blood pressure. That drop may be in systolic blood pressure, diastolic blood pressure, or both. The denervation treatment prediction system 100 may predict an immediate treatment outcome, and / or a treatment outcome three months, six months, and / or other interval after treatment. As described below, an app may be provided that the patient 104 may download to their smartphone, tablet, computer, or other device. Communication between the denervation treatment prediction system 100 and the app may be useful in tracking the physiological characteristics of the patient 104 and may be utilized by the denervation treatment prediction system 100 to update its prediction of a treatment outcome for a patient 104 over time. Predictions made further out in time may be less accurate due to the introduction of confounding factors such as a change in medication administered to a patient 104, one or more additional unexpected disease states of the patient 104, and / or other factors. By communicating these facts to the denervation treatment prediction system 100 via the app or otherwise, the denervation treatment prediction system 100 is able to update its predictions of future outcomes accordingly.

[0217] FIG. 13 shows an example 1300 of treatment information being provided for display to a clinician 108. In particular, in the example 1300, the treatment information is overlay ed over a medical image of the vasculature of the patient 104.

[0218] As shown in the example 1300, the treatment information includes guidance lines 1302 that show a suggested path for a catheter that is inserted as part of a treatment. The suggested path may also display suggested treatment locations for denervation.

[0219] The treatment information also includes various information about relevant blood vessels. For example, the treatment information includes the artery diameter and length at three different points A, B, and C in the main renal artery 1304 of the patient 104. Moreover, the treatment information includes the artery diameter and length at another point D in a first branch 1308 of the artery and the artery diameter and length at two points E and F in a second branch 1306 of the artery.

[0220] FIG. 14 is a flow diagram of an example process for training a neural network to perform segmentation of blood vessels. For convenience, the process 1400 is described as being performed by a system of one or more computers located in one or more locations. For example, the denervation treatment prediction system 100 of FIG. 1, appropriately programmed in accordance with this specification, may perform the process 1400.

[0221] At block 1402, the denervation treatment prediction system 100 obtains a data set of training medical images. The medical images may be of any suitable type, e.g., one of the types described above with reference to block 1202.

[0222] At block 1404, the denervation treatment prediction system 100 generates a respective label for each of the training medical images. The label for a given training medical image identifies, for some or all of the pixels of the image, whether the pixel depicts a blood vessel or not.

[0223] In particular, to generate the label for a given training medical image, the denervation treatment prediction system 100 may pre-process the medical image as described above with reference to block 1204. In some implementations, the denervation treatment prediction system 100 may then provide the pre-processed medical image for presentation to a user and allow a user to submit inputs highlighting the blood vessels that appear in the images.

[0224] In some other implementations, the denervation treatment prediction system 100 may perform adaptive Frangi filtering on the pre-processed medical image as described above at block 1206 to generate the label for the corresponding medical image.

[0225] In yet other implementations, the denervation treatment prediction system 100 may provide a partially-labeled medical image that is generated by performing adaptive Frangifiltering on the pre-processed medical image for presentation to a user and allow a user to submit inputs highlighting any additional blood vessels that appear in the images and have not been identified by the Frangi filtering.

[0226] At block 1406, the denervation treatment prediction system 100 trains the neural network on training examples that each include (i) a medical image from the training data set and (ii) the label for the medical image. For example, the denervation treatment prediction system 100 may train the neural network to minimize any appropriate image segmentation loss function. Examples of image segmentation loss function that may be used include, e.g., binary cross-entropy losses and Dice losses. That is, the denervation treatment prediction system 100 may process a given medical image using the neural network to generate an output that includes respective scores for each of the pixels of the medical image and may then train the neural network, e.g., through gradient descent, on a segmentation loss function that is evaluated from (i) the scores generated by the neural network and (ii) the label for the medical image. The denervation treatment prediction system 100 can train the neural network to minimize the image segmentation loss function using any appropriate training technique, e.g., stochastic gradient descent with backpropagation.

[0227] Instead of or in addition to generating 2D segmentations of medical images to identify blood vessels, the denervation treatment prediction system 100 may generate a 3D representation of the locations, e.g., generate a 3D representation of the vasculature that is depicted in the medical images.

[0228] FIG. 15 shows an example 1500 of a 3D data pipeline that may be used by the denervation treatment prediction system 100 to generate treatment information using 3D vasculature of blood vessels.

[0229] At block 1502, 2D medical images of the interior of the patient 104 are acquired as described above with regard to block 1202. For example, a medical imaging device may capture 2D medical images at N different angles, e.g., between 2-12, e.g., between 3-7, different angles. As a particular example, 2D fluoroscopy images may be acquired from multiple angles along an arc on a C-arm configuration, which does not form a closed loop as used in reconstruction for computed tomography (CT).

[0230] For each of the multiple medical images, at block 1508, a 2D segmentation of blood vessels is generated as described above with reference to FIG. 12. In particular, as described above, this may optionally involve one or more of image pre-processing at block 1504 and / or adaptive Frangi filtering at block 1506.

[0231] Next, at block 1512, the segmented images generated at block 1408 are utilized to generate a 3D model of the vasculature of the patient 104 at a volume of interest. Blocks 1504, 1506, 1508, and / or 1512 may be performed prior to a denervation procedure, and / or may be performed intraoperatively. The 2D medical images, such as fluoroscopy images, are acquired from multiple angles during the procedure, i.e., intra-operatively.

[0232] The 3D model of the vasculature of the patient 104 may be generated using any appropriate technique. As one example, the segmented images generated at block 1404 may be processed by a machine learning model to generate the 3D model. As another example, a computer vision algorithm may be applied to the segmented images generated at block 1404 to generate the 3d model. As yet another example, one or more projection matrices may be used to project the 2d segmented images at block 1404 into 3D space in order to generate the 3D model.

[0233] FIG. 16 illustrates an example 1600 of how segmentations of different 2D images of a volume, taken at different angles, may be combined to provide a 3D representation of a blood vessel or vessel within that volume. In particular, in the example 1600, segmentations of 3 different 2D X-Ray images are projected to generate the 3D geometry of a vessel within the imaged volume.

[0234] After being generated, the constructed 3D volumetric image may be fused with preoperative high-resolution 3D images, e.g., CT or MRI (magnetic resonance imaging). The fused image may provide guidance for the catheterization procedure. For example, the operating surgeon may visualize a planned path based on the fused images. Mechanical models may be applied to the pre-operative images acquired before the procedure based on the intraoperative fluoroscopy images to account for motion and physiological changes associated with the underlying soft tissues.

[0235] Next, at block 1514, information about recommended treatment is generated, and displayed to the clinician 108 and / or user. That generation is performed using the 3D model of the vasculature of the patient 104 at a volume of interest generated at block 1512, as well as other output from the denervation treatment prediction system 100. The information about recommended treatment may include one or more suggested ablation locations, a dose to be given at each suggested ablation location, and labeled treatment-relevant locations. The information may also include any of the information described above with reference to FIG. 12.

[0236] Additionally, after block 1512, instead or in addition to block 1514, the process 1500 may perform block 1516, at which the denervation treatment prediction system 100 generatesguidance for navigating a catheter 602 to a treatment site in the body of the patient 104. Such guidance may be presented as one or more overlays that illustrate a location or locations of a suggested denervation procedure, as well as a pathway through at least part of the vasculature of the patient 104 to that location or locations. That part of the vasculature of the patient 104 may be the part of the vasculature closest to the location or locations of the suggested denervation procedure. Vascular access is generally achieved through the femoral or radial artery, and interventional clinicians are skilled at advancing a guidewire and / or catheter 602 through the vasculature from that vascular access. Block 1516 may be particularly useful in providing terminal guidance of the motion of the catheter 602 to the clinician 108, providing assistance in catheter navigation across a terminal distance in proximity to the location or locations of the suggested denervation procedure. That terminal distance may be in the range of 0.5-60 cm, 0.5-50 cm, 0.5-40 cm, 0.5-30 cm, 0.5-20 cm, 0.5-10 cm, 0.5-5 cm, 0.5-4 cm, 0.5-3 cm, 0.5-2 cm, 0.5-1 cm, 1-60 cm, 1-50 cm, 1-40 cm, 1-30 cm, 1-20 cm, 1-10 cm, 1-5 cm, 1-4 cm, 1-3 cm, 1-2 cm, 2-60 cm, 2-50 cm, 2-40 cm, 2-30 cm, 2-20 cm, 2-10 cm, 2-5 cm, 2-4 cm, 2-3 cm, 3-60 cm, 3-50 cm, 3-40 cm, 3-30 cm, 3-20 cm, 3-10 cm, 3-5 cm, 3-4 cm, 4-60 cm,4-50 cm, 4-40 cm, 4-30 cm, 4-20 cm, 4-10 cm, 4-5 cm, 5-60 cm, 5-50 cm, 5-40 cm, 5-30 cm,5-20 cm, 5-10 cm, 5-9 cm, 5-8 cm, 5-7 cm, 5-6 cm, 10-60 cm, 10-50 cm, 10-40 cm, 10-30 cm, 10-20 cm, 10-15 cm, 20-60 cm, 20-50 cm, 20-40 cm, 20-30 cm, 20-25 cm, 30-60 cm, 30-50 cm, 30-40 cm, 30-35 cm, 40-60 cm, 40-50 cm, 40-45 cm, 50-60 cm, or 50-55 cm. Providing terminal guidance for catheter navigation not only saves computational resources and reduces the number of 2D fluoroscopy images that need to be taken of the patient 104 (thereby reducing radiation exposure of the patient 104), but also avoids annoying the clinician 108 with information about catheter guidance along a long, generally- standard path along the vasculature that the clinician 108 knows from education and experience. Instead of, or in addition to, the one or more overlays that illustrate a location or locations of a suggested denervation procedure, as well as a pathway through at least part of the vasculature of the patient 104 to that location or locations, the denervation treatment prediction system 100 may provide visual “turn-by-turn” instructions on a display and / or audio “turn-by-turn” instructions via a speaker or speakers. Such “turn-by-turn” instructions may include directions for the clinician 108 to steer the distal end of the catheter 602 in a particular direction at each junction between (a) the current blood vessel in which the distal end of the catheter 602 is located and (b) a different blood vessel or branch of the current blood vessel. By providing such direct instructions, the clinician 108 may be able to focus more effectively on the procedure rather than the display.

[0237] In some embodiments, at block 1516, the process 1500 may be utilized to control a robotic, semi-robotic, robotically-enhanced, or other catheter 602 in order to place the distal end of the catheter 602 at a treatment site. Thus, rather than providing directions to a clinician 108 for navigating the distal end of the catheter 602 to a treatment site, the denervation treatment prediction system 100 may control the catheter 602 itself. Of course, the clinician 108 retains control of the catheter 602, and may pause or stop catheter navigation by the denervation treatment prediction system 100 at any time. Robotic control and navigation of a catheter 602 typically utilizes a motor to steer the catheter 602 and associated guidewire and move them forward. Motorized rollers may be used for this purpose. In other embodiments, microfluidic pressure is used to steer and / or move forward the catheter 602 and associated guidewire. In other embodiments, magnetic forces are used to move the tip of the catheter 602 in a particular direction. The catheter 602 may have a first magnet at its tip, which responds to the pull of a second magnet in order to steer. At least one of the magnets may be an electromagnet switchable between off and on states. Magnetic steering may provide the clinician 108 with a higher level of control of the distal end of the catheter 602, because it may provide a wider range of motion and allows rapid changes of direction. The second magnet may be located outside of the patient 104 and may be used to steer and move the distal tip of the catheter 602 by moving the second magnet around the patient 104 and / or by moving the patient 104 around the second magnet. Thus, depending on the role of the second magnet, the robot may either hold the second magnet or the table on which the patient 104 is placed for the denervation procedure, or both.

[0238] In other embodiments, the catheter 602 may be substantially the same as the catheter 602 utilized for human insertion and navigation as described above, and one or more end effectors of a surgical robot may be used to control and navigate that catheter 602 in the same manner as a human. Examples of such a surgical robot include the da Vinci surgical robot (Intuitive Surgical, Sunnyvale, California), the Hugo surgical robot (Medtronic, Minneapolis, Minnesota), the Monarch surgical robot (Ethicon, Raritan, New Jersey), and the SSi Mantra (SS Innovations International Inc., Gurugram, Haryana, India).

[0239] By presenting treatment information to the clinician 108, the use of catheter navigation guidance, direct control of the catheter 602, or some combination of the above, the denervation treatment prediction system 100 allows for a reduced need for contrast injection into the patient 104 by the clinician 108. Contrast agents, also called contrast media, are substances used to enhance the radiodensity of a targeted tissue by altering the way that electromagnetic radiation passes through the body of the patient 104. For denervation procedures, contrast agent isadministered to the patient 104 intravenously, because that contrast agent shows the vasculature well during fluoroscopic imaging. Such contrast agents are typically iodine or barium-based. Iodinated contrast agents are classified based on their osmolality, ranging from approximately 300 to 1200 osmol / kg H2O. Because iodine is the radiopaque substance in all iodinated contrast agents, the radiopacity produced by the administration of these contrast agents depends on their concentration of iodine. Iodinated contrast agents are most frequently administered via intravascular injection, but the substance quickly redistributes to the extravascular space due to the capillary permeability of the contrast molecules. Contrast toxicity occurs when the substances used as contrast agents - iodine, barium, or other agents - cause harmful effects to organic tissues. Those harmful effects are most noticeable in the extravascular space to which the contrast agent redistributes. The most common contrast toxicity is contrast-induced nephropathy (ON). ON is an iatrogenic acute kidney injury characterized by a sudden compromise in renal function within 24 to 48 hours of exposure to a contrast agent. Indeed, contrast toxicity is accepted to be the third most common cause of new acute kidney failure in hospitalized patients 104. Because catheter navigation guidance and control at block 1316 provides either detailed directions, or robotic control, for motion of the catheter 602 in the vasculature, less contrast agent needs to be used in the course of a denervation procedure, allowing for less contrast agent to be used. As a result, the likelihood of occurrence of the unwanted side effect of ON in the patient 104 may be reduced in comparison with standard catheter-based denervation procedures.

[0240] The denervation treatment prediction system 100 has been described using an example of renal denervation for the treatment of hypertension. That example is not limiting, and the denervation treatment prediction system 100 may be utilized to predict the effectiveness of denervation of organs other than or in addition to the kidneys, in the treatment of hypertension and / or other diseases and conditions. As one example, the denervation treatment prediction system 100 may be utilized, additionally or alternatively, to predict the effectiveness, for a specific patient 104, of hepatic artery denervation and / or other denervation to treat diabetes and / or hypertension. Stimulation of sympathetic nerve fibers in the hepatic plexus may increase blood glucose levels by increasing hepatic glucose production. Stimulation of sympathetic nerve fibers of the hepatic plexus may also increase blood glucose levels by decreasing hepatic glucose uptake. Stimulation of sympathetic nerve fibers in the hepatic plexus may also increase sympathetic nerve activity, e.g., central sympathetic nerve activity. Therefore, by disrupting sympathetic nerve signaling (e.g., hepatic denervation, which is the ablation of nerves within and surrounding the common hepatic artery) in the hepatic plexus,blood glucose, triglyceride, norepinephrine, lipid (e.g., lipoprotein), cholesterol levels and / or blood pressure may be decreased or reduced. For example, hepatic denervation may decrease blood glucose levels by 40-50% from baseline. Hepatic denervation may decrease systemic glucose by increasing insulin secretion.

[0241] In addition, or alternatively, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104 and / or a patient population, of denervation of sympathetic nerve fibers surrounding the pancreas. Denervation of the pancreas may be performed to decrease glucagon levels, increase insulin levels and / or decrease central sympathetic nerve activity to e.g., treat high blood pressure.

[0242] In addition, or alternatively, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104 and / or a patient population, of denervation of sympathetic nerve fibers surrounding the adrenal glands. In some embodiments, sympathetic nerve fibers surrounding the adrenal glands are modulated to affect adrenaline, noradrenaline levels, and / or decrease central sympathetic nerve activity to e.g., treat high blood pressure.

[0243] In addition, or alternatively, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104 and / or a patient population, of cavitation of fatty tissue of the liver. Fatty tissue (e.g., visceral fat) of the liver may be targeted to affect glycerol or free fatty acid levels, which may, for example, impact a patient’s ability to maintain normal glucose concentrations. For example, ultrasonic cavitation may be used to remove visceral fat surrounding the liver based on an outcome of the denervation treatment prediction system 100. Such removal of visceral fat could also be predicted to reduce the blood pressure of the specific patient 104.

[0244] In addition, or alternatively, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104 and / or a patient population, of denervation of the celiac plexus. In some embodiments, sympathetic nerve fibers of the celiac plexus are modulated to treat hypertension, and / or to reduce pain associated with pancreatic cancer.

[0245] In addition, or alternatively, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104 and / or a patient population, of denervation of the splanchnic nerve. In some embodiments, the splanchnic nerve is modulated to treat hypertension, and / or heart failure with preserved ejection fraction (HFpEF).

[0246] In addition, or in the alternative, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104, of pulmonary arterydenervation to treat pulmonary hypertension. As another example, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104, of denervation of one or more pulmonary nerves associated with the main bronchi (and / or other nerves associated with lung tissue) to treat chronic obstructive pulmonary disease (COPD). As another example, the denervation treatment prediction system 100 may be utilized to predict the effectiveness, for a specific patient 104, of denervation of one or more nerves associated with musculoskeletal tissue for pain relief.

[0247] In addition, the denervation treatment prediction system 100 may be configured to communicate with a device of the patient 104 after the denervation procedure. The device of the patient 104 may be a smartphone, a computer, a tablet, or other suitable device. An app may be provided that the patient 104 may download to their smartphone. Communication between the denervation treatment prediction system 100 and a device of the patient 104 may be useful in obtaining short-term and long-term results of the denervation procedure on the patient 104. These results may benefit the patient 104, and additionally or instead may be used to train the denervation treatment prediction system 100 to better predict the results of a denervation procedure or procedures for other patients 104.

[0248] The denervation treatment prediction system 100 is described herein as simulating the results of potential denervation procedures, and making suggestions for treatment prior to, during, and after a denervation procedure, and examples have been provided with regard to the relationship between patient specific data 102 and a corpus of patient population data 111. The corpus of patient population data 111 will grow with time, as patient specific data 102 is added to it. The growth of the corpus of patient population data 111 allows the denervation treatment prediction system 100 to be trained as time goes by, which is expected to result in better simulations, predictions, and suggestions.

[0249] This specification uses the term “configured” in connection with systems and computer program components. For a system of one or more computers to be configured to perform particular operations or actions means that the system has installed on it software, firmware, hardware, or a combination of them that in operation cause the system to perform the operations or actions. For one or more computer programs to be configured to perform particular operations or actions means that the one or more programs include instructions that, when executed by data processing apparatus, cause the apparatus to perform the operations or actions.

[0250] Embodiments of the subject matter and the functional operations described in this specification may be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed inthis specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification may be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non-transitory storage medium for execution by, or to control the operation of, data processing apparatus. The computer storage medium may be a machine- readable storage device, a machine -readable storage substrate, a random or serial access memory device, or a combination of one or more of them. Alternatively, or in addition, the program instructions may be encoded on an artificially-generated propagated signal, such as but not limited to a machine-generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus.

[0251] The term “data processing apparatus” refers to data processing hardware and encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers. The apparatus may also be, or further include, special purpose logic circuitry, such as but not limited to an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit). The apparatus may optionally include, in addition to hardware, code that creates an execution environment for computer programs, such as but not limited to code that constitutes processor firmware, a protocol stack, a database management system, and / or an operating system.

[0252] A computer program, which may also be referred to or described as a program, software, a software application, an app, a module, a software module, a script, or code, may be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages; and it may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A program may, but need not, correspond to a file in a file system. A program may be stored in a portion of a file that holds other programs or data, such as but not limited to one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, such as but not limited to files that store one or more modules, sub-programs, or portions of code. A computer program may be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a data communication network.

[0253] In this specification the term “engine” is used broadly to refer to a software-based system, subsystem, or process that is programmed to perform one or more specific functions.Generally, an engine will be implemented as one or more software modules or components, installed on one or more computers in one or more locations. In some cases, one or more computers will be dedicated to a particular engine; in other cases, multiple engines may be installed and running on the same computer or computers.

[0254] The processes and logic flows described in this specification may be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows may also be performed by special purpose logic circuitry, e.g., an FPGA or an ASIC, or by a combination of special purpose logic circuitry and one or more programmed computers.

[0255] Computers suitable for the execution of a computer program may be based on general or special purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. The central processing unit and the memory may be supplemented by, or incorporated in, special purpose logic circuitry. Generally, a computer will also include or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, optical disks, and / or solid state memory. However, a computer need not have such devices. Moreover, a computer may be embedded in another device, such as but not limited to a smartphone, a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, and / or a portable storage device such as a universal serial bus (USB) flash drive.

[0256] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices); magnetic disks e.g., internal hard disks or removable disks); magneto-optical disks; and CD-ROM and DVD-ROM disks.

[0257] To provide for interaction with a user, embodiments of the subject matter described in this specification may be implemented on a computer or other device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor), for displaying information to the user and a keyboard and a pointing device (e.g. , a mouse, a trackball, and / or a touchscreen), by which the user may provide input to the computer. Other kinds of devices may be used to provide for interaction with a user as well; for example, feedback provided tothe user may be any form of sensory feedback, such as but not limited to visual feedback, auditory feedback, or tactile feedback; and input from the user may be received in any form, such as but not limited to acoustic, speech, or tactile input. In addition, a computer may interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user’s device in response to requests received from the web browser. Also, a computer may interact with a user by sending text messages or other forms of message to a personal device, e.g., a smartphone that is running a messaging application, and receiving responsive messages from the user in return.

[0258] Data processing apparatus for implementing machine learning models may also include, for example, special-purpose hardware accelerator units for processing common and computationally-intensive parts of machine learning training or production, i.e., inference and workloads.

[0259] Machine learning models may be implemented and deployed using a machine learning framework, e.g., a TensorFlow framework, or a Jax framework.

[0260] Embodiments of the subject matter described in this specification may be implemented in a computing system that includes a back-end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front-end component, such as but not limited to a client computer having a graphical user interface, a web browser, or an app through which a user may interact with an implementation of the subject matter described in this specification, or any combination of one or more such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication, such as but not limited to a communication network that may be a local area network (LAN) and / or a wide area network (WAN) such as the Internet.

[0261] The computing system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. In some embodiments, a server transmits data, e.g., an HTML page, to a user device, e.g., for purposes of displaying data to and receiving user input from a user interacting with the device, which acts as a client. Data generated at the user device, such as data resulting from the user interaction, may be received at the server from the device.

[0262] In addition to the embodiments described above, the following embodiments are also innovative:

[0263] Embodiment 1 is a method performed by one or more computers, comprising: receiving patient specific data; producing a set of predictive features that characterize the patient specific data; performing one or more simulations of a body of the patient based on the patient specific data to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient; processing the set of predictive features and the one or more simulation results using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

[0264] Embodiment 2 is the method of embodiment 1, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a distribution of the predicted effects of the denervation procedure.

[0265] Embodiment 3 is the method of embodiment 1 or embodiment 2, wherein the predicted effects of the one or more denervation procedures include predicted physiological changes in the body of the patient resulting from the one or more denervation procedures being performed on the body of the patient.

[0266] Embodiment 4 is the method of any one of embodiments 1-3, wherein the data characterizing the predicted effects characterizes, for each denervation procedure, a predicted effectiveness of the denervation procedure.

[0267] Embodiment 5 is the method of any one of embodiments 1-4, when dependent on embodiment 3, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a predicted effectiveness of the denervation procedure based on the predicted physiological changes in the body of the patient.

[0268] Embodiment 6 is the method of any one of embodiments 1-5, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a predicted effectiveness score of the denervation procedure based on the predicted physiological changes in the body of the patient.

[0269] Embodiment 7 is the method of embodiment 5 or embodiment 6, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a categorical effectiveness classification of the denervation procedure based on the predicted physiological changes in the body of the patient.

[0270] Embodiment 8 is the method of any one of embodiments 1-7, wherein the patient specific data includes anatomical data characterizing an internal structure of the body of the patient.

[0271] Embodiment 9 is the method of any one of embodiments 1-8, wherein the patient specific data includes physiological data characterizing processes within the body of the patient.

[0272] Embodiment 10 is the method of any one of embodiments 1-9, wherein producing a set of predictive features that characterize the patient specific data comprises: determining a set of patient specific data features characterizing the received patient specific data; and including the set of patient specific data features within the set of predictive features.

[0273] Embodiment 11 is the method of any one of embodiments 1-10, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: determining parameters for each of the one or more simulations based on the patient specific data.

[0274] Embodiment 12 is the method of any one of embodiments 1-11, when dependent on embodiment 8, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing at least one simulation of a process within a simulated organ system, wherein the structure of the simulated organ system is determined based on the anatomical data for the patient.

[0275] Embodiment 13 is the method of any one of embodiments 1-12, when dependent on embodiment 9, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing at least one simulation of a process within a simulated body, wherein the simulated process is determined based on the physiological data for the patient.

[0276] Embodiment 14 is the method of any one of embodiments 1-13, wherein the simulation results characterize simulated physiological changes in the body of the patient.

[0277] Embodiment 15 is the method of any one of embodiments 1-14, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing simulations of one or more denervation procedures.

[0278] Embodiment 16 is the method of embodiment 15, wherein performing simulations of one or more denervation procedures comprises: performing simulations of multiple denervation procedures in different locations within the body of the patient.

[0279] Embodiment 17 is the method of embodiment 15 or embodiment 16, wherein performing simulations of one or more denervation procedures comprises performing simulations of denervation procedures to multiple organs of the patient.

[0280] Embodiment 18 is the method of any one of embodiments 15-17, wherein the simulations of one or more denervation procedures include a simulation of a denervation procedure to a kidney of the patient.

[0281] Embodiment 19 is the method of any one of embodiments 15-18, wherein the simulations of one or more denervation procedures include a simulation of a denervation procedure to a pancreas of the patient.

[0282] Embodiment 20 is the method of any one of embodiments 15-19, wherein the simulations of one or more denervation procedures include a simulation of a denervation procedure to a liver of the patient.

[0283] Embodiment 21 is the method of any one of embodiments 15-20, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to a duodenum of the patient.

[0284] Embodiment 22 is the method of any one of embodiments 15-21, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to a stomach of the patient.

[0285] Embodiment 23 is the method of any one of embodiments 15-22, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to an intestine of the patient.

[0286] Embodiment 24 is the method of any one of embodiments 15-23, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to a spleen of the patient.

[0287] Embodiment 25 is the method of any one of embodiments 15-24, wherein the simulation results characterize predicted physiological changes in the body of the patient as a result of the one or more simulated denervation procedures.

[0288] Embodiment 26 is the method of any one of embodiments 1-25, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing a simulation of ablating tissue of the patient.

[0289] Embodiment 27 is the method of embodiment 26, wherein performing the simulation of ablating tissue of the patient comprises simulating a transfer of energy from an energy source to the tissue.

[0290] Embodiment 28 is the method of embodiment 27, wherein simulating the transfer of energy from the energy source to the tissue comprises using a finite element method to simulate the transfer of energy from the energy source to the tissue.

[0291] Embodiment 29 is the method of any one of embodiments 1-28, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing a simulation of blood flow characteristics in a patient artery.

[0292] Embodiment 30 is the method of embodiment 29, wherein performing the simulation of blood flow characteristics in the patient artery comprises: simulating the flow of blood in the patient artery; and simulating the response of artery walls of the patient artery to the simulated flow of blood.

[0293] Embodiment 31 is the method of embodiment 29 or 30, wherein performing the simulation of blood flow characteristics in the patient artery comprises using computational fluid dynamics to simulate the flow of blood in the patient artery.

[0294] Embodiment 32 is the method of any one of embodiments 1-31, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing a simulation of a sympathetic nervous system response of the patient.

[0295] Embodiment 33 is the method of any one of embodiments 1-32, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing a simulation of a parasympathetic nervous system response of the patient.

[0296] Embodiment 34 is the method of any one of embodiments 1-33, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises performing a simulation of an autonomic nervous system response of the patient.

[0297] Embodiment 35 is the method of any one of embodiments 1-34, wherein including data characterizing the one or more simulation results within the set of predictive features comprises: determining a set of simulation result features characterizing the one or more simulation results; and including the set of simulation result features within the set of predictive features.

[0298] Embodiment 36 is the method of any one of embodiments 1-35, wherein the predictive machine learning model is a denervation procedure selection model trained using reinforcement learning techniques.

[0299] Embodiment 37 is the method of any one of embodiments 1-36, wherein the predictive machine learning model has been trained to reproduce target predictions for a plurality of example patients based on a set of training data comprising, for each of the plurality of example patients: (i) example patient specific data for the example patient; (ii) one or more example denervation procedures for the example patient; and (iii) one or more target predictions for the example patient.

[0300] Embodiment 38 is the method of embodiment 37, further comprising: receiving one or more target predictions for the patient; and adding patient specific data and target predictions for the patient to the set of training data.

[0301] Embodiment 39 is the method of embodiment 38, further comprising updating the predictive machine learning model based on the patient specific data and target predictions for the patient.

[0302] Embodiment 40 is the method of any one of embodiments 1-39, wherein the predictive machine learning model is a decision tree based model.

[0303] Embodiment 41 is the method of any one of embodiments 1-40, wherein the predictive machine learning model is a neural network.

[0304] Embodiment 42 is the method of any one of embodiments 1-41, further comprising determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures.

[0305] Embodiment 43 is the method of embodiment 42, wherein determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures comprises determining target organs for the one or more suggested denervation procedures.

[0306] Embodiment 44 is the method of embodiment 42 or embodiment 43, wherein determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures comprises determining suggested locations for the one or more suggested denervation procedures.

[0307] Embodiment 45 is the method of any one of embodiments 38-40, when dependent on embodiment 4, wherein determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures comprises determining one or more suggested denervation procedures based on optimizing the predicted effectiveness of the one or more denervation procedures.

[0308] Embodiment 46 is the method of any one of embodiments 42-45, wherein the data characterizing the one or more suggested denervation procedures includes data characterizing the suggested denervation procedures.

[0309] Embodiment 47 is the method of any one of embodiments 1-46, further comprising determining an indication of whether denervation treatment is recommended based on the predicted effects of the one or more denervation procedures.

[0310] Embodiment 48 is the method of embodiment 47, wherein the data characterizing the predicted effects of the one or more denervation procedures includes the indication whether denervation treatment is recommended.

[0311] Embodiment 49 is the method of any one of embodiments 1-48, wherein the received patient specific data includes patient specific data obtained before a denervation treatment for the patient.

[0312] Embodiment 50 is the method of any one of embodiments 1-49, wherein the received patient specific data includes patient specific data obtained during a denervation treatment for the patient.

[0313] Embodiment 51 is the method of any one of embodiments 1-50, wherein the received patient specific data includes follow-up patient specific data obtained after a denervation treatment for the patient.

[0314] Embodiment 52 is the method of any one of embodiments 1-51, when dependent on embodiment 9, wherein the anatomical data for the patient includes one or more medical images for the patient.

[0315] Embodiment 53 is the method of embodiment 52, wherein producing a set of predictive features that characterize the patient specific data comprises: extracting patient-specific geometric and pathologic features from the medical image data; and including the patient specific geometric and pathologic features within the set of predictive features.

[0316] Embodiment 54 is the method of embodiment 52 or embodiment 53, further comprising: producing one or more overlays for the medical images for the patient based on the data characterizing the predicted effects of the one or more denervation procedures; and providing the one or more overlays for display on a user device.

[0317] Embodiment 55 is the method of embodiment 54, when dependent on embodiment 30, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes locations of the one or more suggested denervation procedures.

[0318] Embodiment 56 is the method of embodiment 54, when dependent on embodiment 30, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes the predicted effects of the one or more suggested denervation procedures.

[0319] Embodiment 57 is the method of embodiment 56, when dependent on embodiment 4, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes the predicted effectiveness of the one or more suggested denervation procedures.

[0320] Embodiment 58 is the method of embodiment 54, when dependent on embodiment 13, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes one or more of the simulated denervation procedures.

[0321] Embodiment 59 is the method of any one of embodiments 1-58, when dependent on embodiment 3, wherein the predicted physiological changes in body of the patient resulting from the one or more denervation procedures include a predicted change in the patient’s blood pressure.

[0322] Embodiment 60 is the method of any one of embodiments 1-59, further comprising determining recommendations for the patient based on the patient specific data and the predicted effects of the one or more denervation procedures.

[0323] Embodiment 61 is the method of embodiment 60, when dependent on embodiment 4, wherein determining recommendations for the patient based on the patient specific data and the predicted effects of the one or more denervation procedures comprises determining recommendations for the patient based on optimizing the predicted effectiveness of the one or more denervation procedures.

[0324] Embodiment 62 is the method of embodiment 60 or embodiment 61, wherein the recommendations for the patient include one or more recommended lifestyle changes for the patient.

[0325] Embodiment 63 is the method of any one of embodiments 60-62, wherein the recommendations for the patient include one or more medication recommendations for the patient.

[0326] Embodiment 64 is the method of embodiment 63, wherein the medication recommendations include recommended dosages.

[0327] Embodiment 65 is the method of any one of embodiments 60-64, further comprising providing data characterizing the recommendations for the patient for display on a user device.

[0328] Embodiment 66 is a method performed by one or more computers, comprising: receiving patient specific data, wherein the patient specific data comprises continuously- monitored patient physiological data measured over a continuous time period; producing a setof predictive features that characterize the patient specific data, wherein the set of predictive features comprises one or more aggregation measures computed from the continuously- monitored patient physiological data measured over the continuous time period; processing at least the set of predictive features using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on a body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

[0329] Embodiment 67 is the method of embodiment 66, further comprising: performing one or more simulations of the body of the patient based on the patient specific data to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient, wherein processing at least the set of predictive features using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient comprises: processing at least the set of predictive features and the one or more simulation results using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient.

[0330] Embodiment 68 is the method of embodiments 66 or 67, wherein the continuously- monitored patient physiological data comprises one or more of: ambulatory systolic blood pressure, pulse pressure, heart rate, or orthostatic blood pressure.

[0331] Embodiment 69 is the method of embodiment 68, wherein the one or more aggregation measures comprise one or more of: variability in a slope of systolic blood pressure, variability in a rate of change of a slope of pulse pressure, a maximum rate of change of the slope of systolic blood pressure, a minimum rate of change of the slope of systolic blood pressure, or a mean rate of change of a slope of heart rate.

[0332] Embodiment 70 is the method of any one of embodiments 66-69, further comprising the operations of the respective method of any one of embodiments 2-65.

[0333] Embodiment 71 is the method of any preceding embodiment, wherein the patient specific data comprises data indicating respective levels of one or more hormones of the patient.

[0334] Embodiment 72 is a method performed by one or more computers, comprising: acquiring one or more two-dimensional (2D) medical images of an interior of a patient; for each of the 2D medical images, processing an input generated from the 2D medical image using a neural network to generate a respective segmentation of the 2D medical image that identifies pixels of the 2D medical image that depict one or more blood vessels; determining features ofthe one or more blood vessels from the respective segmentations; and generating treatment information using the features of the one or more blood vessels; and providing the treatment information for display to a clinician.

[0335] Embodiment 73 is the method of embodiment 72, further comprising, for each 2D medical image: pre-processing the 2D medical image generate a pre-processed 2D medical image.

[0336] Embodiment 74 is the method of embodiment 73, further comprising, for each 2D medical image: applying adaptive Frangi filtering to the pre-processed 2D medical images to generate a filtered 2D medical image, wherein the input comprises the filtered 2D medical image and the pre-processed 2D medical image or the 2D medical image.

[0337] Embodiment 75 is the method of embodiment 72, wherein at least one 2D medical image is an X-ray image.

[0338] Embodiment 76 is the method of embodiment 72, wherein at least one 2D medical image is an MRI image.

[0339] Embodiment 77 is the method of embodiment 72, wherein at least one 2D medical image is a CT image.

[0340] Embodiment 78 is the method of embodiment 72, wherein at least one 2D medical image is an ultrasound image.

[0341] Embodiment 79 is the method of embodiment 72, wherein the pre-processing comprises resizing the 2D medical images.

[0342] Embodiment 80 is the method of embodiment 73, wherein the pre-processing comprises grayscaling at least one 2D medical image.

[0343] Embodiment 81 is the method of embodiment 73, wherein the pre-processing comprises performing noise reduction on at least one 2D medical image.

[0344] Embodiment 82 is the method of embodiment 73, wherein the pre-processing comprises performing normalization on at least one 2D medical image.

[0345] Embodiment 83 is the method of embodiment 73, wherein the pre-processing comprises performing binarization on at least one 2D medical image.

[0346] Embodiment 84 is the method of embodiment 73, wherein the pre-processing comprises performing contrast enhancement on at least one 2D medical image using histogram equalization.

[0347] Embodiment 85 is the method of embodiment 73, wherein the identifying further comprises identifying specific features of at least one blood vessel in at least one segmented 2D medical image.

[0348] Embodiment 86 is the method of embodiment 73, wherein the treatment information comprises one or more of: data identifying one or more suggested ablation locations; data identifying a dose to be given at each suggested ablation location; or data identifying labeled treatment-relevant locations within the patient.

[0349] Embodiment 87 is a method performed by one or more computers, comprising: acquiring a respective 2D medical image of an interior of a patient at each of a plurality of angles; generate a respective segmentation of the 2D medical image that identifies pixels of the 2D medical image that depict one or more blood vessels; generating, from the respective segmentations of the 2D medical images, a 3D model of a vasculature of the patient at a volume of interest; generating information about recommended treatment based on the 3D model, and displaying the information about recommended treatment.

[0350] Embodiment 88 is the method of embodiment 87, wherein the generating a 3D model of the vasculature of the patient at the volume of interest is performed prior to a denervation procedure.

[0351] Embodiment 89 is the method of embodiment 87, wherein the generating a 3D model of the vasculature of the patient at the volume of interest is performed during a denervation procedure.

[0352] Embodiment 90 is the method of embodiment 87, further comprising generating guidance for navigating a catheter to a treatment site in the body of the patient using the 3D model of the vasculature.

[0353] Embodiment 91 is the method of embodiment 39, wherein the predictive machine learning model has been trained to generate a ranking of the predictive features, and wherein the updating comprising updating the ranking of the predictive features.

[0354] Embodiment 92 is the method of embodiment 91, wherein the ranking of the predictive features assigns each predictive feature into a hierarchy of features that comprises primary features, secondary features, and tertiary features.

[0355] Embodiment 93 is the method of embodiment 92, wherein updating the ranking comprises: evaluating, using at least the patient specific data and the target predictions for the patient, a performance of the predictive machine learning model using the primary features, the secondary features, and the tertiary features; and updating the ranking based on the evaluating.

[0356] Embodiment 94 is a system comprising: one or more computers; and one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers,cause the one or more computers to perform operations of the respective method of any one of embodiments 1-93.

[0357] Embodiment 95 is one or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations of the respective method of any one of embodiments 1-93.

[0358] While this specification contains many examples of specific implementation details, these should not be construed as limitations on the scope of the present disclosure or on the scope of what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of the present disclosure. Certain features that are described in this specification in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment may also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially be claimed as such, one or more features from a claimed combination may in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

[0359] Similarly, while operations are depicted in the drawings and recited in the claims in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems may generally be integrated together in a single software product or packaged into multiple software products.

[0360] Particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims may be performed in a different order and still achieve desirable results. As one example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some cases, multitasking and parallel processing may be advantageous.

Claims

CLAIMS1. A method performed by one or more computers, comprising: receiving patient specific data; producing a set of predictive features that characterize the patient specific data; performing one or more simulations of a body of the patient based on the patient specific data to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient; processing the set of predictive features and the one or more simulation results using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

2. The method of claim 1, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a distribution of the predicted effects of the denervation procedure.

3. The method of claim 1 or claim 2, wherein the predicted effects of the one or more denervation procedures include predicted physiological changes in the body of the patient resulting from the one or more denervation procedures being performed on the body of the patient.

4. The method of any preceding claim, wherein the data characterizing the predicted effects characterizes, for each denervation procedure, a predicted effectiveness of the denervation procedure.

5. The method of any preceding claim, when dependent on claim 3, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a predicted effectiveness of the denervation procedure based on the predicted physiological changes in the body of the patient.

6. The method of claim 5, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a predicted effectiveness score of the denervation procedure based on the predicted physiological changes in the body of the patient.

7. The method of claim 5 or claim 6, wherein the data characterizing the predicted effects of the one or more denervation procedures characterizes, for each denervation procedure, a categorical effectiveness classification of the denervation procedure based on the predicted physiological changes in the body of the patient.

8. The method of claim any preceding claim, wherein the patient specific data includes anatomical data characterizing an internal structure of the body of the patient.

9. The method of any preceding claim, wherein the patient specific data includes physiological data characterizing processes within the body of the patient.

10. The method of any preceding claim, wherein producing a set of predictive features that characterize the patient specific data comprises: determining a set of patient specific data features characterizing the received patient specific data; and including the set of patient specific data features within the set of predictive features.

11. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: determining parameters for each of the one or more simulations based on the patient specific data.

12. The method of any preceding claim, when dependent on claim 8, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing at least one simulation of a process within a simulated organ system, wherein the structure of the simulated organ system is determined based on the anatomical data for the patient.

13. The method of any dependent claim, when dependent on claim 9, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing at least one simulation of a process within a simulated body, wherein the simulated process is determined based on the physiological data for the patient.

14. The method of any preceding claim, wherein the simulation results characterize simulated physiological changes in the body of the patient.

15. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing simulations of one or more denervation procedures.

16. The method of claim 15, wherein performing simulations of one or more denervation procedures comprises: performing simulations of multiple denervation procedures in different locations within the body of the patient.

17. The method of claim 15 or claim 16, wherein performing simulations of one or more denervation procedures comprises: performing simulations of denervation procedures to multiple organs of the patient.

18. The method of claims 15-17, wherein the simulations of one or more denervation procedures include a simulation of a denervation procedure to a kidney of the patient.

19. The method of claims 15-18, wherein the simulations of one or more denervation procedures include a simulation of a denervation procedure to a pancreas of the patient.

20. The method of claims 15-19, wherein the simulations of one or more denervation procedures include a simulation of a denervation procedure to a liver of the patient.

21. The method of claims 15-20, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to a duodenum of the patient.

22. The method of claims 15-21, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to a stomach of the patient.

23. The method of claims 15-22, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to an intestine of the patient.

24. The method of claims 15-23, wherein the simulations of one or more denervation procedures includes a simulation of a denervation procedure to a spleen of the patient.

25. The method of claims 15-24, wherein the simulation results characterize predicted physiological changes in the body of the patient as a result of the one or more simulated denervation procedures.

26. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing a simulation of ablating tissue of the patient.

27. The method of claim 26, wherein performing the simulation of ablating tissue of the patient comprises: simulating a transfer of energy from an energy source to the tissue.

28. The method of claim 27, wherein simulating the transfer of energy from the energy source to the tissue comprises: using a finite element method to simulate the transfer of energy from the energy source to the tissue.

29. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing a simulation of blood flow characteristics in a patient artery.

30. The method of claim 29, wherein performing the simulation of blood flow characteristics in the patient artery comprises: simulating the flow of blood in the patient artery; and simulating the response of artery walls of the patient artery to the simulated flow of blood.

31. The method of claim 29 or claim 30, wherein performing the simulation of blood flow characteristics in the patient artery comprises: using computational fluid dynamics to simulate the flow of blood in the patient artery.

32. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing a simulation of a sympathetic nervous system response of the patient.

33. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing a simulation of a parasympathetic nervous system response of the patient.

34. The method of any preceding claim, wherein performing the one or more simulations of the body of the patient based on the patient specific data to produce the one or more simulation results comprises: performing a simulation of an autonomic nervous system response of the patient.

35. The method of any preceding claim, wherein including data characterizing the one or more simulation results within the set of predictive features comprises: determining a set of simulation result features characterizing the one or more simulation results; and including the set of simulation result features within the set of predictive features.

36. The method of any preceding claim, wherein the predictive machine learning model is a denervation procedure selection model trained using reinforcement learning techniques.

37. The method of any preceding claim, wherein the predictive machine learning model has been trained to reproduce target predictions for a plurality of example patients based on a set of training data comprising, for each of the plurality of example patients:(i) example patient specific data for the example patient;(ii) one or more example denervation procedures for the example patient; and(iii) one or more target predictions for the example patient.

38. The method of claim 37, further comprising: receiving one or more target predictions for the patient; and adding patient specific data and target predictions for the patient to the set of training data.

39. The method of claim 38, further comprising: updating the predictive machine learning model based on the patient specific data and target predictions for the patient.

40. The method of any preceding claim, wherein the predictive machine learning model is a decision tree based model.

41. The method of any preceding claim, wherein the predictive machine learning model is a neural network.

42. The method of any preceding claim, further comprising: determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures.

43. The method of claim 42, wherein determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures comprises: determining target organs for the one or more suggested denervation procedures.

44. The method of claim 42 or 43, wherein determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures comprises: determining suggested locations for the one or more suggested denervation procedures.

45. The method of claims 38-40 when dependent on claim 4, wherein determining one or more suggested denervation procedures based on the predicted effects of the one or more denervation procedures comprises: determining one or more suggested denervation procedures based on optimizing the predicted effectiveness of the one or more denervation procedures.

46. The method of claims 42-45, wherein the data characterizing the one or more suggested denervation procedures includes data characterizing the suggested denervation procedures.

47. The method of any preceding claim, further comprising: determining an indication of whether denervation treatment is recommended based on the predicted effects of the one or more denervation procedures.

48. The method of claim 47, wherein the data characterizing the predicted effects of the one or more denervation procedures includes the indication whether denervation treatment is recommended.

49. The method of any preceding claim, wherein the received patient specific data includes patient specific data obtained before a denervation treatment for the patient.

50. The method of any preceding claim, wherein the received patient specific data includes patient specific data obtained during a denervation treatment for the patient.

51. The method of any preceding claim, wherein the received patient specific data includes follow-up patient specific data obtained after a denervation treatment for the patient.

52. The method of any preceding claim, when dependent on claim 9, wherein the anatomical data for the patient includes one or more medical images for the patient.

53. The method of claim 52, wherein producing a set of predictive features that characterize the patient specific data comprises: extracting patient specific geometric and pathologic features from the medical image data; and including the patient specific geometric and pathologic features within the set of predictive features.

54. The method of claim 52 or claim 53, further comprising: producing one or more overlays for the medical images for the patient based on the data characterizing the predicted effects of the one or more denervation procedures; and providing the one or more overlays for display on a user device.

55. The method of claim 54, when dependent on claim 30, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes locations of the one or more suggested denervation procedures.

56. The method of claim 54, when dependent on claim 30, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes the predicted effects of the one or more suggested denervation procedures.

57. The method of claim 56, when dependent on claim 4, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes the predicted effectiveness of the one or more suggested denervation procedures.

58. The method of claim 54, when dependent on claim 13, wherein at least one of the overlays, when combined with the medical images for the patient, characterizes one or more of the simulated denervation procedures.

59. The method of any preceding claim, when dependent on claim 3, wherein the predicted physiological changes in body of the patient resulting from the one or more denervation procedures include a predicted change in the patient’ s blood pressure.

60. The method of any preceding claim, further comprising: determining recommendations for the patient based on the patient specific data and the predicted effects of the one or more denervation procedures.

61. The method of claim 60 when dependent on claim 4, wherein determining recommendations for the patient based on the patient specific data and the predicted effects of the one or more denervation procedures comprises: determining recommendations for the patient based on optimizing the predicted effectiveness of the one or more denervation procedures.

62. The method of claim 60 or claim 61, wherein the recommendations for the patient include one or more recommended lifestyle changes for the patient.

63. The method of claims 60-62, wherein the recommendations for the patient include one or more medication recommendations for the patient.

64. The method of claim 63, wherein the medication recommendations include recommended dosages.

65. The method of claims 60-64, further comprising: providing data characterizing the recommendations for the patient for display on a user device.

66. A method performed by one or more computers, comprising: receiving patient specific data, wherein the patient specific data comprises continuously-monitored patient physiological data measured over a continuous time period; producing a set of predictive features that characterize the patient specific data, wherein the set of predictive features comprises one or more aggregation measures computed from the continuously-monitored patient physiological data measured over the continuous time period; processing at least the set of predictive features using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on a body of the patient; and providing data characterizing the predicted effects of the one or more denervation procedures for display on a user device.

67. The method of claim 66, further comprising: performing one or more simulations of the body of the patient based on the patient specific data to produce one or more simulation results that are predictive of effects of denervation procedures being performed to the body of the patient, wherein processing at least the set of predictive features using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient comprises: processing at least the set of predictive features and the one or more simulation results using a predictive machine learning model to generate a predicted output characterizing predicted effects of one or more denervation procedures on the body of the patient.

68. The method of claims 66 or 67, wherein the continuously-monitored patient physiological data comprises one or more of: ambulatory systolic blood pressure, pulse pressure, heart rate, or orthostatic blood pressure.

69. The method of claim 68, wherein the one or more aggregation measures comprise one or more of: variability in a slope of systolic blood pressure, variability in a rate of change of a slope of pulse pressure, a maximum rate of change of the slope of systolic blood pressure, a minimum rate of change of the slope of systolic blood pressure, or a mean rate of change of a slope of heart rate.

70. The method of any one of claims 66-69, further comprising the operations of the respective method of any one of claims 2-65.

71. The method of any preceding claim, wherein the patient specific data comprises data indicating respective levels of one or more hormones of the patient.

72. A method performed by one or more computers, comprising: acquiring one or more two-dimensional (2D) medical images of an interior of a patient; for each of the 2D medical images, processing an input generated from the 2D medical image using a neural network to generate a respective segmentation of the 2D medical image that identifies pixels of the 2D medical image that depict one or more blood vessels; determining features of the one or more blood vessels from the respective segmentations; and generating treatment information using the features of the one or more blood vessels; and providing the treatment information for display to a clinician.

73. The method of claim 72, further comprising, for each 2D medical image: pre-processing the 2D medical image generate a pre-processed 2D medical image.

74. The method of claim 73, further comprising, for each 2D medical image: applying adaptive Frangi filtering to the pre-processed 2D medical images to generate a filtered 2D medical image, wherein the input comprises the filtered 2D medical image and the pre-processed 2D medical image or the 2D medical image.

75. The method of claim 72, wherein at least one 2D medical image is an X-ray image.

76. The method of claim 72, wherein at least one 2D medical image is an MRI image.

77. The method of claim 72, wherein at least one 2D medical image is a CT image.

78. The method of claim 72, wherein at least one 2D medical image is an ultrasound image.

79. The method of claim 72, wherein the pre-processing comprises resizing the 2D medical images to80. The method of claim 73, wherein the pre-processing comprises grayscaling at least one 2D medical image.

81. The method of claim 73, wherein the pre-processing comprises performing noise reduction on at least one 2D medical image.

82. The method of claim 73, wherein the pre-processing comprises performing normalization on at least one 2D medical image.

83. The method of claim 73, wherein the pre-processing comprises performing binarization on at least one 2D medical image.

84. The method of claim 73, wherein the pre-processing comprises performing contrast enhancement on at least one 2D medical image using histogram equalization.

85. The method of claim 73, wherein the identifying further comprises identifying specific features of at least one blood vessel in at least one segmented 2D medical image.

86. The method of claim 73, wherein the treatment information comprises one or more of: data identifying one or more suggested ablation locations; data identifying a dose to be given at each suggested ablation location; or data identifying labeled treatment-relevant locations within the patient.

87. A method performed by one or more computers, comprising: acquiring a respective 2D medical image of an interior of a patient at each of a plurality of angles; generating a respective segmentation of the 2D medical image that identifies pixels of the 2D medical image that depict one or more blood vessels; generating, from the respective segmentations of the 2D medical images, a 3D model of a vasculature of the patient at a volume of interest; generating information about recommended treatment based on the 3D model, and displaying the information about recommended treatment.

88. The method of claim 87, wherein the generating a 3D model of the vasculature of the patient at the volume of interest is performed prior to a denervation procedure.

89. The method of claim 87, wherein the generating a 3D model of the vasculature of the patient at the volume of interest is performed during a denervation procedure.

90. The method of claim 87, further comprising generating guidance for navigating a catheter to a treatment site in the body of the patient using the 3D model of the vasculature.

91. The method of claim 39, wherein the predictive machine learning model has been trained to generate a ranking of the predictive features, and wherein the updating comprising updating the ranking of the predictive features.

92. The method of claim 91, wherein the ranking of the predictive features assigns each predictive feature into a hierarchy of features that comprises primary features, secondary features, and tertiary features.

93. The method of claim 92, wherein updating the ranking comprises: evaluating, using at least the patient specific data and the target predictions for the patient, a performance of the predictive machine learning model using the primary features, the secondary features, and the tertiary features; and updating the ranking based on the evaluating.

94. A system comprising: one or more computers; and one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations of the respective method of any one of claims 1-93.

95. One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations of the respective method of any one of claims 1-93.

Images