Predicting adverse effects of cardiovascular treatments

Abstract

Certain aspects of the disclosure provide a method for predicting adverse effects due to cardiac valve treatments. The method includes obtaining at least one shape representation of a patient heart anatomy; determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating a set of labeled patient data; providing the set of labeled patient data as inputs to a machine learning model; receiving, from the machine learning model, one or more risk scores representing predicted risk probabilities associated with performing one or more cardiac valve treatments on the patient heart anatomy; and generating, for display, an output comprising a risk assessment associated with performing the one or more cardiac valve treatments based on the one or more risk scores.

Description

PREDICTING ADVERSE EFFECTS OF CARDIOVASCULAR TREATMENTSCROSS REFERENCE TO RELATED APPLICATIONS

[0001] This Application claims the benefit of and priority to U.S. Provisional Patent Application No. 63 / 647,421, filed on May 14, 2024, the entire contents of which are hereby incorporated by reference. This Application claims the benefit of and priority to U.S. Provisional Patent Application No. 63 / 647,498, filed on May 14, 2024, the entire contents of which are hereby incorporated by reference. This Application claims the benefit of and priority to U.S. Provisional Patent Application No. 63 / 647,521, filed on May 14, 2024, the entire contents of which are hereby incorporated by reference.BACKGROUNDField

[0002] Aspects of the present disclosure relate to medical treatment planning.Description of Related Art

[0003] Cardiac valve treatments, such as valve repair and valve replacement, are critical procedures to manage valve dysfunctions such as stenosis or regurgitation. While these interventions can significantly improve a patient’s quality of life and increase survival years, these treatments come with risks of adverse effects. Understanding these risks is crucial for informed decision-making and effective patient management.

[0004] Common adverse effects include both intraoperative and postoperative adverse effects, such as: intraoperative and postoperative bleeding, anticoagulation- related bleeding (e.g., patients with mechanical valve replacements often need lifelong anticoagulation therapy, which increases the risk of bleeding complications), endocarditis, surgical-site infections, stroke, valve thrombosis, degeneration of bioprosthetic valves, paravalvular leakage, atrial fibrillation, heart block, and / or the like.

[0005] Some adverse effects are specific to the type of procedure. For example, valve repair can lead to residual regurgitation, stenosis, and / or annuloplasty ring complications. Valve replacement, on the other hand, can lead to prosthesis-patient mismatch (e.g., an improperly sized valve can lead to suboptimal hemodynamics, impacting cardiac function) or hemolysis (e.g., particularly with mechanical valves, turbulent blood flow can cause the destruction of red blood cells, leading to anemia and other complications).

[0006] Finally, each treatment comes with long-term risks and considerations. For example, while they avoid the need for lifelong anticoagulation, bioprosthetic valves typically have a lifespan of 10-20 years, necessitating future replacement, especially in younger patients. Mechanical valve recipients, on the other hand, must adapt to anticoagulation therapy, including dietary and activity modifications to minimize bleeding risks.

[0007] Cardiac valve treatments, while lifesaving and improving overall cardiac function, carry significant risks of adverse effects. Next to meticulous surgical techniques, vigilant postoperative care, patient education and regular follow-up, comprehensive preoperative assessment is essential to mitigate these risks.SUMMARY

[0008] Certain aspects provide a method for predicting adverse effects due to cardiac valve treatments. The method includes obtaining at least one shape representation of a patient heart anatomy; determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy; providing the set of labeled patient data as inputs to a machine learning model; receiving, from the machine learning model, one or more risk scores representing predicted risk probabilities associated with performing one or more cardiac valve treatments on the patient heart anatomy; and generating, for display, an output comprising a risk assessment associated with performing the one or more cardiac valve treatments based on the one or more risk scores.

[0009] Certain aspects provide a method for predicting adverse effects due to cardiac valve treatments. The method includes obtaining at least one shape representation of a patient heart anatomy; determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy; providing the set of labeled patient data and a set of other patient datasets as inputs to a first machine learning model; receiving, from the first machine learning model: a subset of other patient datasets, of theset of other patient datasets, that comprises patient data that is similar to the set of labeled patient data, the subset of other patient datasets comprising: one or more first other patient datasets, of the subset of other patient datasets, that are associated with positive treatment outcome; and one or more second other patient datasets, of the subset of other patient datasets, that are associated with negative treatment outcome; and a first set of factors associated with positive treatment outcome and a second set of factors associated with negative treatment outcome; providing, to a second machine learning model, the subset of other patient datasets, the first set of factors associated with positive treatment outcome, the second set of factors associated with negative treatment outcome, and the set of labeled patient data; and receiving, from the second machine learning model, an output indicating a cohort predicted to have positive treatment outcome, the cohort comprising one or more third other patient datasets, of the subset of other patient datasets, sharing a set of similar patient characteristics, and a risk assessment comprising one or more risk scores for one or more negative treatment outcomes.

[0010] Other aspects provide processing systems configured to perform the aforementioned methods as well as those described herein; non-transitory, computer- readable media comprising instructions that, when executed by a processors of a processing system, cause the processing system to perform the aforementioned methods as well as those described herein; a computer program product embodied on a computer readable storage medium comprising code for performing the aforementioned methods as well as those further described herein; and a processing system comprising means for performing the aforementioned methods as well as those further described herein.

[0011] The following description and the related drawings set forth in detail certain illustrative features of one or more aspects.DESCRIPTION OF THE DRAWINGS

[0012] The appended figures depict certain aspects and are therefore not to be considered limiting of the scope of this disclosure.

[0013] FIG. 1 depicts a process flowchart associated with a process for predicting adverse effects of potential cardiovascular treatments.

[0014] FIG. 2 depicts a process flowchart associated with a process for predicting adverse effects of potential cardiovascular treatments using a plurality of Al models.

[0015] FIG. 3 depicts a process flowchart associated with an example subprocess for acquiring patient data.

[0016] FIG. 4 depicts a process flowchart associated with another example subprocess for acquiring patient data.

[0017] FIG. 5 depicts a process flowchart associated with an example subprocess for extracting features from acquired patient data.

[0018] FIG. 6 depicts a process flowchart associated with another example subprocess for extracting features from acquired patient data.

[0019] FIG. 7 depicts a process flowchart associated with an example subprocess for determining a preliminary treatment selection.

[0020] FIG. 8 depicts a process flowchart associated with an example subprocess for using Al model(s) to predict risks of adverse effects of cardiovascular treatments.

[0021] FIG. 9 depicts a process flowchart associated with an example subprocess for using a plurality of Al models to predict risks of adverse effects of cardiovascular treatments.

[0022] FIGS. 10-11 depict a process flowchart associated with example subprocesses for using a plurality of Al models to predict risks of adverse effects of cardiovascular treatments.

[0023] FIG. 12 depicts a process flowchart associated with an example subprocess for generating a risk assessment report.

[0024] FIG. 13 depicts an example process flowchart associated with a method for predicting adverse effects of cardiovascular treatments.

[0025] FIG. 14 depicts an example process flowchart associated with a method for predicting adverse effects of cardiovascular treatments.

[0026] FIG. 15 depicts an example processing system with which aspects of the present disclosure can be performed.

[0027] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the drawings. It is contemplated that elements and features of one embodiment may be beneficially incorporated in other embodiments without further recitation.DETAILED DESCRIPTION

[0028] Many medical conditions may be treated with a variety of treatment options. For example, certain pathologies of cardiac valves may be treated with open heart surgery, with transcatheter valve repair, or with transcatheter valve replacement. Treatments involving implantation of one or more implantable devices may, moreover, be performed using devices of varying manufacturers, brands, product families, types, sizes, or the like. Such devices may be implanted in different locations or orientations. Which of the many possible combinations of parameters constitutes the optimal treatment option depends on the individual patient’s characteristics.

[0029] Existing preoperative assessment practices focus mainly on simulations of individual treatments within representations of the full heart or a part of the heart containing the valve to be treated. Running such simulations is a labor-intensive approach, as each treatment option requires a separate simulation, and such a simulation only delivers insight into a limited set of risks. For example, virtually placing a 3D representation of a prosthetic mitral valve into a medical image set or a virtual 3D model of the left ventricle may give an indication of the risk of poor blood flow due to the implant blocking part of the left ventricle outflow tract but gives limited insight into the many other possible adverse effects. Thus, the labor intensiveness of simulation of any one treatment option for a particular patient prohibits evaluating all of them and selecting the best one for the patient.

[0030] Furthermore, in current practice, not all imaging modalities are widely used for studying the three-dimensional geometry of patient anatomy. In cardiac applications in particular, computer tomography (CT) is generally regarded as a reliable imaging modality for generating dimensionally accurate representations (e.g., 2D images, volume renderings, virtual 3D models, virtual 4D models, or the like) of a patient’s anatomy. CT data would therefore be a valuable source of information for any prescreening. However, due to the risks associated with exposure to radiation, the limited availability of CT equipment and the elevated cost of CT imaging, CT images are generally only acquired after prescreening, when the clinician has already decided to perform a more detailed analysis of a specific treatment option.

[0031] Echocardiography imaging, on the other hand, is routinely utilized in early patient examinations, and echocardiography data is commonly available at theprescreening stage. However, deriving dimensionally accurate representations of a patient’s anatomy from echocardiography data remains a challenge. Therefore, there is a technical problem as to how to perform preoperative assessment to predict adverse effects of cardiac treatments, such as to determine a suitable treatment for a patient. In particular, there is a technical problem as to how to efficiently compute such predictions, such as based on limited datasets. For example, current techniques may be computationally infeasible to perform due to the computational complexity of the number of simulations that may need to be run to assess multiple cardiac treatments.

[0032] Furthermore, in current practice, studies of the three-dimensional geometry of the patient anatomy generally only take certain parts of the anatomy into account, such as the blood pool volume, more particularly the blood pool volume around the valve in question, valve annuli and vessel lumina. However, certain other anatomical structures, such as valve leaflets, may have an important impact on the functioning of the heart, both before and after treatment. Including the geometry of the valve itself, inclusive of the valve leaflets, into the analysis is likely to provide a more accurate assessment.

[0033] Certain aspects herein provide techniques, such as methods and systems, which may provide a technical solution to these technical problems by leveraging advanced artificial-intelligence (Al) technology to analyze patient data. Certain aspects may accordingly provide the technical benefit of reduced computational complexity and / or reduced data needs (e.g., reliance on echocardiography imaging instead of CT) to assess multiple cardiac treatments, thereby improving the performance of the computing device performing such computations. By assessing, (e.g., among others), the patient’s anatomy’s geometric features as present in echocardiography data, certain aspects of techniques disclosed herein aim to provide personalized risk assessments for adverse outcomes, such as bleeding, infection, thromboembolism, structural deterioration, arrhythmia, etc. This predictive capability may assist clinicians in making informed decisions, such as about treatment strategies and / or postoperative management.

[0034] In some aspects, the present disclosure relates to systems and methods for predicting adverse effects due to cardiac valve treatments, such as valve repair, valve replacement, or the like. Certain aspects leverage advanced artificial-intelligence (Al) technology to analyze patient data, such as the geometry of a patient’s valve and, more particularly, the geometry of the valve leaflets. By assessing the valve’s geometricfeatures, the certain aspects herein aim to provide personalized risk assessments for adverse outcomes, such as bleeding, infection, thromboembolism, structural deterioration, arrhythmia, etc. This predictive capability may assist clinicians in making informed decisions about treatment strategies and postoperative management.

[0035] In some aspects, the present disclosure also relates to systems and methods for identifying the medical treatment with sufficient (e.g., the highest, above a threshold, etc.) chance of success based on patient characteristics. Although the remainder of this disclosure uses cardiac valve treatments, such as valve repair, valve replacement, or the like, as an example, the person skilled in the art will readily appreciate that the teachings equally apply to other kinds of medical treatments and other parts of the anatomy. Further, though certain aspects are discussed with respect to identifying the medical treatment with the highest chance of success, the teachings equally may apply to identifying a medical treatment with sufficient chance of success.Process Flowcharts for Predicting Adverse Effects

[0036] FIG. 1 depicts a process flowchart 100 associated with predicting adverse effects of potential cardiovascular treatments. In some aspects, process flowchart 100 includes process steps for data acquisition, feature extraction, preliminary treatment selection, using an Al model, and generating output.

[0037] At block 102, a (computing) system, such as system 1500 of FIG. 15, acquires data, including patient data for a target patient. Examples of detailed process flowcharts associated with data acquisition are depicted in FIG. 3-4. In some aspects, patient data may be acquired by means of a patient consultation, anamnesis, and / or examination. Alternatively, or additionally, in some aspects, patient data may be downloaded from a file, database, and / or data storage device. Patient data may comprise one or more of the following: patient history data, medical images, and / or a multi-dimensional model of anatomy of the patient, such as a patient heart anatomy. Patient history data may include one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, prior examination results, or other patient history data. Examples of lifestyle attributes include: active or passive lifestyle (e.g., as determined by an activity level of patient), smoking or non-smoking, alcohol use, and / or other lifestyle activities, habits, or attributes.

[0038] Medical images may comprise images of a patient’s heart, including partial views, such as of valve structures, and / or whole views of the patient’s heart. Medical images may be acquired using one or more scanning devices and / or may be loaded from a fde, database or data storage device. Imaging modalities associated with the medical images may include one or more of: echocardiography, magnetic resonance imaging (MRI), or computed tomography (CT) scans. Medical images may be taken with or without a contrast agent. In some aspects, data acquisition may relate to single-phase data, such as 3D data, multi-phase data, such as 4D data, and / or moving image data.

[0039] In some aspects, multi-dimensional models are generated based on the medical images acquired as part of data acquisition associated with block 102. In particular, reconstruction software may be used to reconstruct virtual multi-dimensional (e.g., 3D or 4D) models of the anatomy of the patient’s heart. The virtual multi-dimensional models may be generated fully automatically by a computing system, semi-automatically, or manually by a user of a computing system. The virtual multi-dimensional models may comprise any type of virtual multi-dimensional model known in the art, such as voxel clouds, solid models, boundary representations, polygon meshes, triangle meshes, parametric surface models, NURBS surface models... or any combination thereof. In some aspects, the virtual models include the valve leaflets, based on the patient’s medical images. In some aspects, the medical images are reconstructed using a process referred to as image segmentation. Image segmentation is the process of dividing a digital image into multiple segments or sets of pixels to simplify its representation and make it more meaningful for analysis. Image segmentation can be used to create a virtual model by identifying and extracting objects from multiple 2D medical images, allowing the segmented components to be reconstructed in multi-dimensional space through techniques such as the marching-cubes technique, photogrammetry or structure-from- motion. Beneficially, medical images from different sources and / or different imaging modalities may be combined, such as by registering image data sets and / or by registering 3D or 4D models created based on the image data sets.

[0040] In some aspects, the output of the subprocess associated with data acquisition comprises at least one shape representation of the patient heart anatomy. The shape representation includes the valve to be treated, or more particularly in certain aspects, one or more valve leaflets. In some aspects, the shape representation comprises an echocardiography dataset and / or a virtual 3D or 4D model generated based on theechocardiography dataset. In some aspects, the shape representation may comprise imaging data such as MRI or CT data that is combined to generate one or more virtual 3D or 4D models.

[0041] At block 104, the system extracts one or more features from the acquired data. Examples of detailed process flowcharts associated with feature extraction are depicted in FIG. 5-6. In some aspects, the extracted features include anatomical landmarks of the patient heart anatomy that may be identified and labeled based on the available patient data. In some aspects, the features are extracted from a shape representation that is generated based on the acquired patient data. Anatomical landmarks may include any features or anatomical parts generally present in the patient population. Some examples of anatomical landmarks include valve leaflets, papillary muscle heads, commissures, valve annuli, chambers, blood flow inlets or outlets, ostia of blood flow inlets or outlets, and / or vessel lumina. Some additional examples of anatomical landmarks may include any geometric primitives, including one or more primitive shapes (e.g., cylinders, (truncated) cones, spheres, cubes or prisms), points, lines, line segments, curves, polylines, splines, planes or surfaces, derived from said anatomical parts or used to represent said anatomical parts. For example, anatomical landmarks may comprise geometric center points of chambers or valve annuli, apex points of ventricles, best-fit planes through valve annuli, and / or curves, such as splines, along valve annuli.

[0042] Other anatomical landmarks may include previously implanted devices, such as prosthetic valves or clips, and / or components of such devices. Some example components include leaflets of a previously implanted prosthetic valve. Thus, “leaflets” herein may refer to native leaflets of the valve of the patient heart anatomy (e.g., aortic valve, mitral valve, tricuspid valve, or pulmonary valve) and / or leaflets of a previously implanted prosthetic valve.

[0043] Beneficially, various geometric parameters of the anatomical landmarks may be analyzed, such as one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane, and / or edge lengths. The spatial relationships, such as distances and / or orientations, between different anatomical landmarks may also be examined.

[0044] In some aspects, additional morphological features, such as one or more of: a leaflet calcification, mobility, or any anomalies, may be identified and quantified. In someaspects, one or more functional parameters may also be identified, such as ejection fraction or fractional shortening, based on the anatomical landmarks, the geometric parameters, and / or the morphological features. Ejection fraction is a measurement that indicates the percentage of blood that leaves the heart each time it contracts and serves as an indicator of how efficiently the heart is pumping blood. Fractional shortening is the percentage reduction in the diameter of the left ventricle during systole compared to diastole and serves as an indicator of how effectively the cardiac muscle shortens to pump blood through the circulatory system. Systole is the phase of the cardiac cycle when the heart muscle contracts and actively pumps blood out of the heart chambers, while diastole is the relaxation phase when the heart chambers fill with blood in preparation for the next contraction.

[0045] Anatomical landmarks, including valves and valve leaflets, as the case may be, may be identified manually or automatically (e.g., using feature-recognition techniques) on one or more of the available virtual (multi-dimensional) models of the patient heart anatomy and / or medical images. In some aspects, one or more virtual models may be labeled in their entirety as representing certain anatomical landmarks, such as leaflets. Additionally, or alternatively, one or more parts of a virtual model may be sectioned off and labeled as representing certain anatomical landmarks.

[0046] In some aspects, anatomical landmarks are labeled by adding one or more primitive objects, such as one or more control points, lines, line segments, polylines, splines, polygons, planes, primitive shapes (e.g., cylinders, (truncated) cones, prisms, cubes, spheres...), or NURBS surfaces to the available virtual models and / or medical images. In the case of previously implanted devices, or components of such devices, device and / or component geometry may be represented by primitive objects, such as the ones listed above, and / or may be derived from a virtual model of the prosthetic device, such as a computer-aided design (CAD) model. In some aspects, the components of such devices include leaflets of a previously implanted prosthetic device.

[0047] Thus, the labels for the anatomical landmarks may comprise one or more virtual primitive objects, virtual surface models, contours, control points, and / or any combination thereof. Some examples of virtual surface models include polygon mesh, triangle mesh, and / or parametric surface model. Some examples of contours include polylines and / or splines.

[0048] At block 106, the system selects one or more treatment options to be analyzed as potential treatments for the target patient based on the available information about the patient and the patient heart anatomy. An example of a detailed process flowchart associated with treatment selection is depicted in FIG. 7. In some aspects, the treatment options are selected by a user or automatically by a system, such as system 1500 of FIG. 15. In some aspects, the selection of one or more treatment options may be hardcoded or pre-programmed in the system.

[0049] In some aspects, each treatment option is associated with a specific combination of a plurality of treatment parameters. An example of a treatment parameter may be the type of treatment, such as valve repair or valve replacement. As such, a treatment option may include a particular type of treatment. Valve repair is a surgical procedure that preserves the patient’s own heart valve by reconstructing or reshaping the valve to restore proper function. Some techniques for valve repair include annuloplasty, leaflet repair, or chordal reconstruction. Valve replacement is a surgical procedure where a damaged or diseased heart valve is removed and replaced with a (prosthetic) mechanical or biological tissue valve or where such a mechanical or biological tissue valve is inserted into a damaged or diseased valve. Some biological tissue valves include valves made from human, pig, or cow tissue. In some aspects, valve repair is preferred over valve replacement, when feasible, because it maintains autologous valve anatomy and may avoid the need for long-term anticoagulation therapy typically needed with valve replacement. On the other hand, valve replacement is used when valve repair is not feasible due to extensive damage, calcification, or when the valve structure is too compromised to function properly even after a repair has been performed.

[0050] Another example of a treatment parameter may be whether the treatment would involve an implantable device and which implantable device. As such, a treatment option may include one or more implantable devices, such as a prosthetic valve, a clip, a suture, or the like, as part of either valve repair or valve replacement. Each implantable device may be further characterized by one or more parameters, such as device type, brand, product family, or size. Although typically meant for prescreening purposes, a treatment option may also include a target position for one or more implantable devices. The target position may refer to envisaged location and / or orientation of an implantable device after insertion into the patient anatomy. In some aspects, the target position is virtually simulated.

[0051] Another example of a treatment parameter may cover different ways of delivering the one or more implantable devices. As such, a treatment option may include information about the delivery of the one or more implantable devices, such as one or more of: the delivery pathway, the geometry or an aspect of the geometry of (part of) the trajectory, brand, type, size of delivery device, or any combination thereof. Examples of the delivery pathway may be via the inferior vena cava (I VC), which is a major vein that delivers deoxygenated blood from the lower body to the heart’s right atrium, or via the superior vena cava (SPC), which is a major vein that delivers deoxygenated blood from the upper body to the heart’s right atrium.

[0052] At block 108, the system is configured to provide inputs to one or more Al models. The one or more Al models may be used to predict the risk of adverse effects based on the available patient data and selected treatments. Any suitable Al algorithm may be employed to facilitate the prediction of the adverse effects, including neural networks, statistical shape models, support vectoring machines, machine learning algorithms, or any combination thereof. The inputs may comprise the acquired data associated with block 102, extracted features associated with block 104, and / or selected treatment options associated with block 106.

[0053] Before use, the Al model(s) may need to be trained. In some aspects, the Al model(s) are trained on a plurality of patient datasets comprising historical data relating to a set of previously treated patients. The patient datasets may comprise preoperative shape representations of the patients’ heart anatomies, labeled anatomical landmarks, data regarding previously administered treatments, and / or postoperative outcomes. In some aspects, the patient datasets comprise historical data, that is comparable or equivalent to the patient history data described above in relation to the data acquired at block 102. Anatomical landmark labels included in the plurality of patient datasets may also be comparable or topologically equivalent to the anatomical landmark labels generated with respect to block 104 when performing feature extraction. In some aspects, the patient datasets include echocardiography datasets, CT scans, and / or virtual models generated based on the echocardiography datasets and / or CT scans. In some aspects, the data regarding a previously administered treatment may comprise one or more of the treatment parameters described above. In some aspects, the postoperative outcomes may include information regarding the occurrence of one or more complications or adverse effects, such as any of the adverse effects mentioned above and / or other adverse effects. In someaspects, the postoperative outcomes include patient survival years and / or quality-of-life scores.

[0054] After training, the Al model(s) may be provided with any combination of gathered, acquired, and / or generated patient data. In some aspects, an Al model is provided with at least one shape representation of a target patient heart anatomy. In some aspects, the inputs processed by the Al model may include one or more labels representing anatomical landmarks, such as valve leaflets. In cases where one or more treatment options have been selected, the inputs may include data describing these treatment options, such as one or more of the treatment parameters described above. Inputs may include one or more of single-phase, multiple-phase, moving image data, and / or virtual models generated based on the image data.

[0055] One or more Al model(s) may be trained to correlate specific geometric features with the likelihood of one or more various adverse effects. Accordingly, at block 110, the system receives output from the Al model(s). The Al model(s) may generate risk scores and / or probabilities for each adverse effect recorded in the historical data. The adverse effects may comprise any of the adverse effects mentioned above and / or other adverse effects. If multiple treatment options are selected for analysis, for example in an order of preference, risk prediction may be performed by the Al model(s) in the order of preference indicated with the treatment options. In some aspects, the risk scores are compared against a threshold to further analyze and refine the treatments. For example, when a treatment option is reported to have a predicted risk below a certain threshold, no further treatment options may need to be evaluated because the treatment option has an acceptably low risk. In some aspects, the threshold may be a preset value in the system, a value chosen by a user, and / or an implicit part of the user’s assessment. In some instances, the system is configured to present the user with one or more predicted risks, wherein the system is configured to receive user input indicating whether the one or more predicted risks is acceptable or unacceptable without use of a defined or fixed threshold.

[0056] As described above, one or more Al model(s) may be used to facilitate the prediction of adverse effects of different treatment options. A potential drawback of using a single Al model to predict a treatment outcome for a target patient, might be that, to make the system function for a broad population of target patients, the Al model needs to be trained on historical data pertaining to patients with widely varying patientcharacteristics who have received varying treatments. This might reduce the reliability of the Al model’s prediction. A solution to this problem is to use a multi-step approach, in which a first Al model is used to select, from historical data pertaining to a large set of patients, only those datasets corresponding to patients that are similar to the target patient. Those patients will have undergone varying treatments with varying outcomes. A second Al model can then be trained on only that subset of datasets. Because it is trained on data with a high relevance for the target patient, its predictions when applied on the target patient’s data will be more reliable. Accordingly, FIG. 2 depicts a process flowchart 200 associated with predicting adverse effects of potential cardiovascular treatments using a plurality of Al models. Process flowchart 200 includes process steps for data acquisition (block 202), feature extraction (block 204), using a first Al model (block 206), receiving output from the first Al model (block 208), using a second Al model (block 210), and receiving output from the second Al model (block 212). In some aspects, block 202 is comparable to or the same as block 102 of FIG. 1, block 204 is comparable to or the same as block 104 of FIG. 1, blocks 206 and 210 are comparable to or the same as block 108 of FIG. 1, and block 212 is comparable to or the same as block 110 of FIG. 1.

[0057] In some aspects, a first Al model is trained as a pattern-recognition model. The first Al model is configured to receive a set of patient datasets associated with a plurality of previously treated patients and a target patient dataset associated with a target patient. The patient datasets may comprise the information as described above. The first Al model is then trained to identify a subset of patient datasets belonging to patients that are most similar to the target patient that will be treated. The subset of patient datasets is identified by the first Al model based on patient data acquired by the system at block 202 and / or based on features extracted by the system at block 204. In some aspects, the subset of patient datasets is identified as being similar to the target patient based on analyzing specific geometric features of the target patient anatomy and the geometric features included in the set of patient datasets.

[0058] This subset of patient datasets may be subdivided into a first plurality of patient datasets and a second plurality of patient datasets. The first plurality of patient datasets includes patient datasets that are similar to the target patient dataset and are associated with successful postoperative outcomes corresponding to one or more treatments that each patient corresponding to a dataset of the plurality of datasets received. The second plurality of patient datasets includes patient datasets that are similar to thetarget patient dataset and that are associated with unsuccessful postoperative outcomes corresponding to one or more treatments that each patient received.

[0059] In some aspects, the overall number of similar patient datasets to be identified may be predicted in the system or obtained as input from a user. Alternatively, a number of patient datasets may be predefined in the system or obtained as input from a user for each of the first and second pluralities of similar patient datasets (e.g., more or less patient datasets may be associated with either the first or second plurality of patient datasets).

[0060] Based on a comparison of the first plurality of patient datasets and the second plurality of patient datasets, the first Al model may predict which factors contribute to an increased risk of an unsuccessful postoperative outcome (e.g., risk of adverse effects) or an increased chance of a successful postoperative outcome. In some aspects, the first Al model may further predict the weight of each of these factors in the procedural outcome. Factors may be based on any of the data items included in the training dataset, such as patient characteristics or treatment option parameters. Thus, the system receives output from the first Al model at block 208, wherein the output comprises the first plurality of patient datasets, the second plurality of patient datasets, optionally a first set of factors associated with successful treatment outcomes, and optionally a second set of factors associated with unsuccessful treatment outcomes.

[0061] In some aspects, as depicted in FIG. 2, the system uses a second Al model to further facilitate the prediction of adverse effects of cardiovascular treatments. For example, at block 210, the system provides inputs to the second Al model. The inputs may include data acquired at block 202 and / or features extracted at block 204. In some aspects, the output received from the first Al model at block 208, which comprises the first plurality of patient datasets, the second plurality of datasets, and optionally the first and second sets of factors associated with successful and unsuccessful treatment outcomes, respectively, is used to train the second Al model to correlate patient characteristics, such as specific geometric features (of the patients’ hearts), with the treatment(s) (e.g., most) likely to deliver a successful postoperative outcome. The second Al model is then able to predict, among the various treatments described in the first and second pluralities of patient datasets, which treatment(s) may have a high likelihood of success for the target patient. These treatments may then be considered as treatment options for the target patient.

[0062] In some aspects, the output of the second Al model is the cohort that is (e.g. most) likely to deliver a successful postoperative outcome. “Cohort” may refer to a patient group sharing the same values for a defined, or predefined, selection of patient characteristics, such as any of the characteristics or parameters described in the first and second pluralities of patient datasets. For example, the selection of patient characteristics may comprise treatment data, such as treatment type, device brand, or any other feature that may be identified within a patient dataset, such as features associated with extracting features at block 204.

[0063] Thus, in some aspects, all patients being treated with the same treatment type (e.g., valve repair or valve replacement) may be included in a cohort. In another example, all patients being treated with the same treatment type and same device brand form a cohort. In another example, all patients being treated with the same treatment type, same device brand, and same device size may form a cohort. Thus, a cohort comprises a plurality of patient datasets that share at least one patient characteristic but can share any number of patient characteristics.

[0064] In some aspects, the output provided by the second Al model is an optimal cohort of patient datasets. Because it was trained with patient datasets that are similar to the target patient dataset, the second Al model is able to predict whether a particular treatment that had a successful postoperative treatment for a patient similar to the target patient will also lead to a successful postoperative treatment for the target patient. In some aspects, the optimal postoperative outcome is the postoperative outcome that has the lowest predicted risk of an unsuccessful postoperative outcome of the cardiovascular treatment, out of all available cohorts of patient datasets.

[0065] In addition to predicting successful treatment options, in some aspects, the second Al model is also configured to predict the risk of adverse effects of the treatment options predicted to have a successful postoperative outcome for the target patient. For example, even if a particular treatment option is predicted to have a successful postoperative outcome for the target patient, the target patient could still experience one or more adverse effects of the treatment. The second Al model may also be configured to identify any parameters included in the patient datasets that may increase or reduce the risk of experiencing those adverse effects.

[0066] In some aspects, the second Al model may be trained to correlate patient characteristics with a likelihood of various adverse effects. For example, the second Al model may generate risk scores and / or probabilities for each adverse effect recorded in the historical data of the patient datasets. In some aspects, the second Al model may be trained to correlate treatment option parameters with the likelihood of various adverse effects. Some treatment option parameters include remaining treatment parameters that were not used to define the cohort. The second Al model may identify the weights of individual treatment option parameters on the likelihood of the adverse effects. Additionally, or alternatively, the second Al model may be trained to predict to what extent a particular treatment option parameter should be adjusted to in order to obtain a change in the likelihood of an adverse effect (e.g., a decrease in the likelihood of the adverse effect).

[0067] In some aspects, the second Al model may be trained to correlate patient characteristics and a specific cohort with an expected treatment outcome. Some treatment outcomes include patient survival years and / or quality-of-life scores. Patient survival years refers to the number of years that the patient is predicted to live after the cardiovascular treatment is performed. Quality-of-life scores refer to a score, defined on a scale, which indicate a quality of life that is predicted to be experienced by the patient after the cardiovascular treatment is performed. In some aspects, the second Al model may generate a quality-of-life score that is an average for all survival years. In some aspects, the second Al model may generate a quality-of-life score for each year predicted in the survival years.Data Acquisition

[0068] FIG. 3 depicts an example of a process flowchart associated with a subprocess for acquiring patient data and generating a shape representation of a patient’s heart. In some aspects, the subprocess depicted in FIG. 3 is associated with acquiring data at block 102 of FIG. 1 and / or block 202 of FIG. 2. Process flowchart 300 begins at block 302, which represents the start of the depicted subprocess associated with data acquisition. In particular, a system, such as system 1500 of FIG. 15, gathers patient data at block 304, including patient history 306, as part of a target patient dataset associated with a target patient. The target patient is the patient who will be receiving a cardiovascular treatment. As described above, the patient history may include reported symptoms, previouspathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, and / or prior examination results of the target patient. The system then acquires echocardiography (“echo”) data at block 308. In some aspects, the system only acquires echocardiography data at this stage as part of the prescreening process for the target patient. This may avoid unnecessarily taking additional images such as MRI or CT scans, saving on expenses time and logistical burden. Accordingly, as another benefit to only acquiring echocardiography data at this stage, the target patient’s exposure to radiation is reduced or eliminated. The echocardiography data 310 may be 2D, 3D, or 4D. In some aspects, the echocardiography data 310 may be acquired using an external probe, transesophageal echocardiography (TEE / TOE), or intracardiac echocardiography (ICE), or other methods.

[0069] External probe echocardiography involves placing an ultrasound transducer on the chest wall to capture images of the heart through skin and tissues. This non- invasive approach is a common method for cardiac imaging and allows visualization of heart chambers, valves, and blood flow. TEE / TOE involves inserting a specialized ultrasound probe into the esophagus, positioning it directly behind the heart. This provides superior image quality by eliminating interference from the chest wall and lungs, making it particularly valuable for detailed assessment of heart valves, detecting blood clots, and guiding certain cardiac procedures. ICE utilizes a catheter-based ultrasound probe that is inserted through a blood vessel and advanced into the heart chambers. This invasive approach offers real-time, high-resolution imaging from within the heart itself and is primarily used during interventional procedures such as transcatheter valve repair, closure of septal defects, or ablation procedures for arrythmias.

[0070] As illustrated by decision block 312, in certain embodiments, the system generates a virtual (multi-dimensional) model of the patient heart anatomy based on echocardiography data 310. In other embodiments, the system does not generate a virtual model of the patient heart anatomy. In those embodiments where the system does generate a virtual model, the system proceeds to block 314 and generates the virtual model, such as virtual model 316, of the patient heart anatomy. In some aspects, the virtual model is a 3D or 4D model. As described above, the system is able to generate the virtual model through image segmentation. In some aspects, the image segmentation is a manual segmentation, an automatic segmentation, an Al-assisted segmentation, or a combination thereof.

[0071] Some automatic segmentation methods may use a statistical shape model (SSM) fit. SSM fit is a mathematical technique used in image segmentation that applies a pre -trained statistical model of anatomical shape variations to identify and extract relevant structures from medical images. The SSM captures the statistical distribution of shape variations across a population and constrains the segmentation process to anatomically plausible shapes, helping to create accurate virtual models, even when working with noisy or incomplete image data. Thus, an SSM fit may help ensure that segmented heart structures, like chambers and valves, conform to realistic anatomical patterns, thereby improving the accuracy of virtual heart models used for diagnosis and treatment planning.

[0072] The system generates the virtual model 316, which may, for example, be configured as a voxel cloud, boundary representation, or SSM instance. Some examples of boundary representations include parametric shapes or polygon meshes, such as triangle meshes.

[0073] The echocardiography data 310 and / or the virtual model 316, if available, may serve as at least one shape representation 318. The shape representation 318 may comprise echocardiography data, virtual model data, or a combination thereof. Process flowchart 300 ends at block 320, which represents the end of the depicted subprocess associated with data acquisition. Shape representation 318 may serve as input for subsequent subprocesses, such as extracting features at block 104 of FIG. 1 or extracting features at block 204 of FIG. 2.

[0074] FIG. 4 depicts a process flowchart associated with another example of a subprocess for acquiring patient data and generating a virtual model of a patient’s heart. In some aspects, the subprocess depicted in FIG. 4 is associated with acquiring data at block 102 of FIG. 1 and / or block 202 of FIG. 2. Process flowchart 400 begins at block 401, which represents the start of the depicted subprocess associated with data acquisition. In particular, a system, such as system 1500 of FIG. 15, gathers patient data at block 402 for a target patient. In some aspects, the patient data includes patient history 404 as part of a target patient dataset associated with the target patient. As described above, the patient history may include reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, and / or prior examination results of the target patient.

[0075] As illustrated by decision block 406, in certain embodiments, the system generates a virtual (multi-dimensional) model of the patient heart anatomy based on medical images. In other embodiments, the system does not generate a virtual model of the patient heart anatomy. In those embodiments where the system does generate a virtual model, the system first acquires the necessary medical images. As illustrated by decision block 408, in certain embodiments, the system acquires CT data, in other embodiments echocardiography data, or, in yet other embodiments, a combination of both. If the system only acquires echocardiography data, the system acquires the echocardiography data at block 410. At block 412, the system then generates a virtual model, such as virtual model 414, based on the echocardiography data. Virtual model 414, optionally combined with the echocardiography data acquired at block 410 may serve as at least one shape representation 432. Process flowchart 400 then ends at block 416, which represents the end of the depicted subprocess associated with data acquisition.

[0076] If the system only acquires CT data, the system acquires the CT data at block 418. At block 420, the system generates a virtual model, such as virtual model 422, based on the CT data. Virtual model 422, optionally combined with the CT data acquired at block 418 may serve as at least one shape representation 432. Process flowchart 400 then ends at block 416, which represents the end of the depicted subprocess associated with data acquisition.

[0077] If the system acquires both echocardiography data and CT data to generate the virtual model of the patient heart anatomy, the system acquires the echocardiography data at block 410 and acquires the CT data at block 418. Next, the system generates a virtual model based on the echocardiography data at block 412 and generates a virtual model based on the CT data at block 420. After generating virtual model 414 and virtual model 422, the system registers the virtual model 414 and virtual model 422 at block 424. Registering a set of virtual models may include identifying one or more correlation points within each virtual model that can be matched, in order to correlate the different virtual models with each other, including correlating a position and an orientation of each virtual model so as to make both virtual models coincide in a common coordinate system. Subsequently, the system combines virtual model 414 and virtual model 422 at block 426 to generate at least one shape representation 432. Combining virtual models may comprise generating a new virtual model comprising features derived from both source models. For example, certain anatomical features, such as blood pool volume, valveannuli, blood flow inlets or outlets, ostia of blood flow inlets or outlets, or vessel lumina, may be well discernable in CT data with higher dimensional accuracy than in echocardiography data, while other anatomical features, such as features with a thickness below the resolution of conventional CT data (e.g., valve leaflets), may be more easily discernable in echocardiography data. Combining virtual model 414 and virtual model 422 may then comprise complimenting virtual model 422, which may comprise dimensionally accurate representations of the blood pool volume around the valve and the valve annulus, with representations of the valve leaflets taken from virtual model 414. Process flowchart ends at block 416, which represents the end of the depicted subprocess associated with data acquisition.

[0078] In those embodiments where the system does not generate a virtual model, the system acquires image data, such as image data 430 at block 428. Image data 430 may comprise echocardiography data, CT data, MRI data, and / or data related to any other medical imaging modality known in the art. Image data 430 may serve as at least one shape representation 432. Process flowchart 400 then ends at block 416, which represents the end of the depicted subprocess associated with data acquisition.

[0079] Shape representation 432 may serve as input for subsequent subprocesses, such as extracting features at block 104 of FIG. 1 or extracting features at block 204 of FIG. 2.

[0080] Acquiring image data at block 428, acquiring echocardiography data at block 410 and acquiring CT data at block 418 may each comprise using an image acquisition device, such as an ultrasound probe, a CT scanning device and the like. Additionally, or alternatively, each of these steps may comprise loading image data from a fde, database or data storage device.

[0081] Generating model at block 412 and block 420 may each comprise using any of the techniques described above, such as at block 314 of FIG. 3.

[0082] Virtual (multi-dimensional) model 414 and / or virtual model 422 may be 2D, 3D or 4D virtual models and may, for example, be configured as a voxel cloud, boundary representation, or SSM instance. Some examples of boundary representations include parametric shapes or polygon meshes, such as triangle meshes.Feature Extraction

[0083] As mentioned above, FIGS. 5-6 are associated with extracting features at block 104 of FIG. 1 and / or block 204 of FIG. 2. Accordingly, FIG. 5 depicts a process flowchart associated with a subprocess for extracting features, including anatomical landmarks identified in the shape representation based on target patient dataset. In particular, process flowchart 500 begins at block 502, which represents the start of the depicted subprocess associated with feature extraction. At block 504, the system then identifies anatomical landmarks within the shape representation of the patient heart anatomy, such as shape representation 318 of FIG. 3 or shape representation 432 of FIG. 4. Landmarking may be performed manually, automatically (e.g., using an Al model), using an SSM fit, or a combination thereof. The anatomical landmark labels generated at block 506 may include coordinates of anatomical landmarks, primitive objects (such as described above), dimensions of anatomical landmarks, derived functional parameters, and / or SSM parameters. Examples of anatomical landmarks include native landmarks associated with the patient heart, such as papillary muscle heads, commissures, valve annuli, chambers, chamber centers, inlets / outlets, ostia of inlets / outlets, and / or apex of ventricles, or non-native landmarks associated with a previously implanted device. Process flowchart 500 ends at block 508, which represents the end of the depicted subprocess associated with feature extraction.

[0084] FIG. 6 depicts a process flowchart associated with another example of a subprocess for feature extraction. In some aspects, the extracted features include leaflets of the patient heart anatomy. Accordingly, FIG. 6 depicts process flowchart 600 for identifying and labeling leaflets of the patient heart anatomy. In particular, process flowchart 600 starts at block 602, which represents the start of the depicted subprocess associated with feature extraction. At block 604, the system identifies leaflets of one or more valves of the patient heart anatomy within the shape representation of the patient heart anatomy, such as shape representation 318 of FIG. 3 or shape representation 432 of FIG. 4. After identifying the leaflets, the system labels the leaflets at block 606. In some aspects, the system, such as system 1500 of FIG. 15, labels an entire virtual model or shape representation as a leaflet. In some aspects, the system labels part of the virtual model or shape representation as a leaflet. The system may add control points, lines, polylines, splines, polygons, and / or NURBS surfaces to a virtual model or images to label the leaflets. In some aspects, the system utilizes an Al model trained to perform featurerecognition in order to generate Al-generated labels for anatomical landmarks, such as leaflets.

[0085] Leaflets that are labeled may include native leaflets of a valve, such as the aortic valve, mitral valve, tricuspid valve, and / or pulmonary valve. Labeled leaflets may also include leaflets of previously implanted prosthetic valves. In the case of previously implanted prosthetic valves, leaflet geometry may be labeled based on image data, CAD fdes, and / or a virtual model of the previously implanted prosthetic valves. The leaflet labels may be configured as surface models, such as polygon meshes (e.g., triangle meshes) or parametric shapes, contours, such as polylines and splines, and / or control points.

[0086] After labeling the leaflets, process flowchart ends at block 608, which represents the end of the depicted subprocess associated with feature extraction.

[0087] The subprocess depicted in process flowchart 600 of FIG. 6 may be a subprocess of the subprocess depicted in process flowchart 500 of FIG. 5. As such, block 604 of identifying leaflets may be a substep of block 504 of identifying landmarks, and block 606 of labeling leaflets may be a substep of block 506 of labeling landmarks.Preliminary Treatment Selection

[0088] FIG. 7 depicts a process flowchart associated with selecting one or more preliminary treatment options, for example, as described with respect to block 106 of FIG. 1. The preliminary treatment selection comprises one or more treatment options that will be analyzed to determine the risk of adverse effects that may be experienced by the target patient. In some aspects, the treatment options are cardiovascular treatments, including different versions of valve repair and / or valve replacement treatments. Process flowchart 700 begins at block 702, which represents the start of the depicted subprocess. The system determines one or more treatment options as part of a preliminary treatment selection at block 704. The system outputs treatment selection 706 comprising the one or more treatment options. In some aspects, each treatment option includes one or more treatment attributes, such as type of treatment, device type, device position, and / or delivery pathway. Some examples of types of treatments include valve repair and valve replacement. Process flowchart 700 ends at block 708, which represents the end of the depicted subprocess.Using an Al Model for Predicting Adverse Effects

[0089] FIG. 8 depicts a process flowchart associated with a subprocess for predicting adverse effects using one or more Al models. In some aspects, process flowchart 800 depicts a subprocess associated with blocks 108-110 of FIG. 1. Process flowchart begins at block 802, which represents the start of the depicted subprocess. The system first gathers inputs to provide to the Al model(s) at block 804. In some aspects, the inputs comprise extracted features (such as the features extracted in block 104 of FIG. 1, or block 204 in FIG.2), landmark labels (such as labels generated in block 506 of FIG. 5), leaflet labels (such as labels generated in block 606 of FIG. 6), virtual models, shape representations, image data, single-phase data, multi-phase data, moving-image data, the preliminary treatment selection, and / or patient history. In some aspects, the Al model(s) are configured as neural networks, SSMs, support vectoring machines, machine learning models, or combination thereof. The Al model(s) are configured to predict adverse effects of the treatment selection, such as treatment selection 706, for the target patient based on the available inputs gathered at block 804.

[0090] At block 808, the system receives output from the Al model(s). In some aspects, block 808 is comparable to block 110 of FIG. 1. In some aspects, the output includes one or more predicted adverse effects of the one or more treatments included in the treatment selection. Some examples of adverse effects include vessel obstruction, paravalvular leakage, malpositioning of device, and / or device migration. Vessel obstruction refers to a blockage or narrowing of a blood vessel resulting from thrombosis, device components, or tissue growth that impedes blood flow. Paravalvular leakage refers to the backward flow of blood around, rather than through, a prosthetic heart valve due to incomplete sealing between the valve and the native tissues of the patient heart. Malpositioning of the device refers to the improper placement or orientation of a prosthetic implant that reduces its effectiveness and may cause further complications. Device migration refers to the movement of an implanted device from its intended position, which can lead to device dysfunction, vessel damage, or obstruction.

[0091] After receiving the output from the Al model(s), process flowchart 800 ends at block 810, which represents the end of the depicted subprocess.Using a Plurality of Al Models for a Plurality of Treatment Options

[0092] As described above, one or more Al model(s) may be used to facilitate the prediction of adverse effects of different treatment options. A potential drawback of using a single Al model to predict a treatment outcome for a target patient, might be that, to make the system function for a broad range of treatment options, the Al model needs to be trained on historical data pertaining to patients with widely varying patient characteristics who have received varying treatments. This might reduce the reliability of the Al model’s prediction. A solution to this problem is to use a plurality of Al models, each trained on historical data pertaining to a single treatment option.

[0093] Accordingly, FIG. 9 depicts a process flowchart associated with predicting adverse effects using a plurality of Al models. In some aspects, the process flowchart 900 depicted in FIG. 9 is an alternate implementation and configuration of blocks 106-110 depicted in FIG. 1. FIG. 9 shows an example elaborated for a limited set of two eligible treatment options: valve repair and valve replacement. The skilled person will readily understand that the principle can be applied to any plurality of treatment options. Process flowchart 900 begins at block 902, which represents the start of the depicted subprocess. At decision block 904, the system determines whether to prioritize valve repair as a treatment. If the system, such as system 1500 of FIG. 15, determines or receives instructions from a user of the system to prioritize valve repair, the system provides inputs to the first Al model at block 906. The inputs may include patient history, echocardiography data, landmark data, virtual model data, or a combination thereof. At block 908, the system receives output from the first Al model, including one or more risks of adverse effects associated with valve repair as a treatment option for the target patient. In some aspects, the first Al model is configured to predict risks of adverse effects associated with various valve repair treatment options selected in a treatment selection, such as treatment selection 706 of FIG. 7.

[0094] At decision block 910, the system determines whether the predicted one or more risks are acceptable. To this end, the system may comprise one or more built-in or pre-determined thresholds to evaluate the predicted risks. Alternatively, the system may present information regarding the predicted risks to a user and may receive from the user an instruction as to whether the one or more risks are acceptable. If the one or more risks are acceptable, the system plans the valve repair treatment at block 912. If the risks arenot acceptable or if the system determines or receives instructions from a user of the system, at decision block 904, to not prioritize valve repair, the system utilizes a second Al model at block 914. In some aspects, the second Al model is configured to predict one or more risks of adverse effects of valve replacement as a treatment option for the target patient.

[0095] At block 914, the system provides inputs to the second Al model. Similar to the inputs provided to the first Al model, the input data provided to the second Al model may include patient history data, echocardiography data, landmark data, virtual model data, or a combination thereof. At block 916, the system receives output from the second Al model. The output includes one or more risks of adverse effects associated with valve replacement as a treatment option for the target patient. In some aspects, the second Al model is configured to predict risks of adverse effects associated with various valve replacement treatment options selected in a treatment selection, such as treatment selection 706 of FIG. 7.

[0096] The system determines whether the risks are acceptable at decision block 918. To this end, the system may comprise one or more built-in or pre-determined thresholds to evaluate the predicted risks. Alternatively, the system may present information regarding the predicted risks to a user and may receive from the user an instruction as to whether the one or more risks are acceptable. If the risks are acceptable, the system, a user of the system, a medical professional or a trained non-medical professional plans the valve replacement procedure for the target patient at block 920. If the one or more risks are not acceptable, the system, a user of the system, a medical professional or a trained non-medical professional evaluates alternative treatment options at block 922. Some examples of alternative treatment options include medication, transcatheter procedures, balloon valvuloplasty, or lifestyle modifications. In some aspects, the system uses one or more further Al models configured to predict risks of adverse effects associated with one or more of these alternative treatment options, respectively.Using a Plurality of Al Models for Cohort Selection

[0097] In some aspects, process flowchart 1000 depicted in FIG. 10 is a subprocess associated with the first Al model referenced in FIG. 2 and process flowchart 1100 depicted in FIG. 11 is a subprocess associated with the second Al model referenced in FIG. 2.

[0098] As shown in FIG. 10, process flowchart 1000 begins at block 1002, which represents the start of the depicted subprocess from the perspective of the first Al model. The first Al model processes the inputs at block 1004. The inputs may include patient history data, image data, landmark data, virtual model data, or a combination thereof for a plurality of previously treated patients and for a target patient. At block 1006, the first Al model identifies a set of one or more patients of the plurality of previously treated patients that are similar to the target patient. To this end, any suitable pattern recognition technique known in the field of artificial intelligence, or more particularly machine learning, such as statistical pattern recognition, syntactic pattern recognition, neural pattern recognition or template matching. For example, a neural network may analyze the inputs relating to the previously treated patients to derive and uncover meaningful features. The extracted features may then be segmented into what constitutes the patterns. The identified inherent patterns and insights and the dataset for the target patient may be fed into the model for either class or cluster prediction. The set of previously treated patients that fall in the same class or cluster as the target patient may then serve as the identified set of one or more patients of the plurality of previously treated patients that are similar to the target patient. The first Al model classifies one or more patients of the identified set of similar previously treated patients into a first subset of patients with unsuccessful outcomes (at block 1010) and a second subset of patients with successful outcomes (at block 1008). At block 1012, the first Al model is further configured to compare the first and second set of patients with each other. At optional block 1014, the first Al model predicts one or more risk factors associated with treatments performed on patients corresponding to the first and second set of patients based on comparing the first and second subsets of patients with each other. Additionally or alternatively, in some aspects, the first Al model is configured to predict which factors contribute to successful outcomes and which factors contribute to unsuccessful outcomes. Some example factors may include size of valve annulus, presence of cardiac arrythmias, obesity, diabetes, and / or blood cholesterol levels.

[0099] Additionally, in some aspects, the first Al model is configured to determine the weight of each factor in the procedural outcome (e.g., how much a particular factor influenced either the successful or unsuccessful postoperative outcome). For example, for the risk of stroke, the presence of arrythmias may be weighted higher than the size of the valve annulus. In another example, for the risk of patient prosthesis mismatch, the weightof the size of the valve annulus may be higher than cholesterol blood values. Process flowchart 1000 ends at block 1016, which represents the end of the depicted subprocess.

[0100] As mentioned above, in some aspects, process flowchart 1100 depicted in FIG. 11 is a subprocess associated with the second Al model referenced in FIG. 2. As shown in FIG. 11, process flowchart 1100 begins at block 1102, which represents the start of the depicted subprocess from the perspective of the second Al model. In some aspects, block 1102 is comparable to block 1016 of FIG. 10, such that the end of process flowchart 1000 is the beginning of process flowchart 1100.

[0101] The system provides inputs to the second Al model at block 1104. In some aspects, the inputs include a first subset of patient datasets associated with previously treated patients that are similar to the target patient that have successful outcomes, a second subset of patient datasets associated with previously treated patients that are similar to the target patient associated with patients that have unsuccessful outcomes, and a target patient dataset corresponding to the target patient. Each patient dataset, of the first set of patient datasets, second set of patient datasets, and / or the target patient dataset, may include patient history data, image data, landmark data, virtual model data, or combination thereof. Each patient dataset of the first set of patient datasets and the second set of patient datasets further includes data regarding previously administered treatments, and postoperative outcomes. In some aspects, the data regarding a previously administered treatment may comprise one or more of the treatment parameters described above. In some aspects, the postoperative outcomes may include information regarding the occurrence of one or more complications or adverse effects, such as any of the adverse effects mentioned above and / or other adverse effects. In some aspects, the postoperative outcomes may comprise patient survival years and / or quality-of-life scores.

[0102] At block 1106, the second Al model is configured to generate one or more Al model predictions. In some aspects, the Al model predictions include a treatment that is predicted to have a successful postoperative outcome for the target patient, as described above in the context of FIG. 2. In some aspects, the treatment is predicted based on a patient cohort with a successful procedure outcome corresponding to the treatment. Additionally, in some aspects, the Al model predictions include one or more risks associated with the predicted treatment. The second Al model may further determinedependencies or parameters included in the patient cohort data that may influence the likelihood of experiencing adverse effects.

[0103] In some aspects, the system then determines whether the one or more risks are acceptable at optional block 1108. To this end, the system may comprise one or more built-in or pre-determined thresholds to evaluate the predicted risks. Alternatively, the system may present information regarding the predicted risks to a user and may receive from the user an instruction as to whether the one or more risks are acceptable. If the one or more risks are acceptable, the second Al model generates additional output at optional block 1110. In some aspects the generated additional output includes survival years and / or quality-of-life score for the target patient, if the target patient were to receive the procedure corresponding to the patient cohort and successful procedure outcome. What additional output can be generated depends on what postoperative data is available in the patient datasets used to train the second Al model. The system, a user of the system, a medical professional or a trained non-medical professional then plans the procedure for the target patient at block 1112. This may comprise preparing a detailed preoperative virtual simulation of the proposed treatment option on the target patient shape representation and final determination of some or all treatment parameters.

[0104] If the one or more risks are not acceptable, the second Al model may identify which one or more parameters (e.g., of the set of labeled patient data or of the treatment option) can be adjusted and to what extent to obtain a reduced risk of the adverse effects at optional block 1114, based on the earlier determined dependencies or parameters included in the patient cohort data that may influence the likelihood of experiencing adverse effects. For example, the Al model may determine that the risk of a certain adverse effect will decrease to an acceptable level if the target patient loses a certain amount of weight. For example, the second Al model may identify that lowering the blood cholesterol levels of the patient may reduce the risk of an adverse effect, or resolving an arrythmia may lower the risk of the adverse effect.

[0105] Not all of the proposed parameter changes may be feasible. For example, it may not be possible to adjust certain parameters at all, such as certain dimensions of anatomical features. For other parameters, whether it is possible to adjust them will depend on the specifics of the target patient. For example, for some patients, it may be possible to adjust other parameters, such as blood cholesterol levels, through a change inlifestyle, while other patients might not be willing to make such a change in lifestyle. Yet other patients may be in such an urgent need of treatment that there is not enough time for a change of lifestyle to take effect. Thus, in some aspects, the system or a user of the system then determines whether the adjustments to the parameters are feasible at block 1116. If the parameter adjustment is feasible, the system, a user of the system, a medical professional or a trained non-medical professional plans the updated procedure for the target patient with the updated parameters at block 1118. In this manner, the second Al model is configured to predict a procedure and procedure parameters that will yield a high or highest likelihood of postoperative success for the target patient, as compared to different procedures and corresponding procedure parameters.

[0106] If the adjustments are not feasible, the system, a user of the system, a medical professional or a trained non-medical professional evaluates alternative treatments at block 1120. In some aspects, the system evaluates alternative treatments, such as alternative treatments associated with block 922 of FIG. 9.

[0107] Some parameter adjustments may not be feasible, such as changing the size of a valve annulus diameter. This is because no preoperative treatment, therapy, or lifestyle change may affect the size of the valve annulus diameter. In contrast, other types of parameters adjustments may be feasible, such as activity level, cholesterol blood level, smoking, or other lifestyle-related parameter adjustments. In some aspects, the feasibility of lifestyle-related parameter adjustments may be determined by a patient’s willingness to make lifestyle adjustments or may be related to whether the cardiac treatment can be postponed long enough for the lifestyle change to affect the identified treatment parameter.Risk Assessment Reports

[0108] FIG. 12 depicts a process flowchart associated with a subprocess for generating a risk assessment report. As shown in FIG. 12, process flowchart 1200 begins at block 1202, which represents the start of the depicted subprocess. The system then displays one or more risk assessments at block 1204. In some aspects, the risk assessments are based on predicted adverse effects associated with output associated with block 110 of FIG. 1 and / or output associated with block 210 of FIG. 2.

[0109] In some aspects, the system then generates a report 1208 at optional block 1206. The report may present one or more predicted risks of adverse effects associated with one or more treatment options in any suitable form, such as numerically or graphically. Report 1208 may further comprise some or all available target patient data acquired, gathered or generated in any of process steps 102-104 of FIG. 1, process steps 202-204 of FIG. 2, or the subprocesses of FIG. 3-6. Report 1208 may be a digital file in any suitable format known in the art, such as a PDF file, or in a dedicated format for displaying the report in the system or in a dedicated software, such as on a mobile device. Additionally, or alternatively, report 1208 may come in tangible form as a printed report.

[0110] Process flowchart 1200 ends at block 1210, which represents the end of the depicted subprocess.Example Methods for Predicting Adverse Effects due to Cardiac Valve Treatments

[0111] FIG. 13 depicts an example method 1300 for predicting adverse effects due to cardiac valve treatments. In one aspect, method 1300 can be implemented by the processing system 1500 of FIG. 15 or can be performed by a user using processing system 1500.

[0112] Method 1300 begins at block 1305 with obtaining at least one shape representation of a patient heart anatomy.

[0113] Method 1300 then proceeds to block 1310 with determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy.

[0114] Method 1300 then proceeds to block 1315 with generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy.

[0115] Method 1300 then proceeds to block 1320 with providing the set of labeled patient data as inputs to a machine learning model.

[0116] Method 1300 then proceeds to block 1325 with receiving, from the machine learning model, one or more risk scores representing predicted risk probabilities associated with performing one or more cardiac valve treatments on the patient heart anatomy.

[0117] Method 1300 then proceeds to block 1330 with generating, for display, an output comprising a risk assessment associated with performing the one or more cardiac valve treatments based on the one or more risk scores.

[0118] In some aspects, the patient heart anatomy is associated with a set of patient data and block 1305 includes generating the at least one shape representation based on the set of patient data.

[0119] In some aspects, the set of patient data further comprises patient history data.

[0120] In some aspects, the patient history data includes one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, or prior examination results of a patient associated with the patient heart anatomy.

[0121] In some aspects, the set of patient data further comprises one or more images of the patient heart anatomy, wherein each image depicts at least part of an anatomical landmark of the one or more anatomical landmarks.

[0122] In some aspects, the at least one shape representation comprises one or more of the one or more images of the patient heart anatomy.

[0123] In some aspects, the one or more images of the patient heart anatomy comprise one or more of: echocardiography images, MRI images, or CT images.

[0124] In some aspects, at least one image of the one or more images of the patient heart anatomy comprises an image taken with contrast.

[0125] In some aspects, the one or more images of the patient heart anatomy comprise one or more of: single-phase data or moving-image data.

[0126] In some aspects, generating the at least one shape representation comprises: generating a virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy, wherein the at least one shape representation comprises the virtual model of the patient heart anatomy.

[0127] In some aspects, generating the virtual model of the patient heart anatomy comprises: generating a plurality of preliminary virtual models, each virtual model based on one or more images, registering the plurality of preliminary virtual models so as to make them coincide in a common coordinate system, and combining features of theplurality of preliminary virtual models to generate the virtual model of the patient heart anatomy. In some aspects, the one or more images of the patient heart anatomy comprises images of a plurality of image modalities and each of the plurality of preliminary virtual models corresponds to one of the plurality of image modalities. In some aspects, the plurality of image modalities comprises CT and echocardiography. In some aspects, combining features of the plurality of preliminary virtual models associated with a plurality of image modalities comprises combining features of a plurality of feature types, and selecting as a source for each feature the specific preliminary virtual model of the plurality of preliminary virtual models associated with the specific image modality in which the feature’s feature type is (well / best) discernable. In some aspects, combining features of the plurality of virtual models comprises combining one or more of a blood pool volume, a valve annulus, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen sourced from a virtual model associated with CT data, with one or more valve leaflets sourced from a virtual model associated with echocardiography data.

[0128] In some aspects, generating the virtual model of the patient heart anatomy is based on using image segmentation on the one or more images of the patient heart anatomy.

[0129] In some aspects, generating the virtual model of the patient heart anatomy comprises: providing the one or more images of the patient heart anatomy to a machine learning model configured to generate virtual models; and receiving, from the machine learning model, the virtual model of the patient heart anatomy.

[0130] In some aspects, the virtual model of the patient heart anatomy is a three- dimensional model.

[0131] In some aspects, the virtual model of the patient heart anatomy is a fourdimensional model.

[0132] In some aspects, the one or more anatomical landmarks comprise one or more anatomical parts of a patient heart.

[0133] In some aspects, the one or more anatomical parts comprise one or more of: a valve, a valve leaflet, a papillary muscle head, a commissure, a valve annulus, a chamber, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen.

[0134] In some aspects, the one or more anatomical landmarks comprise one or more implanted devices present within the patient heart anatomy.

[0135] In some aspects, the one or more implanted devices comprise one or more of: a prosthetic valve, a leaflet of a prosthetic valve, or a prosthetic clip.

[0136] In some aspects, the one or more anatomical landmarks comprise one or more geometric parameters.

[0137] In some aspects, the one or more geometric parameters comprise one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane, or an edge length associated with the one or more anatomical landmarks. In some aspects, the one or more geometric parameters further comprise parameters describing the spatial relationship of two or more of the one or more anatomical landmarks, such as a distance between two anatomical landmarks or an orientation of one anatomical landmark relative to another.

[0138] In some aspects, the one or more anatomical landmarks comprise one or more morphological features.

[0139] In some aspects, the one or more morphological features comprise one or more of: leaflet calcification, mobility, or one or more anomalies associated with the one or more anatomical landmarks.

[0140] In some aspects, block 1315 includes adding one or more primitive objects to the at least one shape representation of the patient heart anatomy.

[0141] In some aspects, the one or more primitive objects comprises one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape (such as a cylinder, a (truncated) cone, a prism, a cube, a sphere), or a NURBS surface.

[0142] In some aspects, a set of labels corresponding to labeling the one or more anatomical landmarks comprises one or more of: a primitive object, a virtual surface model, a contour, or a control point indicating the one or more anatomical landmarks.

[0143] In some aspects, method 1300 further includes selecting one or more preliminary treatment options from a plurality of preliminary treatment options.

[0144] In some aspects, method 1300 further includes providing an indication of the one or more preliminary treatment options as an additional input to the machine learning model, wherein the risk assessment is based on at least one risk score for each of the one or more preliminary treatment options.

[0145] In some aspects, the one or more preliminary treatment options comprise at least one of a valve repair treatment or a valve replacement treatment.

[0146] In some aspects, the one or more preliminary treatment options comprise one or more of a treatment type, a device type, a device position, and a device delivery.

[0147] In some aspects, selecting the one or more preliminary treatment options from the plurality of preliminary treatment options is based on an order of preference associated with the one or more preliminary treatment options.

[0148] In some aspects, the order of preference is based on an order of increasing complexity, risk of adverse effects, or invasiveness.

[0149] In some aspects, the risk assessment further comprises one or more additional risk scores associated with one or more potential complications for one or more cardiovascular treatments.

[0150] In some aspects, the risk assessment further comprises one or more correlations between one or more geometric parameters of the patient heart anatomy and the one or more potential complications.

[0151] In some aspects, the at least one shape representation comprises a valve of the patient heart anatomy.

[0152] In some aspects, the valve is one of a mitral valve, an aortic valve, a pulmonary valve or a tricuspid valve.

[0153] Note that FIG. 13 is just one example of a method, and other methods including fewer, additional, or alternative operations are possible consistent with this disclosure.

[0154] FIG. 14 depicts an example method 1400 for predicting adverse effects due to cardiac valve treatments. In one aspect, method 1400 can be implemented by the processing system 1500 of FIG. 15.

[0155] Method 1400 begins at block 1405 with obtaining at least one shape representation of a patient heart anatomy.

[0156] Method 1400 then proceeds to block 1410 with determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy.

[0157] Method 1400 then proceeds to block 1415 with generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy.

[0158] Method 1400 then proceeds to block 1420 with providing the set of labeled patient data and a set of other, historic, patient datasets as inputs to a first machine learning model.

[0159] Method 1400 then proceeds to block 1425 with receiving, from the first machine learning model: a subset of other patient datasets, of the set of other patient datasets, that comprises patient data that is similar to the set of labeled patient data, the subset of other patient datasets comprising: one or more first other patient datasets, of the subset of other patient datasets, that are associated with positive treatment outcome; and one or more second other patient datasets, of the subset of other patient datasets, that are associated with negative treatment outcome; and optionally a first set of factors associated with positive treatment outcome and a second set of factors associated with negative treatment outcome.

[0160] Method 1400 then proceeds to block 1430 with providing, to a second machine learning model, the subset of other patient datasets, optionally the first set of factors associated with positive treatment outcome, optionally the second set of factors associated with negative treatment outcome, and the set of labeled patient data.

[0161] Method 1400 then proceeds to block 1435 with receiving, from the second machine learning model, an output indicating a cohort predicted to have positive treatment outcome, the cohort comprising one or more third other patient datasets, of the subset of other patient datasets, sharing a set of similar patient characteristics, and optionally a risk assessment comprising one or more risk scores for one or more negative treatment outcomes.

[0162] In some aspects, the cohort is associated with a set of values for treatment parameters shared by all of the one or more third patient datasets.

[0163] In some aspects, the at least one shape representation comprises a valve of the patient heart anatomy.

[0164] In some aspects, the valve is one of a mitral valve, an aortic valve, a pulmonary valve or a tricuspid valve.

[0165] In some aspects, the at least one shape representation comprises one or more valve leaflets of the patient heart anatomy.

[0166] In some aspects, the one or more anatomical landmarks comprise one or more functional parameters.

[0167] In some aspects, the one or more functional parameters comprise one or more of: an ejection fraction or fractional shortening.

[0168] In some aspects, method 1400 further includes receiving, from the second machine learning model, a risk score, indicating a likelihood of an adverse effect.

[0169] In some aspects, method 1400 further includes receiving, from the second machine learning model, a proposed adjustment to a parameter of a cardiovascular treatment predicted to positively change the risk score.

[0170] In some aspects, method 1400 further includes receiving, from the second machine learning model, an expected treatment outcome for a patient corresponding to the patient heart anatomy.

[0171] In some aspects, the expected treatment outcome comprises one or more of: a prediction of patient survival years or one or more quality-of-life scores.

[0172] In some aspects, the patient heart anatomy is associated with a set of patient data and block 1405 includes generating the at least one shape representation based on the set of patient data.

[0173] In some aspects, the set of patient data further comprises patient history data.

[0174] In some aspects, the patient history data includes one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, or prior examination results of a patient associated with the patient heart anatomy.

[0175] In some aspects, the set of patient data further comprises one or more images of the patient heart anatomy, wherein each image depicts at least part of an anatomical landmark of the one or more anatomical landmarks.

[0176] In some aspects, the at least one shape representation comprises one or more of the one or more images of the patient heart anatomy.

[0177] In some aspects, the one or more images of the patient heart anatomy comprise one or more of: echocardiography images, MRI images, or CT images.

[0178] In some aspects, at least one image of the one or more images of the patient heart anatomy comprises an image taken with contrast.

[0179] In some aspects, the one or more images of the patient heart anatomy comprise one or more of: single-phase data or moving-image data.

[0180] In some aspects, generating the at least one shape representation comprises: generating a virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy, wherein the at least one shape representation comprises the virtual model of the patient heart anatomy.

[0181] In some aspects, generating the virtual model of the patient heart anatomy comprises: generating a plurality of preliminary virtual models, each virtual model based on one or more images, registering the plurality of preliminary virtual models so as to make them coincide in a common coordinate system, and combining features of the plurality of preliminary virtual models to generate the virtual model of the patient heart anatomy. In some aspects, the one or more images of the patient heart anatomy comprises images of a plurality of image modalities and each of the plurality of preliminary virtual models corresponds to one of the plurality of image modalities. In some aspects, the plurality of image modalities comprises CT and echocardiography. In some aspects, combining features of the plurality of preliminary virtual models associated with a plurality of image modalities comprises combining features of a plurality of feature types, and selecting as a source for each feature the specific preliminary virtual model of the plurality of preliminary virtual models associated with the specific image modality in which the feature’s feature type is (well / best) discernable. In some aspects, combining features of the plurality of virtual models comprises combining one or more of a blood pool volume, a valve annulus, a blood flow inlet or outlet, an ostium of a blood flow inletor outlet, or a vessel lumen sourced from a virtual model associated with CT data, with one or more valve leaflets sourced from a virtual model associated with echocardiography data.

[0182] In some aspects, generating the virtual model of the patient heart anatomy is based on using image segmentation on the one or more images of the patient heart anatomy.

[0183] In some aspects, generating the virtual model of the patient heart anatomy comprises: providing the one or more images of the patient heart anatomy to a machine learning model configured to generate virtual models; and receiving, from the machine learning model, the virtual model of the patient heart anatomy.

[0184] In some aspects, the virtual model of the patient heart anatomy is a three- dimensional model.

[0185] In some aspects, the virtual model of the patient heart anatomy is a fourdimensional model.

[0186] In some aspects, the one or more anatomical landmarks comprise one or more anatomical parts of a patient heart.

[0187] In some aspects, the one or more anatomical parts comprise one or more of: a valve, a valve leaflet, a papillary muscle head, a commissure, a valve annulus, a chamber, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen.

[0188] In some aspects, the one or more anatomical landmarks comprise one or more implanted devices present within the patient heart anatomy.

[0189] In some aspects, the one or more implanted devices comprise one or more of: a prosthetic valve, a leaflet of a prosthetic valve, or a prosthetic clip.

[0190] In some aspects, the one or more anatomical landmarks comprise one or more geometric parameters.

[0191] In some aspects, the one or more geometric parameters comprise one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane, or an edge length associated with the one or more anatomical landmarks. In some aspects, the one or more geometric parameters further comprise parameters describing the spatial relationship of two or more of the one or more anatomicallandmarks, such as a distance between two anatomical landmarks or an orientation of one anatomical landmark relative to another.

[0192] In some aspects, the one or more anatomical landmarks comprise one or more morphological features.

[0193] In some aspects, the one or more morphological features comprise one or more of: leaflet calcification, mobility, or one or more anomalies associated with the one or more anatomical landmarks.

[0194] In some aspects, block 1415 includes adding one or more primitive objects to the at least one shape representation of the patient heart anatomy.

[0195] In some aspects, the one or more primitive objects comprises one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape (such as a cylinder, a (truncated) cone, a prism, a cube, a sphere), or a NURBS surface.

[0196] In some aspects, a set of labels corresponding to labeling the one or more anatomical landmarks comprises one or more of: a primitive object, a virtual surface model, a contour, or a control point indicating the one or more anatomical landmarks.

[0197] In some aspects, the risk assessment further comprises one or more additional risk scores associated with one or more potential complications for one or more cardiovascular treatment options.

[0198] In some aspects, the risk assessment further comprises one or more correlations between one or more geometric parameters of the patient heart anatomy and the one or more potential complications.

[0199] Note that FIG. 14 is just one example of a method, and other methods including fewer, additional, or alternative operations are possible consistent with this disclosure.Example Processing System for Predicting Adverse Effects of Cardiac Valve Treatments

[0200] FIG. 15 depicts an example processing system 1500 configured to perform various aspects described herein, including, for example, method 1300 as described above with respect to FIG. 13 and / or method 1400 as described above with respect to FIG. 14.

[0201] Processing system 1500 is generally an example of an electronic device configured to execute computer-executable instructions, such as those derived from compiled computer code, including without limitation personal computers, tablet computers, servers, smart phones, smart devices, wearable devices, augmented and / or virtual reality devices, and others.

[0202] Processing system 1500 may be a single electronic device or a distributed system of a plurality of electronic devices communicatively connected to each other, e.g., via a cable, a wireless connection, a local network, an intranet, the Internet or any other form of electronic communication.

[0203] In the depicted example, processing system 1500 includes one or more processor(s) 1502, one or more input / output device(s) 1504, one or more display device(s) 1506, one or more network interface(s) 1508 through which processing system 1500 is connected to one or more network(s) (e.g., a local network, an intranet, the Internet, or any other group of processing systems communicatively connected to each other), and computer-readable medium 1512. In the depicted example, the aforementioned components are coupled by a bus 1510, which may generally be configured for data exchange amongst the components. Bus 1510 may be representative of multiple buses, while only one is depicted for simplicity.

[0204] Processor(s) 1502 are generally configured to retrieve and execute instructions stored in one or more memories, including local memories like computer-readable medium 1512, as well as remote memories and data stores. Similarly, processor(s) 1502 are configured to store application data residing in local memories like the computer- readable medium 1512, as well as remote memories and data stores. More generally, bus 1510 is configured to transmit programming instructions and application data among the processor(s) 1502, display device(s) 1506, network interface(s) 1508, and / or computer- readable medium 1512. In certain embodiments, processor(s) 1502 are representative of one or more central processing units (CPUs), graphics processing unit (GPUs), tensor processing unit (TPUs), accelerators, or other processing devices.

[0205] Input / output device(s) 1504 may include any device, mechanism, system, interactive display, and / or various other hardware and software components for communicating information between processing system 1500 and a user of processing system 1500. For example, input / output device(s) 1504 may include input hardware, suchas a keyboard, touch screen, buton, microphone, speaker, and / or other device for receiving inputs from the user and sending outputs to the user.

[0206] Display device(s) 1506 may generally include any sort of device configured to display data, information, graphics, user interface elements, and the like to a user. For example, display device(s) 1506 may include internal and external displays such as an internal display of a tablet computer or an external display for a server computer or a projector. Display device(s) 1506 may further include displays for devices, such as augmented, virtual, and / or extended reality devices, more particularly head-mounted devices. In various embodiments, display device(s) 1506 may be configured to display a graphical user interface.

[0207] Network interface(s) 1508 provide processing system 1500 with access to external networks and thereby to external processing systems. Network interface(s) 1508 can generally be any hardware and / or software capable of transmitting and / or receiving data via a wired or wireless network connection. Accordingly, network interface(s) 1508 can include a communication transceiver for sending and / or receiving any wired and / or wireless communication.

[0208] Computer-readable medium 1512 may be a volatile memory, such as a random-access memory (RAM), or a nonvolatile memory, such as nonvolatile randomaccess memory (NVRAM), or the like. In this example, computer-readable medium 1512 includes obtaining component 1514, determining component 1516, generating component 1518, providing component 1520, receiving component 1522, adding component 1524, and selecting component 1526. Processing of the components 1514- 1526 may enable and cause the processing system 1500 to perform the method 1300 described with respect to FIG. 13, or any aspect related to it, and / or the method 1400 described with respect to FIG. 14, or any aspect related to it.

[0209] In certain embodiments, obtaining component 1514 is configured to obtain at least one shape representation of a patient heart anatomy, as described in FIG. 13 with reference to block 1305. In certain embodiments, determining component 1516 is configured to determine one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy, as described in FIG. 13 with reference to block 1310. In certain embodiments, generating component 1518 is configured to generate a set of labeled patient data, at least in part, by labelingone or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy, as described in FIG. 13 with reference to block 1315. In certain embodiments, providing component 1520 is configured to provide the set of labeled patient data as inputs to a machine learning model, as described in FIG. 13 with reference to block 1320. In certain embodiments, receiving component 1522 is configured to receive, from the machine learning model, one or more risk scores representing predicted risk probabilities associated with performing one or more cardiac valve treatments on the patient heart anatomy, as described in FIG. 13 with reference to block 1325. In certain embodiments, generating component 1518 is configured to generate, for display, an output comprising a risk assessment associated with performing the one or more cardiac valve treatments based on the one or more risk scores, as described in FIG. 13 with reference to block 1330.

[0210] In certain embodiments, obtaining component 1514 is configured to obtain at least one shape representation of a patient heart anatomy, as described in FIG. 14 with reference to block 1405. In certain embodiments, determining component 1516 is configured to determine one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy, as described in FIG. 14 with reference to block 1410. In certain embodiments, generating component 1518 is configured to generate a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy, as described in FIG. 14 with reference to block 1415. In certain embodiments, providing component 1520 is configured to provide the set of labeled patient data and a set of other patient datasets as inputs to a first machine learning model, as described in FIG. 14 with reference to block 1420. In certain embodiments, receiving component 1522 is configured to receive, from the machine learning model: a subset of other patient datasets, of the set of other patient datasets, that comprises patient data that is similar to the set of labeled patient data, the subset of other patient datasets comprising: one or more first other patient datasets, of the subset of other patient datasets, that are associated with positive treatment outcome; and one or more second other patient datasets, of the subset of other patient datasets, that are associated with negative treatment outcome; and optionally a first set of factors associated with positive treatment outcome and a second set of factors associated with negative treatment outcome, as described in FIG. 14 with reference to block 1425.In certain embodiments, providing component 1520 is configured to provide, to a second machine learning model, the subset of other patient datasets, optionally the first set of factors associated with positive treatment outcome, optionally the second set of factors associated with negative treatment outcome, and the set of labeled patient data, as described in FIG. 14 with reference to block 1430. In certain embodiments, receiving component 1522 is configured to receive, from the second machine learning model, an output indicating a cohort predicted to have positive treatment outcome, the cohort comprising one or more third other patient datasets, of the subset of other patient datasets, sharing a set of similar patient characteristics, and a risk assessment comprising one or more risk scores for one or more negative treatment outcomes, as described in FIG. 14 with reference to block 1435.

[0211] Note that FIG. 15 is just one example of a processing system consistent with aspects described herein, and other processing systems having additional, alternative, or fewer components are possible consistent with this disclosure.Example Clauses

[0212] Implementation examples are described in the following numbered clauses:

[0213] Clause 1 : A method for predicting adverse effects due to cardiac valve treatments, comprising: obtaining, by a computing device, at least one shape representation of a patient heart anatomy; determining, by a user using the computing device or by the computing device, one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating, by the user using the computing device or by the computing device, a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy; providing, by the computing device, the set of labeled patient data as inputs to a machine learning model; receiving, by the computing device, from the machine learning model, one or more risk scores representing predicted risk probabilities associated with performing one or more cardiac valve treatments on the patient heart anatomy; and generating, by the computing device, for display, an output comprising a risk assessment associated with performing the one or more cardiac valve treatments based on the one or more risk scores.

[0214] Clause 2: The method of Clause 1, wherein: the patient heart anatomy is associated with a set of patient data and obtaining the at least one shape representation comprises generating the at least one shape representation based on the set of patient data.

[0215] Clause 3 : The method of Clause 2, wherein the set of patient data comprises patient history data.

[0216] Clause 4: The method of Clause 3, wherein the patient history data includes one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, or prior examination results of a patient associated with the patient heart anatomy.

[0217] Clause 5 : The method of Clause 3 or 4, wherein the set of patient data further comprises one or more images of the patient heart anatomy, wherein each image depicts at least part of an anatomical landmark of the one or more anatomical landmarks.

[0218] Clause 6: The method of Clause 5, wherein the at least one shape representation comprises one or more of the one or more images of the patient heart anatomy.

[0219] Clause 7 : The method of Clause 5 or 6, wherein the one or more images of the patient heart anatomy comprise one or more of: echocardiography images, MRI images, or CT images.

[0220] Clause 8: The method of any one of Clauses 5-7, wherein at least one image of the one or more images of the patient heart anatomy comprises an image taken with contrast.

[0221] Clause 9: The method of any one of Clauses 5-8, wherein the one or more images of the patient heart anatomy comprise one or more of: single-phase data or moving-image data.

[0222] Clause 10: The method of any one of Clauses 5-9, wherein generating the at least one shape representation comprises: generating a virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy, wherein the at least one shape representation comprises the virtual model of the patient heart anatomy.

[0223] Clause 11 : The method of clause 10, wherein generating the virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy comprises: generating a plurality of preliminary virtual models, each based on one ormore images; registering the plurality of preliminary virtual models so as to make them coincide in a common coordinate system; and combining features of the plurality of preliminary virtual models to generate the virtual model of the patient heart anatomy.

[0224] Clause 12: The method of Clause 11, wherein the one or more images of the patient heart anatomy comprises images of a plurality of image modalities and each of the plurality of preliminary virtual models corresponds to one of the plurality of image modalities.

[0225] Clause 13: The method of Clause 12, wherein the plurality of image modalities comprises CT and echocardiography.

[0226] Clause 14: The method of Clause 12 or 13, wherein combining features of the plurality of preliminary virtual models comprises combining features of a plurality of feature types, and selecting as source for each feature a preliminary virtual model of the plurality of preliminary virtual models corresponding to an image modality of the plurality of image modalities in which the feature’ s feature type is (well / best) discernable.

[0227] Clause 15: The method of any one of clauses 11-14, wherein combining features of the plurality of preliminary virtual models comprises combining one or more of a blood pool volume, a valve annulus, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen from a preliminary virtual model based on CT data with one or more valve leaflets from a preliminary virtual model based on echocardiography data.

[0228] Clause 16: The method of any one of Clauses 10-15, wherein generating the virtual model of the patient heart anatomy is based on using image segmentation on the one or more images of the patient heart anatomy.

[0229] Clause 17: The method of any one of Clauses 10-16, wherein generating the virtual model of the patient heart anatomy comprises: providing the one or more images of the patient heart anatomy to a machine learning model configured to generate virtual models; and receiving, from the machine learning model, the virtual model of the patient heart anatomy.

[0230] Clause 18 : The method of any one of Clauses 10-17, wherein the virtual model of the patient heart anatomy is a three-dimensional model.

[0231] Clause 19: The method of any one of Clauses 10-17, wherein the virtual model of the patient heart anatomy is a four-dimensional model.

[0232] Clause 20: The method of any one of Clauses 1-19, wherein the one or more anatomical landmarks comprise one or more anatomical parts of a patient heart.

[0233] Clause 21 : The method of Clause 20, wherein the one or more anatomical parts comprise one or more of: a valve, a valve leaflet, a papillary muscle head, a commissure, a valve annulus, a chamber, a blood flow inlet or outlet, an ostium of blood flow inlet or outlet, or a vessel lumen.

[0234] Clause 22: The method of any one of Clauses 1-21, wherein the one or more anatomical landmarks comprise one or more implanted devices present within the patient heart anatomy.

[0235] Clause 23: The method of Clause 22, wherein the one or more implanted devices comprise one or more of: a prosthetic valve, a leaflet of a prosthetic valve, or a prosthetic clip.

[0236] Clause 24: The method of any one of Clauses 1-23, wherein the one or more anatomical landmarks comprise one or more geometric parameters.

[0237] Clause 25: The method of Clause 24, wherein the one or more geometric parameters comprise one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane, or an edge length associated with the one or more anatomical landmarks.

[0238] Clause 26: The method of Clause 24 or 25, wherein the one or more geometric parameters further comprises parameters describing a spatial relationship between two or more of the one or more anatomical landmarks.

[0239] Clause 27 : The method of Clause 26, wherein the parameters describing the spatial relationship between two or more of the one or more anatomical landmarks comprise one or more of a distance between two anatomical landmarks or an orientation of one anatomical landmark relative to another anatomical landmark.

[0240] Clause 28: The method of any one of Clauses 1-27, wherein the one or more anatomical landmarks comprise one or more morphological features.

[0241] Clause 29: The method of Clause 28, wherein the one or more morphological features comprise one or more of: leaflet calcification, mobility, or one or more anomalies associated with the one or more anatomical landmarks.

[0242] Clause 30: The method of any one of Clauses 1-29, wherein generating the set of labeled patient data comprises adding one or more primitive objects to the at least one shape representation of the patient heart anatomy.

[0243] Clause 31 : The method of Clause 30, wherein the one or more primitive objects comprise one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

[0244] Clause 32: The method of Clause 31, wherein the primitive shape is one of a cylinder, a cone, a truncated cone, a sphere, a cube or a prism.

[0245] Clause 33: The method of any one of Clauses 1-32, wherein a set of labels corresponding to labeling the one or more anatomical landmarks comprises one or more of: a primitive object, a virtual surface model, a contour, or a control point indicating the one or more anatomical landmarks.

[0246] Clause 34: The method of Clause 33, wherein the primitive object comprises one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

[0247] Clause 35: The method of Clause 34, wherein the primitive shape is one of a cylinder, a cone, a truncated cone, a sphere, a cube or a prism.

[0248] Clause 36: The method of any one of Clauses 1-35, further comprising: selecting one or more preliminary treatment options from a plurality of preliminary treatment options; and providing an indication of the one or more selected preliminary treatment options as an additional input to the machine learning model, wherein the risk assessment is based on at least one risk score for each of the one or more selected preliminary treatment options.

[0249] Clause 37: The method of Clause 36, wherein the one or more selected preliminary treatment options comprise at least one of a valve repair treatment or a valve replacement treatment.

[0250] Clause 38: The method of Clause 36 or 37, wherein the one or more selected preliminary treatment options comprise one or more of a treatment type, a device type, a device position, and a device delivery.

[0251] Clause 39: The method of any one of Clauses 36-38, wherein selecting the one or more preliminary treatment options from the plurality of preliminary treatment options is based on an order of preference associated with the one or more preliminary treatment options.

[0252] Clause 40: The method of Clause 39, wherein the order of preference is based on an order of increasing complexity, risk of adverse effects, or invasiveness.

[0253] Clause 41 : The method of any one of Clauses 1-40, wherein the risk assessment further comprises one or more additional risk scores associated with one or more potential complications for one or more cardiovascular treatments.

[0254] Clause 42: The method of Clause 41, wherein the risk assessment further comprises one or more correlations between one or more geometric parameters of the patient heart anatomy and one or more or the one or more additional risk scores associated with the one or more potential complications.

[0255] Clause 43 : The method of any one of Clauses 1 -42, wherein the at least one shape representation comprises a valve of the patient heart anatomy.

[0256] Clause 44: The method of Clause 43, wherein the valve is one of a mitral valve, an aortic valve, a pulmonary valve or a tricuspid valve.

[0257] Clause 45: A method for predicting adverse effects due to cardiac valve treatments, comprising: obtaining at least one shape representation of a patient heart anatomy; determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy; providing the set of labeled patient data and a set of other patient datasets as inputs to a first machine learning model; receiving, from the first machine learning model: a subset of other patient datasets, of the set of other patient datasets, that comprises patient data that is similar to the set of labeled patient data, the subset of other patient datasets comprising: one or more first other patientdatasets, of the subset of other patient datasets, that are associated with positive treatment outcome; and one or more second other patient datasets, of the subset of other patient datasets, that are associated with negative treatment outcome; providing, to a second machine learning model, the subset of other patient datasets, and the set of labeled patient data; and receiving, from the second machine learning model, an output indicating a cohort predicted to have positive treatment outcome, the cohort comprising one or more third other patient datasets, of the subset of other patient datasets, sharing a set of similar patient characteristics.

[0258] Clause 46: The method of Clause 45, further comprising: receiving, from the first machine learning model: a first set of factors associated with positive treatment outcome and a second set of factors associated with negative treatment outcome.

[0259] Clause 47: The method of Clause 46, further comprising: providing, to the second machine learning model, the first set of factors associated with positive treatment outcome and the second set of factors associated with negative treatment outcome.

[0260] Clause 48: The method of any one of Clauses 45-47, further comprising: receiving, from the second machine learning model, a risk assessment comprising one or more risk scores for one or more negative treatment outcomes.

[0261] Clause 49: The method of Clause 48, wherein the risk assessment further comprises one or more additional risk scores associated with one or more potential complications for one or more cardiovascular treatments.

[0262] Clause 50: The method of Clause 49, wherein the risk assessment further comprises one or more correlations between one or more geometric parameters of the patient heart anatomy and one or more of the one or more additional risk scores associated with one or more potential complications.

[0263] Clause 51 : The method of any one of Clauses 45-50, wherein the at least one shape representation comprises a valve of the patient heart anatomy.

[0264] Clause 52: The method of Clause 51, wherein the valve is one of a mitral valve, an aortic valve, a pulmonary valve or a tricuspid valve.

[0265] Clause 53: The method of any one of Clauses 45-52, wherein the at least one shape representation comprises one or more valve leaflets of the patient heart anatomy.

[0266] Clause 54: The method of any one of Clauses 45-53, wherein the one or more anatomical landmarks comprise one or more functional parameters.

[0267] Clause 55: The method of Clause 54, wherein the one or more functional parameters comprise one or more of: an ejection fraction or fractional shortening.

[0268] Clause 56: The method of any one of Clauses 45-55, further comprising receiving, from the second machine learning model, a risk score, indicating a likelihood of an adverse effect.

[0269] Clause 57: The method of Clause 56, further comprising receiving, from the second machine learning model, a proposed adjustment to a parameter of the set of labeled patient data or of a cardiovascular treatment predicted to positively change the risk score.

[0270] Clause 58: The method of any one of Clauses 45-57, further comprising receiving, from the second machine learning model, an expected treatment outcome for a patient corresponding to the patient heart anatomy.

[0271] Clause 59: The method of Clause 58, wherein the expected treatment outcome comprises one or more of: a prediction of patient survival years or one or more quality- of-life scores.

[0272] Clause 60: The method of any one of Clauses 45-59, wherein: the patient heart anatomy is associated with a set of patient data and obtaining the at least one shape representation comprises generating the at least one shape representation based on the set of patient data.

[0273] Clause 61 : The method of Clause 60, wherein the set of patient data further comprises patient history data.

[0274] Clause 62: The method of Clause 61, wherein the patient history data includes one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, or prior examination results of a patient associated with the patient heart anatomy.

[0275] Clause 63 : The method of any one of Clauses 60-62, wherein the set of patient data further comprises one or more images of the patient heart anatomy, wherein each image depicts at least part of an anatomical landmark of the one or more anatomical landmarks.

[0276] Clause 64: The method of Clause 63, wherein the at least one shape representation comprises one or more of the one or more images of the patient heart anatomy.

[0277] Clause 65 : The method of Clause 63 or 64, wherein the one or more images of the patient heart anatomy comprise one or more of: echocardiography images, MRI images, or CT images.

[0278] Clause 66: The method of any one of Clauses 63-65, wherein at least one image of the one or more images of the patient heart anatomy comprises an image taken with contrast.

[0279] Clause 67: The method of any one of Clauses 63-66, wherein the one or more images of the patient heart anatomy comprise one or more of: single-phase data or moving-image data.

[0280] Clause 68: The method of any one of Clauses 63-67, wherein generating the at least one shape representation comprises: generating a virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy, wherein the at least one shape representation comprises the virtual model of the patient heart anatomy.

[0281] Clause 69: The method of clause 68, wherein generating the virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy comprises: generating a plurality of preliminary virtual models, each based on one or more images; registering the plurality of preliminary virtual models so as to make them coincide in a common coordinate system; and combining features of the plurality of preliminary virtual models to generate the virtual model of the patient heart anatomy.

[0282] Clause 70: The method of Clause 69, wherein the one or more images of the patient heart anatomy comprise images of a plurality of image modalities and each of the plurality of preliminary virtual models corresponds to one of the plurality of image modalities.

[0283] Clause 71 : The method of Clause 70, wherein the plurality of image modalities comprises CT and echocardiography.

[0284] Clause 72: The method of Clause 70 or 71, wherein combining features of the plurality of preliminary virtual models comprises combining features of a plurality of feature types, and selecting as source for each feature a preliminary virtual model of theplurality of preliminary virtual models corresponding to an image modality of the plurality of image modalities in which the feature’ s feature type is (well / best) discernable.

[0285] Clause 73: The method of any one of clauses 69-72, wherein combining features of the plurality of preliminary virtual models comprises combining one or more of a blood pool volume, a valve annulus, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen from a preliminary virtual model based on CT data with one or more valve leaflets from a preliminary virtual model based on echocardiography data.

[0286] Clause 74: The method of any one of Clauses 68-73, wherein generating the virtual model of the patient heart anatomy is based on using image segmentation on the one or more images of the patient heart anatomy.

[0287] Clause 75: The method of any one of Clauses 68-74, wherein generating the virtual model of the patient heart anatomy comprises: providing the one or more images of the patient heart anatomy to a machine learning model configured to generate virtual models; and receiving, from the machine learning model, the virtual model of the patient heart anatomy.

[0288] Clause 76: The method of any one of Clauses 68-75, wherein the virtual model of the patient heart anatomy is a three-dimensional model.

[0289] Clause 77 : The method of any one of Clauses 68-75, wherein the virtual model of the patient heart anatomy is a four-dimensional model.

[0290] Clause 78: The method of any one of Clauses 45-77, wherein the one or more anatomical landmarks comprise one or more anatomical parts of a patient heart.

[0291] Clause 79: The method of Clause 78, wherein the one or more anatomical parts comprise one or more of: a valve, a valve leaflet, a papillary muscle head, a commissure, a valve annulus, a chamber, a blood flow inlet or outlet, an ostium of blood flow inlet or outlet, or a vessel lumen.

[0292] Clause 80: The method of any one of Clauses 45-79, wherein the one or more anatomical landmarks comprise one or more implanted devices present within the patient heart anatomy.

[0293] Clause 81 : The method of Clause 80, wherein the one or more implanted devices comprise one or more of: a prosthetic valve, a leaflet of a prosthetic valve, or a prosthetic clip.

[0294] Clause 82: The method of any one of Clauses 45-81, wherein the one or more anatomical landmarks comprise one or more geometric parameters.

[0295] Clause 83: The method of Clause 82, wherein the one or more geometric parameters comprise one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane or an edge length associated with the one or more anatomical landmarks.

[0296] Clause 84: The method of Clause 82 or 83, wherein the one or more geometric parameters further comprise parameters describing a spatial relationship between two or more of the one or more anatomical landmarks.

[0297] Clause 85: The method of Clause 84, wherein the parameters describing the spatial relationship between two or more of the one or more anatomical landmarks comprise one or more of a distance between two anatomical landmarks or an orientation of one anatomical landmark relative to another.

[0298] Clause 86: The method of any one of Clauses 45-85, wherein the one or more anatomical landmarks comprise one or more morphological features.

[0299] Clause 87: The method of Clause 86, wherein the one or more morphological features comprise one or more of: leaflet calcification, mobility, or one or more anomalies associated with the one or more anatomical landmarks.

[0300] Clause 88: The method of any one of Clauses 45-87, wherein generating the set of labeled patient data comprises adding one or more primitive objects to the at least one shape representation of the patient heart anatomy.

[0301] Clause 89: The method of Clause 88, wherein the one or more primitive objects comprise one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

[0302] Clause 90: The method of Clause 89, wherein the primitive shape is one of a cylinder, a cone, a truncated cone, a sphere, a cube or a prism.

[0303] Clause 91 : The method of any one of Clauses 45-90, wherein a set of labels corresponding to labeling the one or more anatomical landmarks comprises one or more of: a primitive object, a virtual surface model, a contour, or a control point indicating the one or more anatomical landmarks.

[0304] Clause 92: The method of Clause 91, wherein the primitive object comprises one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

[0305] Clause 93 : The method of Clause 92, wherein the primitive shape is one of a cylinder, a cone, a truncated cone, a sphere, a cube or a prism.

[0306] Clause 94: A processing system, comprising: memory comprising computerexecutable instructions; and one or more processors configured to execute the computerexecutable instructions and cause the processing system to perform a method in accordance with any one of Clauses 1-93.

[0307] Clause 95 : A processing system, comprising means for performing a method in accordance with any one of Clauses 1-93.

[0308] Clause 96: A non-transitory computer-readable medium storing program code for causing a processing system to perform the steps of any one of Clauses 1-93.

[0309] Clause 97: A computer program product embodied on a computer-readable storage medium comprising code for performing a method in accordance with any one of Clauses 1-93.Additional Considerations

[0310] The preceding description is provided to enable any person skilled in the art to practice the various embodiments described herein. The examples discussed herein are not limiting of the scope, applicability, or embodiments set forth in the claims. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments. For example, changes may be made in the function and arrangement of elements discussed without departing from the scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as appropriate. For instance, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Also, features described with respectto some examples may be combined in some other examples. For example, an apparatus may be implemented, or a method may be practiced using any number of the aspects set forth herein. In addition, the scope of the disclosure is intended to cover such an apparatus or method that is practiced using other structure, functionality, structure, and functionality in addition to, or other than, the various aspects of the disclosure set forth herein. It should be understood that any aspect of the disclosure disclosed herein may be embodied by one or more elements of a claim.

[0311] As used herein, a phrase referring to “at least one of’ a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover a, b, c, a-b, a-c, b-c, and a-b-c, as well as any combination with multiples of the same element (e.g., a-a, a-a-a, a-a-b, a-a-c, a-b-b, a-c-c, b-b, b-b-b, b-b-c, c-c, and c-c-c or any other ordering of a, b, and c).

[0312] As used herein, the term “determining” encompasses a wide variety of actions. For example, “determining” may include calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, a database, or another data structure), ascertaining and the like. Also, “determining” may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory) and the like. Also, “determining” may include resolving, selecting, choosing, establishing and the like.

[0313] The methods disclosed herein comprise one or more steps or actions for achieving the methods. The method steps and / or actions may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims. Further, the various operations of methods described above may be performed by any suitable means capable of performing the corresponding functions. The means may include various hardware and / or software component(s) and / or module(s), including, but not limited to a circuit, an application specific integrated circuit (ASIC), or processor. Generally, where there are operations illustrated in figures, those operations may have corresponding counterpart means-plus-function components with similar numbering.

[0314] Unless explicitly mentioned, all method steps or actions for achieving the methods disclosed herein may be performed automatically by a computing system, semi- automatically or manually by a user using a computing system. Unless explicitlymentioned, the word “user” as used herein may refer to a medical professional or a nonmedical professional, such as an engineer or a trained technician.

[0315] The following claims are not intended to be limited to the embodiments shown herein but are to be accorded the full scope consistent with the language of the claims. Within a claim, reference to an element in the singular is not intended to mean “one and only one” unless specifically so stated, but rather “one or more.” Unless specifically stated otherwise, the term “some” refers to one or more. No claim element is to be construed under the provisions of 35 U.S.C. § 112(f) unless the element is expressly recited using the phrase “means for” or, in the case of a method claim, the element is recited using the phrase “step for.” All structural and functional equivalents to the elements of the various aspects described throughout this disclosure that are known or later come to be known to those of ordinary skill in the art are expressly incorporated herein by reference and are intended to be encompassed by the claims. Moreover, nothing disclosed herein is intended to be dedicated to the public regardless of whether such disclosure is explicitly recited in the claims.

Claims

CLAIMS1. A method for predicting adverse effects due to cardiac valve treatments, comprising: obtaining at least one shape representation of a patient heart anatomy; determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy; providing the set of labeled patient data as inputs to a machine learning model; receiving, from the machine learning model, one or more risk scores representing predicted risk probabilities associated with performing one or more cardiac valve treatments on the patient heart anatomy; and generating, for display, an output comprising a risk assessment associated with performing the one or more cardiac valve treatments based on the one or more risk scores.

2. The method of Claim 1, wherein: the patient heart anatomy is associated with a set of patient data; and obtaining the at least one shape representation comprises generating the at least one shape representation based on the set of patient data.

3. The method of Claim 2, wherein the set of patient data comprises patient history data.

4. The method of Claim 3, wherein the patient history data includes one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, or prior examination results of a patient associated with the patient heart anatomy.

5. The method of Claim 3, wherein the set of patient data further comprises one or more images of the patient heart anatomy, wherein each image depicts at least part of an anatomical landmark of the one or more anatomical landmarks.

6. The method of Claim 5, wherein the at least one shape representation comprises one or more of the one or more images of the patient heart anatomy.

7. The method of Claim 5, wherein the one or more images of the patient heart anatomy comprise one or more of: echocardiography images, magnetic resonance imaging (MRI) images, or computed tomography (CT) images.

8. The method of Claim 5, wherein at least one image of the one or more images of the patient heart anatomy comprises an image taken with contrast.

9. The method of Claim 5, wherein the one or more images of the patient heart anatomy comprise one or more of: single-phase data or moving- image data.

10. The method of Claim 5, wherein generating the at least one shape representation comprises: generating a virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy, wherein the at least one shape representation comprises the virtual model of the patient heart anatomy.

11. The method of Claim 10, wherein generating the virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy comprises: generating a plurality of preliminary virtual models, wherein each preliminary virtual model is based on one or more images of the patient heart anatomy; registering the plurality of preliminary virtual models in a common coordinate system; and combining one or more features of the plurality of preliminary virtual models to generate the virtual model of the patient heart anatomy.

12. The method of Claim 11, wherein: the one or more images of the patient heart anatomy comprise one or more images of a plurality of image modalities; and each of the plurality of preliminary virtual models corresponds to one of the plurality of image modalities.

13. The method of Claim 12, wherein the plurality of image modalities comprises CT scans and echocardiography images.

14. The method of Claim 12, wherein combining features of the plurality of preliminary virtual models comprises: combining one or more features of a plurality of feature types; and selecting, as a source for each respective feature, a preliminary virtual model of the plurality of preliminary virtual models corresponding to an image modality of the plurality of image modalities in which a feature type of the respective feature is discernable.

15. The method of Claim 11, wherein combining the one or more features of the plurality of preliminary virtual models comprises combining one or more of a blood pool volume, a valve annulus, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen from a preliminary virtual model based on CT data with one or more valve leaflets from a preliminary virtual model based on echocardiography data.

16. The method of Claim 10, wherein generating the virtual model of the patient heart anatomy is based on using image segmentation on the one or more images of the patient heart anatomy.

17. The method of Claim 10, wherein generating the virtual model of the patient heart anatomy comprises: providing the one or more images of the patient heart anatomy to a machine learning model configured to generate virtual models; and receiving, from the machine learning model, the virtual model of the patient heart anatomy.

18. The method of Claim 10, wherein the virtual model of the patient heart anatomy is a three-dimensional model.

19. The method of Claim 10, wherein the virtual model of the patient heart anatomy is a four-dimensional model.

20. The method of Claim 1, wherein the one or more anatomical landmarks comprise one or more anatomical parts of a patient heart.

21. The method of Claim 20, wherein the one or more anatomical parts comprise one or more of: a valve, a valve leaflet, a papillary muscle head, a commissure, a valveannulus, a chamber, a blood flow inlet or outlet, an ostium of blood flow inlet or outlet, or a vessel lumen.

22. The method of Claim 1, wherein the one or more anatomical landmarks comprise one or more implanted devices present within the patient heart anatomy.

23. The method of Claim 22, wherein the one or more implanted devices comprise one or more of: a prosthetic valve, a leaflet of a prosthetic valve, or a prosthetic clip.

24. The method of Claim 1, wherein the one or more anatomical landmarks comprise one or more geometric parameters.

25. The method of Claim 24, wherein the one or more geometric parameters comprise one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane, or an edge length associated with the one or more anatomical landmarks.

26. The method of Claim 24, wherein the one or more geometric parameters further comprises one or more parameters describing a spatial relationship between two or more of the one or more anatomical landmarks.

27. The method of Claim 26, wherein the one or more parameters describing the spatial relationship between two or more of the one or more anatomical landmarks comprise a distance between two anatomical landmarks and / or an orientation of at least one anatomical landmark relative to another anatomical landmark.

28. The method of Claim 1, wherein the one or more anatomical landmarks comprise one or more morphological features.

29. The method of Claim 28, wherein the one or more morphological features comprise one or more of: leaflet calcification, mobility, or one or more anomalies associated with the one or more anatomical landmarks.

30. The method of Claim 1, wherein generating the set of labeled patient data comprises adding one or more primitive objects to the at least one shape representation of the patient heart anatomy.

31. The method of Claim 30, wherein the one or more primitive objects comprise one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

32. The method of Claim 31, wherein the primitive shape is one of: a cylinder, a cone, a truncated cone, a sphere, a cube, or a prism.

33. The method of Claim 1, wherein a set of labels corresponding to labeling the one or more anatomical landmarks comprises one or more of: a primitive object, a virtual surface model, a contour, or a control point indicating the one or more anatomical landmarks.

34. The method of Claim 33, wherein at least one primitive object comprises one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

35. The method of Claim 34, wherein the primitive shape is one of: a cylinder, a cone, a truncated cone, a sphere, a cube, or a prism.

36. The method of Claim 1, further comprising: selecting one or more preliminary treatment options from a plurality of preliminary treatment options; and providing an indication of the one or more selected preliminary treatment options as an additional input to the machine learning model, wherein the risk assessment is based on at least one risk score for each of the one or more selected preliminary treatment options.

37. The method of Claim 36, wherein the one or more selected preliminary treatment options comprise at least one of a valve repair treatment or a valve replacement treatment.

38. The method of Claim 36, wherein the one or more selected preliminary treatment options comprise one or more of: a treatment type, a device type, a device position, and a device delivery.

39. The method of Claim 36, wherein selecting the one or more selected preliminary treatment options from the plurality of preliminary treatment options is based on an order of preference associated with the one or more selected preliminary treatment options.

40. The method of Claim 39, wherein the order of preference is based on an order of increasing complexity, risk of adverse effects, or invasiveness.

41. The method of Claim 1, wherein the risk assessment further comprises one or more additional risk scores associated with one or more potential complications for one or more cardiovascular treatments.

42. The method of Claim 41, wherein the risk assessment further comprises one or more correlations between one or more geometric parameters of the patient heart anatomy and one or more or the one or more additional risk scores associated with the one or more potential complications.

43. The method of Claim 1, wherein the at least one shape representation comprises a valve of the patient heart anatomy.

44. The method of Claim 43, wherein the valve is one of a mitral valve, an aortic valve, a pulmonary valve or a tricuspid valve.

45. A method for predicting adverse effects due to cardiac valve treatments, comprising: obtaining at least one shape representation of a patient heart anatomy; determining one or more features of the patient heart anatomy based on analyzing the at least one shape representation of the patient heart anatomy; generating a set of labeled patient data, at least in part, by labeling one or more anatomical landmarks of the patient heart anatomy in the at least one shape representation based on the one or more features of the patient heart anatomy; providing the set of labeled patient data and a set of other patient datasets as inputs to a first machine learning model; receiving, from the first machine learning model: a subset of other patient datasets, of the set of other patient datasets, which comprises patient data that is similar to the set of labeled patient data, the subset of other patient datasets comprising: one or more first other patient datasets, of the subset of other patient datasets, which are associated with positive treatment outcome; and one or more second other patient datasets, of the subset of other patient datasets, which are associated with negative treatment outcome;providing, to a second machine learning model, the subset of other patient datasets and the set of labeled patient data; and receiving, from the second machine learning model, an output indicating a cohort predicted to have positive treatment outcome, the cohort comprising one or more third other patient datasets, of the subset of other patient datasets, sharing a set of similar patient characteristics.

46. The method of Claim 45, further comprising receiving, from the first machine learning model, a first set of factors associated with a positive treatment outcome and a second set of factors associated with a negative treatment outcome.

47. The method of Claim 46, further comprising providing, to the second machine learning model, the first set of factors associated with the positive treatment outcome and the second set of factors associated with the negative treatment outcome.

48. The method of Claim 45, further comprising receiving, from the second machine learning model, a risk assessment comprising one or more risk scores for one or more negative treatment outcomes.

49. The method of Claim 48, wherein the risk assessment further comprises one or more additional risk scores associated with one or more potential complications for one or more cardiovascular treatments.

50. The method of Claim 49, wherein the risk assessment further comprises one or more correlations between one or more geometric parameters of the patient heart anatomy and one or more of the one or more additional risk scores associated with one or more potential complications.

51. The method of Claim 45, wherein the at least one shape representation comprises a valve of the patient heart anatomy.

52. The method of Claim 51, wherein the valve is one of a mitral valve, an aortic valve, a pulmonary valve or a tricuspid valve.

53. The method of Claim 45, wherein the at least one shape representation comprises one or more valve leaflets of the patient heart anatomy.

54. The method of Claim 45, wherein the one or more anatomical landmarks comprise one or more functional parameters.

55. The method of Claim 54, wherein the one or more functional parameters comprise one or more of: an ejection fraction or fractional shortening.

56. The method of Claim 45, further comprising receiving, from the second machine learning model, a risk score, indicating a likelihood of an adverse effect.

57. The method of Claim 56, further comprising receiving, from the second machine learning model, a proposed adjustment to a parameter of the set of labeled patient data or of a cardiovascular treatment predicted to positively change the risk score.

58. The method of Claim 45, further comprising receiving, from the second machine learning model, an expected treatment outcome for a patient corresponding to the patient heart anatomy.

59. The method of Claim 58, wherein the expected treatment outcome comprises one or more of: a prediction of patient survival years or one or more quality-of-life scores.

60. The method of Claim 45, wherein: the patient heart anatomy is associated with a set of patient data, and obtaining the at least one shape representation comprises generating the at least one shape representation based on the set of patient data.

61. The method of Claim 60, wherein the set of patient data further comprises patient history data.

62. The method of Claim 61 , wherein the patient history data includes one or more of: reported symptoms, previous pathology, age, sex, height, weight, lifestyle attributes, hereditary disorders, or prior examination results of a patient associated with the patient heart anatomy.

63. The method of Claim 60, wherein the set of patient data further comprises one or more images of the patient heart anatomy, wherein each image depicts at least part of an anatomical landmark of the one or more anatomical landmarks.

64. The method of Claim 63, wherein the at least one shape representation comprises one or more of the one or more images of the patient heart anatomy.

65. The method of Claim 63, wherein the one or more images of the patient heart anatomy comprise one or more of: echocardiography images, magnetic resonance imaging (MRI) images, or computed tomography (CT) images.

66. The method of Claim 63, wherein at least one image of the one or more images of the patient heart anatomy comprises an image taken with contrast.

67. The method of Claim 63, wherein the one or more images of the patient heart anatomy comprise one or more of: single-phase data or moving- image data.

68. The method of Claim 63, wherein generating the at least one shape representation comprises: generating a virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy, wherein the at least one shape representation comprises the virtual model of the patient heart anatomy.

69. The method of Claim 68, wherein generating the virtual model of the patient heart anatomy based on the one or more images of the patient heart anatomy comprises: generating a plurality of preliminary virtual models, wherein each preliminary virtual model is based on one or more images of the patient heart anatomy; registering the plurality of preliminary virtual models in a common coordinate system; and combining one or more features of the plurality of preliminary virtual models to generate the virtual model of the patient heart anatomy.

70. The method of Claim 69, wherein: the one or more images of the patient heart anatomy comprise one or more images of a plurality of image modalities; and each of the plurality of preliminary virtual models corresponds to one of the plurality of image modalities.

71. The method of Claim 70, wherein the plurality of image modalities comprises CT scans and echocardiography images.

72. The method of Claim 70, wherein combining features of the plurality of preliminary virtual models comprises: combining one or more features of a plurality of feature types; andselecting, as a source for each respective feature, a preliminary virtual model of the plurality of preliminary virtual models corresponding to an image modality of the plurality of image modalities in which a feature type of a respective feature is discernable.

73. The method of Claim 69, wherein combining the one or more features of the plurality of preliminary virtual models comprises combining one or more of a blood pool volume, a valve annulus, a blood flow inlet or outlet, an ostium of a blood flow inlet or outlet, or a vessel lumen from a preliminary virtual model based on CT data with one or more valve leaflets from a preliminary virtual model based on echocardiography data.

74. The method of Claim 68, wherein generating the virtual model of the patient heart anatomy is based on using image segmentation on the one or more images of the patient heart anatomy.

75. The method of Claim 68, wherein generating the virtual model of the patient heart anatomy comprises: providing the one or more images of the patient heart anatomy to a machine learning model configured to generate virtual models; and receiving, from the machine learning model, the virtual model of the patient heart anatomy.

76. The method of Claim 68, wherein the virtual model of the patient heart anatomy is a three-dimensional model.

77. The method of Claim 68, wherein the virtual model of the patient heart anatomy is a four-dimensional model.

78. The method of Claim 45, wherein the one or more anatomical landmarks comprise one or more anatomical parts of a patient heart.

79. The method of Claim 78, wherein the one or more anatomical parts comprise one or more of: a valve, a valve leaflet, a papillary muscle head, a commissure, a valve annulus, a chamber, a blood flow inlet or outlet, an ostium of blood flow inlet or outlet, or a vessel lumen.

80. The method of Claim 45, wherein the one or more anatomical landmarks comprise one or more implanted devices present within the patient heart anatomy.

81. The method of Claim 80, wherein the one or more implanted devices comprise one or more of: a prosthetic valve, a leaflet of a prosthetic valve, or a prosthetic clip.

82. The method of Claim 45, wherein the one or more anatomical landmarks comprise one or more geometric parameters.

83. The method of Claim 82, wherein the one or more geometric parameters comprise one or more of: location, shape, size, thickness, curvature, area, circumference, geometric center point, best-fit plane, or an edge length associated with the one or more anatomical landmarks.

84. The method of Claim 82, wherein the one or more geometric parameters further comprises one or more parameters describing a spatial relationship between two or more of the one or more anatomical landmarks.

85. The method of Claim 84, wherein the one or more parameters describing the spatial relationship between two or more of the one or more anatomical landmarks comprise a distance between two anatomical landmarks and / or an orientation of at least one anatomical landmark relative to another anatomical landmark.

86. The method of Claim 45, wherein the one or more anatomical landmarks comprise one or more morphological features.

87. The method of Claim 86, wherein the one or more morphological features comprise one or more of: leaflet calcification, mobility, or one or more anomalies associated with the one or more anatomical landmarks.

88. The method of Claim 45, wherein generating the set of labeled patient data comprises adding one or more primitive objects to the at least one shape representation of the patient heart anatomy.

89. The method of Claim 88, wherein the one or more primitive objects comprise one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

90. The method of Claim 89, wherein the primitive shape is one of a cylinder, a cone, a truncated cone, a sphere, a cube or a prism.

91. The method of Claim 45, wherein a set of labels corresponding to labeling the one or more anatomical landmarks comprises one or more of: a primitive object, a virtualsurface model, a contour, or a control point indicating the one or more anatomical landmarks.

92. The method of Claim 91, wherein the primitive object comprises one or more of: a control point, a line, a line segment, a polyline, a spline, a polygon, a plane, a primitive shape, or a NURBS surface.

93. The method of Claim 92, wherein the primitive shape is one of: a cylinder, a cone, a truncated cone, a sphere, a cube, or a prism.

Images