Deep learning for intracranial pressure estimation

Abstract

The presently claimed and described technology provides methods for a deep learning approach for intracranial pressure estimation using non-cranial waveform measurements.

Description

DEEP LEARNING FOR INTRACRANIAL PRESSURE ESTIMATIONRELATED APPLICATIONS

[0001] The present patent application claims the priority benefit of U.S. Provisional Patent Application Ser. No. 63 / 626,051, filed January 28, 2024, the content of which is hereby incorporated by reference in its entirety into this disclosure.ACKNOWLEDGEMENT OF GOVERNMENT SUPPORT

[0002] This invention was made with government support under Grant No. DK107908 awarded by The National Institutes of Health. The government has certain rights in the invention.BACKGROUND

[0003] Elevation in intracranial pressure (ICP) is common in severe acute brain injuries (SABI), such as stroke and traumatic brain injuries, contributing to secondary neurological damage. (Le Roux ct al. 2014) (Fernando ct al. 2019; Hawryluk ct al. 2022). The current gold standard for ICP monitoring is an invasive monitor, which carries risks of infection and hemorrhage, limiting its use (Baggiani et al. 2023).

[0004] Non-invasive ICP estimation methods, like transcranial doppler (TCD) and optic nerve sheath diameter (ONSD), (Muller et al. 2023) (Dubourg et al. 2011; Fernando et al. 2019) show promise with high area under the receiver operating characteristic curve (AUROC) values as high as 0.92 and 0.94 respectively for detection of intracranial hypertension. However, their utility is constrained by limited availability of specialized skills and equipment settings (Lau and Amtfield 2017; W. Chen et al. 2023) and the requirement of high clinical suspicion to administer these tests, possibly overlooking subclinical ICP abnormalities. Additionally, there is substantial variability in accuracy and clinical relevance for ICP monitoring using these technologies. (Chesnut et al. 2012; Nattino ct al. 2023; Robba ct al. 2021).

[0005] Recognizing these limitations, recent research seeks to correlate physiological data with neurological conditions. (Badnjevic, Skrbic, and Pokvic 2019; Nair et al. 2023). Yet, such studies are often restricted by their sample size, stringent data filters limiting real world application, andlack of external validation limiting generalizability. (Brasil et al. 2021 ; Megjhani et al. 2023;Lazaridis et al. 2022; Nair et al. 2023).BRIEF SUMMARY

[0006] A system of one or more computers can be configured to perform particular operations or actions by virtue of having software, firmware, hardware, or a combination of them installed on the system that in operation causes or cause the system to perform the actions. One or more computer programs can be configured to perform particular operations or actions by virtue of including instructions that, when executed by data processing apparatus, cause the apparatus to perform the actions.

[0007] One general aspect includes a computer-implemented method of predicting or detecting increased intracranial pressure in a subject. The computer - implemented method also includes establishing, by using an alCP model, an alCP prediction model based on at least a plurality of training data sets that are related to a plurality of subjects, each of the plurality of training data sets including at least one extracranial waveform. The computer - implemented method also includes receiving, via one or more processors, a first data set for the subject. The computer - implemented method also includes determining, via one or more processors, an alCP value corresponding to the subject by analyzing the first data set using the alCP prediction model. The computer - implemented method also includes comparing, via one or more processors, the alCP value to a first threshold value. The computer - implemented method also includes where the alCP value is equal to or greater than the first threshold value, generating a determination of alCP. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0008] One general aspect includes a computer-implemented method of predicting or detecting increased intracranial pressure in a subject. The computer - implemented method also includes receiving, via one or more processors, a first data set for the subject. The computer - implemented method also includes determining, via one or more processors, an alCP value corresponding to the subject by analyzing the first data set using a trained machine learning model. The computer - implemented method also includes comparing, via one or more processors,the alCP value to a first threshold value. The computer - implemented method also includes where the alCP value is equal to or greater than the first threshold value, generating a determination of alCP. In an aspect, the trained machine learning model has been trained using a plurality of training data sets that are related to a plurality of subjects, each of the plurality of training data sets including at least one extracranial waveform. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0009] One general aspect includes a non-transitory computer readable medium having instructions stored thereon, wherein execution of the instructions by a processor causes the processor to receive a first data set for the subject; determine an ICP value corresponding to the subject by analyzing the first data set for the subject using a trained machine learning model; and generate a determination of ICP by comparing the ICP value to a first threshold value.

[0010] In an aspect, execution of the instructions by a processor further causes the processor to receive a second data set for the subject; determine a risk score by analyzing the second data set for the subject; and generate a prediction of frequency of dysregulation in ICP by comparing the risk score to a second threshold value.

[0011] In an aspect, execution of the instructions by a processor further causes the processor to: determine a classification by comparing the risk score to a third threshold value.

[0012] In an aspect, the first threshold value is at least about 15 mm of Hg. In an aspect, the first data set comprises extracranial waveform data. In an aspect, the extracranial waveform data comprises arterial blood pressure (ABG) waveforms, electrocardiogram (EKG) leads II and / or V waveforms, respiratory waveforms, plethysmography waveforms, SpCh waveforms, and / or Pho toplethy smogram (PPG) waveforms.

[0013] In an aspect, the methods further include receiving, via one or more processors, a second data set for the subject; determining, via the one or more processors, a risk score by analyzing the second data set for the subject; comparing, via the one or more processors, the risk score to a second threshold value; and when the risk score meets the second threshold value, generating a prediction of frequency of dysregulation in ICP. In an aspect, the risk prediction model is a slicelevel prediction model

[0014] In an aspect, the second data set is generated from the first data set.

[0015] In an aspect, the methods further include determining, via one or more processors, a classification by comparing the risk score to a third threshold value.

[0016] In an aspect, (i) when the risk score is lower than the third threshold value, the subject is classified as low risk; or (ii) when the risk score is equal to or greater than the third threshold value, the subject is classified as high risk

[0017] In an aspect, the methods further include evaluating the association of aCIP using digital clinical data for the subject.

[0018] In an aspect, the methods are used to (i) provide intracranial monitoring to subjects that have contraindications to invasive monitoring; (ii) screen subjects for intracranial hypertension; (iii) to evaluate the significance of association of alCP with clinical phenotypes; and / or (iv) detect intracranial pressure abnormalities.

[0019] In an aspect, the subject is a human. In an aspect, the subject is an adult.

[0020] These and other advantages, aspects, and novel features of the present disclosure, as well as details of illustrated embodiments thereof, will be more fully understood from the following description and drawings.BRIEF DESCRIPTION OF THE FIGURES

[0021] Embodiments of the present disclosure will now be described, by way of example only, with reference to the attached Figures, wherein:

[0022] FIGs. 1A and IB depict preprocessing of datasets and development of the alCP model. FIG. 1A is a Schema for AlCP (i) Initial dataset (ii) Filtering by monitoring modality (iii) Filtering waveforms, and (iv) Final dataset. FIG. IB is a depiction of Model Architecture, Training and Output.

[0023] FIG. 2 a categorization of phenotypes for demographic analysis.

[0024] FIGs. 3A and 3B show the performance of the alCP model. FIG. 3A is an AUROC curve on the internal (MIMIC-ICP) and external (MSH-ICP) test cohorts. FIG.3B is an AUPRC curve for internal (MIMIC-ICP) and external (MSH-ICP) test cohorts.

[0025] FIGs. 4A and 4B are visualization of waveforms and alCP predictions of two different waveforms from two different patients with different alCP. FIG. 4A shows a low predictedintracranial hypertension risk (alCP < 0.50). FIG. 4B shows an elevated risk of intracranial hypertension (alCP > 0.50).

[0026] FIGs. 5A - 5C are visualization of alCP predictions and actual ICP measurements over 30 minutes. Three different waveforms from three different patients with different intracranial waveforms over approximately 30 minutes after smoothing and filtering. FIG. 5A: Patient with low levels of variability. FIG. 5B: Patient with high levels of variability. FIG. 5C: Patient with intermediate levels of variability.DETAILED DESCRIPTIONI. Introduction

[0027] Increased intracranial pressure (ICP) >15mmHg is associated with adverse neurological outcomes, but needs invasive intracranial monitoring. Described herein is a artificial intelligence- derived biomarker for elevated ICP (alCP) for adult patients. (FIG. 1A and IB.) The model utilizes physiologic extracranial waveforms that are routinely collected in intensive care to generate a second-by-second prediction of whether intracranial pressure is present.

[0028] Described herein is the largest model (10 million parameters) to-date for intracranial pressure, trained on over 270 hours of waveform data from a publicly available dataset. The model utilizes physiologic extracranial waveforms that are routinely collected in intensive care to generate a second-by-second prediction of whether intracranial pressure is elevated. 258 h of waveform data from the Mount Sinai Hospital was used to externally validate this approach.

[0029] There has also been growing interest in the role of neuromonitoring across a wide range of different phenotypes including pregnancy, liver disease, kidney disease, sepsis, myocardial infarction, and acute respiratory distress syndrome. Disclosed herein are associations of the model’s predictions with clinical phenotypes both a priori and via unbiased phenome-wide scans.II. Definitions

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the methods described herein belong. Any reference to standard methods refers to the most recent available version of the method at the time of filing of this disclosure unless otherwise indicated.

[0031] For any method disclosed herein that includes discrete steps, the steps may be conducted in any feasible order. And, as appropriate, any combination of two or more steps may be conducted simultaneously.

[0032] All headings are for the convenience of the reader and should not be used to limit the meaning of the text that follows the heading, unless so specified.

[0033] The words "preferred" and "preferably" refer to embodiments of the invention that may afford certain benefits, under certain circumstances. However, other embodiments may also be preferred, under the same or other circumstances. Furthermore, the recitation of one or more preferred embodiments or aspects does not imply that other embodiments or aspects are not useful and is not intended to exclude other embodiments or aspects from the scope of the invention.

[0034] The term "comprises" and variations thereof do not have a limiting meaning where these terms appear in the description and claims. Such terms will be understood to imply the inclusion of a stated step or element or group of steps or elements but not the exclusion of any other step or element or group of steps or elements.

[0035] By "consisting of" is meant including, and limited to, whatever follows the phrase "consisting of." Thus, the phrase "consisting of" indicates that the listed elements are required or mandatory, and that no other elements may be present. By "consisting essentially of" is meant including any elements listed after the phrase, and limited to other elements that do not interfere with or contribute to the activity or action specified in the disclosure for the listed elements. Thus, the phrase "consisting essentially of" indicates that the listed elements are required or mandatory, but that other elements are optional and may or may not be present depending upon whether or not they materially affect the activity or action of the listed elements.

[0036] The singular form "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. These articles refer to one or to more than one (i.e., to at least one). As used herein, the term "or" is generally employed in its usual sense including "and / or" unless the content clearly dictates otherwise. The term "and / or" means any one or more of the items in the list joined by "and / or". As an example, "x and / or y" means any element of the three -element set { (x), (y), (x, y) } . In other words, "x and / or y" means "one or both of x and y". As another example, "x, y, and / or z" means any element of the seven-element set { (x), (y), (z), (x, y), (x, z), (y, z), (x, y, z) } . In other words, "x, y and / or z" means "one or more of x, y and z".

[0037] Where ranges are given, endpoints include all numbers subsumed within that range (e.g., 1 to 5 includes 1, 1.5, 2, 2.75, 3, 3.80, 4, 5, etc.). Furthermore, unless otherwise indicated or otherwise evident from the context and understanding of one of ordinary skill in the art, values that are expressed as ranges can assume any specific value or subrange within the stated ranges in different embodiments of the disclosure, to the tenth of the unit of the lower limit of the range, unless the context clearly dictates otherwise. Herein, "up to" a number (for example, up to 50) includes the number (for example, 50). The term "in the range" or "within a range" (and similar statements) includes the endpoints of the stated range.

[0038] Reference throughout this specification to "one aspect," "an aspect," "certain aspects," or "some aspects," "one embodiment," "an embodiment," "certain embodiment," or "some embodiment," etc., means that a particular feature, configuration, composition, or characteristic described in connection with the aspect is included in at least one aspect of the disclosure. Thus, the appearances of such phrases in various places throughout this specification are not necessarily referring to the same embodiment of the disclosure. Furthermore, the particular features, configurations, compositions, or characteristics may be combined in any suitable manner in one or more aspects.

[0039] Unless otherwise indicated, all numbers expressing quantities of components, molecular weights, and so forth used in the specification and claims are to be understood as being modified in all instances by the term "about." As used herein in connection with a measured quantity, the term "about" refers to that variation in the measured quantity as would be expected by the skilled artisan making the measurement and exercising a level of care commensurate with the objective of the measurement and the precision of the measuring equipment used. The term "about" as used in connection with a numerical value throughout the specification and the claims denotes an interval of accuracy, familiar and acceptable to a person skilled in the art. In general, such interval of accuracy is + / -10%. Accordingly, unless otherwise indicated to the contrary, the numerical parameters set forth in the specification and claims are approximations that may vary depending upon the desired properties sought to be obtained by the present invention. At the very least, and not as an attempt to limit the doctrine of equivalents to the scope of the claims, each numerical parameter should at least be construed in light of the number of reported significant digits and by applying ordinary rounding techniques.

[0040] Notwithstanding that the numerical ranges and parameters setting forth the broad scope of the invention arc approximations, the numerical values set forth in the specific examples arc reported as precisely as possible. All numerical values, however, inherently contain a range necessarily resulting from the standard deviation found in their respective testing measurements.

[0041] The term "exemplary" means serving as a non-limiting example, instance, or illustration. As utilized herein, the terms "e.g.," and "for example" set off lists of one or more non-limiting aspects, examples, instances, or illustrations.

[0042] As used herein, the term "substantially" refers to the qualitative condition of exhibiting total or near-total extent or degree of a characteristic or property of interest. Biological and chemical phenomena rarely, if ever, go to completion and / or proceed to completeness or achieve or avoid an absolute result. The term "substantially" is therefore used herein to capture the potential lack of completeness inherent in many biological and chemical phenomena. For example, "substantially" may refer to being within at least about 20%, alternatively at least about 10%, alternatively at least about 5% of a characteristic or property of interest.

[0043] The invention is defined in the claims. However, below is a non-exhaustive listing of nonlimiting exemplary aspects. Any one or more of the features of these aspects may be combined with any one or more features of another example, embodiment, or aspect described herein.

[0044] As used herein, the terms "subject", "individual", and "patient" are interchangeable, and relate to vertebrates, preferably mammals. For example, mammals in the context of the disclosure are humans, non-human primates, domesticated animals such as dogs, cats, sheep, cattle, goats, pigs, horses, etc., laboratory animals such as mice, rats, rabbits, guinea pigs, etc., as well as animals in captivity such as animals in zoos. The term "animal" as used herein includes humans. The term "subject" may also include a patient, i.e., an animal, having a disease. In exemplary aspects, a subject, individual, or patient refers to a human (e.g., a man, a woman, or a child).

[0045] As used herein, the term “non-transitory computer-readable medium” refers to a physical storage medium that can store data or instructions for use by a computer or other processing device. Non-limiting examples of non-transitory computer-readable media include hard drives, solid-state drives, USB flash drives, optical discs (e.g., CDs and DVDs), and memory cards.

[0046] As used herein, the term “trained machine learning model”, “machine trained model”, or “trained prediction model” refers to computational systems developed through machine learning techniques, where algorithms are trained on large datasets to for example, make predictions. During training, the model adjusts its parameters by analyzing data.III. Deep Learning Approach For Detecting Intracranial Hypertension

[0031] alCP serves as a digital biomarker of intracranial hypertension. The inventors developed and validated alCP both internally and externally for 1CP greater than 15 mmHg in patients with invasive ICP monitors. In patients without ICP monitoring, the inventors show that alCP is strongly associated with relevant neurological phenotypes, indicating that it may be able to serve as a screening tool in patients not subjected to monitoring but with extracranial monitors.

[0032] One aspect of the disclosure is a computer-implemented method of predicting or detecting increased intracranial pressure in a subject. Intracranial pressure (ICP) refers to the pressure exerted by fluids such as cerebrospinal fluid (CSF) and blood within the skull and brain tissue. It is a critical physiological parameter that helps maintain proper brain function and circulation. Normal ICP typically ranges from 7 to 15 mmHg in adults, but elevated ICP can occur due to conditions like traumatic brain injury, tumors, hydrocephalus, or cerebral edema. Increased ICP can compress brain structures, reduce blood flow, and lead to serious complications such as brain herniation or ischemia. Monitoring and managing ICP are essential in neurocritical care to prevent permanent neurological damage and ensure adequate cerebral perfusion.

[0033] In an embodiment, the method includes establishing, by using an alCP model, a prediction model based on at least a plurality of training data sets that are related to a plurality of subjects, each of the plurality of training data sets including at least one extracranial waveform.

[0034] In some embodiments, the training data sets include data for at least 20 patients. In some embodiments, the training data sets include data for at least 30 patients. In some embodiments, the training data sets include data for at least 40 patients. In some embodiments, the training data sets include data for at least 50 patients. In some embodiments, the training data sets include data for at least 60 patients. In some embodiments, the training data sets include data for at least 70 patients. In some embodiments, the training data sets include data for at least 80 patients. In some embodiments, the training data sets include data for at least 90 patients. In some embodiments, the training data sets include data for at least 100 patients. In some embodiments, the training data setsinclude data for at least 120 patients. In some embodiments, the training data sets include data for more than 100 patients.

[0035] In some embodiments, the training data sets include at least 100,000 waveforms. In some embodiments, the training data sets include at least 200,000 waveforms. In some embodiments, the training data sets include at least 300,000 waveforms. In some embodiments, the training data sets include at least 400,000 waveforms. In some embodiments, the training data sets include at least 500,000 waveforms. In some embodiments, the training data sets include at least 600,000 waveforms. In some embodiments, the training data sets include at least 700,000 waveforms. In some embodiments, the training data sets include at least 800,000 waveforms. In some embodiments, the training data sets include at least 900,000 waveforms. In some embodiments, the training data sets include at least 1,00,000 waveforms. In some embodiments, the training data sets include more than 1,000,000 waveforms.

[0036] In some embodiments, training data sets include at least two different extracranial waveforms. In some embodiments, training data sets include at least three different extracranial waveforms. In some embodiments, training data sets include at least four different extracranial waveforms. In some embodiments, training data sets include at least five different extracranial waveforms. In some embodiments, training data sets include at least six different extracranial waveforms. In some embodiments, training data sets include six or more different extracranial waveforms.

[0037] Extracranial waveforms refer to the patterns of blood flow or other physiological signals detected outside the skull, often through non-invasive techniques like Doppler ultrasound or plethysmography. These waveforms are commonly used to assess blood flow in extracranial arteries, such as the carotid arteries, providing valuable insights into vascular health and potential conditions like stenosis, atherosclerosis, or reduced perfusion. By analyzing waveform characteristics, including amplitude, velocity, and frequency, clinicians can evaluate blood flow dynamics, detect abnormalities, and guide diagnostic or therapeutic decisions. Extracranial waveform analysis plays a critical role in monitoring cerebrovascular and systemic circulatory health.

[0038] In an embodiment, the extracranial waveforms include arterial blood pressure waveforms. Arterial blood pressure waveforms represent the dynamic changes in blood pressure within arteries during a cardiac cycle, typically displayed as a graph of pressure versus time. These waveformsconsist of key features: the systolic upstroke (rise in pressure during ventricular contraction), the systolic peak (maximum pressure), the dicrotic notch (caused by aortic valve closure), and the diastolic decline (pressure drop as the heart relaxes). Analyzing these waveforms provides valuable insights into cardiovascular health, including heart function, arterial stiffness, and blood flow dynamics.

[0039] In an embodiment, the extracranial waveforms include electrocardiogram (EKG) leads II and / or V waveforms. Electrocardiogram (EKG) leads II and V are commonly used configurations for monitoring cardiac electrical activity. Lead II, part of the standard limb leads, measures the electrical potential difference between the right arm and left leg, offering a clear view of the heart's electrical activity along the axis most aligned with normal cardiac conduction. It is particularly useful for detecting arrhythmias and observing P waves, which represent atrial depolarization. Lead V, part of the precordial (chest) leads, includes various placements such as VI through V6, each positioned to provide a cross-sectional view of the heart. The waveforms produced by these leads, including the P wave, QRS complex, and T wave, help in diagnosing various heart conditions such as arrhythmias, heart attacks, and other cardiac abnormalities by showing the timing and progression of electrical impulses throughout the heart.

[0040] In an embodiment, the extracranial waveforms include respiratory waveforms. Respiratory waveforms are graphical representations of airflow, pressure, or volume changes in the respiratory system over time, often displayed during mechanical ventilation or respiratory monitoring. These waveforms provide critical insights into a patient’s breathing patterns, lung mechanics, and the interaction between the ventilator and the patient. Common types include volume-time, pressuretime, and flow-time waveforms, each offering unique information, such as tidal volume, airway pressures, and inspiratory / expiratory flow.

[0041] In an embodiment, the extracranial waveforms include plethysmography waveforms. Plethysmography waveforms arc graphical representations of changes in volume within an organ or body part, typically used to monitor blood flow, respiratory function, or other physiological parameters. These waveforms are generated by plethy sinographs, which measure variations in the size or shape of a body part, such as a finger or limb, in response to pulsatile blood flow or other physiological events.

[0042] In an embodiment, the extracranial waveforms include SpO2 waveforms. SpO2 waveforms are graphical representations of the oxygen saturation levels in a person's blood, typicallygenerated by a pulse oximeter. These waveforms illustrate the periodic fluctuations in light absorption that occur as blood flows through capillaries, allowing the device to measure the ratio of oxygenated to deoxygenated hemoglobin. The SpO2 waveform often includes both a plethysmographic (pleth) signal, showing the pulsatile nature of blood flow, and a numerical value indicating the percentage of oxygen saturation in the blood (SpO2).

[0043] In an embodiment, the extracranial waveforms include PPG waveforms. PPG (Photoplethysmogram) waveforms are graphical representations of the blood volume changes in the microvascular bed of tissue, typically captured using light-based sensors like LEDs and photodetectors. These waveforms are primarily used to monitor physiological parameters such as heart rate, blood oxygen saturation (SpO2), and circulatory health. The PPG signal reflects the cyclic variations in blood flow caused by the heartbeat, with each pulse producing a characteristic waveform.

[0044] In an aspect, the method further includes receiving, via one or more processors, a first data set for the subject. In an embodiment, the first data set includes at least one extracranial waveform data from the subject. In an embodiment, the first data set includes at least two different extracranial waveform data from the subject. In an embodiment, the first data set includes at least three different extracranial waveform data from the subject. In an embodiment, the first data set includes at least four different extracranial waveform data from the subject. In an embodiment, the first data set includes at least five different extracranial waveform data from the subject. In an embodiment, the first data set includes at least six different extracranial waveform data from the subject. In an embodiment, the first data set includes 5 or more different extracranial waveform data from the subject.

[0045] In an aspect, the method further includes determining, via one or more processors, an ICP value corresponding to the subject by analyzing the first data set for the subject using the prediction model or a trained machine learning model. In an embodiment, the trained machine learning model has been trained using a plurality of training data sets that are related to a plurality of subjects, each of the plurality of training data sets including at least one extracranial waveform.

[0046] In an aspect, the method further includes comparing, via one or more processors, the ICP value to a first threshold value and when the ICP value is equal to or greater than the first threshold value, generating a determination of ICP.

[0047] While TCP can fluctuate, an TCP greater than 20-25 mm of Hg (mmHg) is considered elevated, and an TCP greater than 40 mm of Hg (mmHg) is severely elevated. In an embodiment, the first threshold value is at least about 15 mmHg. In an embodiment, the first threshold value is about 15 mmHg. In an embodiment, the first threshold value is about 16 mmHg. In an embodiment, the first threshold value is about 17 mmHg. In an embodiment, the first threshold value is about 18 mmHg. In an embodiment, the first threshold value is about 19 mmHg. In an embodiment, the first threshold value is about 20 mmHg. In an embodiment, the first threshold value is about 21 mmHg. In an embodiment, the first threshold value is about 22 mmHg. In an embodiment, the first threshold value is about 23 mmHg. In an embodiment, the first threshold value is about 24 mmHg. In an embodiment, the first threshold value is about 25 mmHg. In an embodiment, the first threshold value is about 26 mmHg. In an embodiment, the first threshold value is about 27 mmHg. In an embodiment, the first threshold value is about 28 mmHg. In an embodiment, the first threshold value is about 29 mmHg. In an embodiment, the first threshold value is about 30 mmHg. In an embodiment, the first threshold value is at least about 35 mmHg. In an embodiment, the first threshold value is at least about 40 mmHg.

[0048] In an embodiment, as depicted in FIG. IB, if the subject’s ICP value is greater than or equal to 15 mmHg, the subject is determined to have or be at risk of having ICP.

[0049] In an aspect, the method further includes receiving, via one or more processors, a second data set for the subject. In an embodiment, the second data set is a subset of the first data set. In another embodiment, the second data set is generated or derived from the first data set.

[0050] In an aspect, the method further includes determining, via the one or more processors, a risk score by analyzing the second data set. In an embodiment, the second data set is a slice-level prediction model, as depicted in FIG. IB. A slice-level prediction model is a machine learning model that makes predictions based on distinct subsets, or "slices," of the input data. Each slice represents a specific segment or category of the data, such as a particular demographic group, time period, or other relevant features. In an aspect, each slice of the slice-level prediction model represents a specific time segment.

[0051] In an aspect, the method further includes determining, via one or more processors, a classification by comparing the risk score to a third threshold value. In an embodiment, when the risk score is lower than the third threshold value, the subject is classified as low risk. In an alternative embodiment, when the risk score is equal to or greater than the third threshold value,the subject is classified as high risk. A subject who is classified as low risk includes subject who arc unlikely to experience elevated ICP. A subject who is classified as high risk includes subjects who are likely to experience elevated pressure within the skull. In some embodiments, this elevated pressure may be associated with or caused by conditions such as traumatic brain injury, hydrocephalus, or stroke. Accurate classification enables timely intervention, improving patient outcomes by guiding monitoring strategies and therapeutic decisions.

[0052] In an aspect, the method further includes evaluating the association of aCIP using digital clinical data. In an embodiment, the digital clinical data may include electronic health records (EHRs), imaging reports, and / or other patient datasets. By combining digital clinical data with the disclosed methods, clinicians can uncover insights into procedure efficacy, disparities in care, and potential areas for improvement. Such evaluations are critical for optimizing care delivery, informing evidence-based practices, and improving patient outcomes in acute care settings.

[0053] The disclosed methods are non-invasive and can provide significant benefit to patients. The disclosed methods allow for safer, more accessible, and repeatable monitoring without the risks of invasive procedures. The disclosed methods also allow for large-scale studies and continuous monitoring.

[0054] In an aspect, the disclosed methods may be used to provide intracranial monitoring to subjects that have contraindications to invasive monitoring or detect intracranial pressure abnormalities. These subjects may have conditions such as coagulopathy, infections, or anatomical challenges that make traditional invasive methods, like intracranial catheter placement, unsafe. The disclosed methods offer a non-invasive way to monitor ICP trends, enabling timely diagnosis and management of intracranial hypertension while reducing the risks associated with invasive procedures.

[0055] In an aspect, the disclosed methods may be used to screen subjects for intracranial hypertension. Non-invasive screening for intracranial hypertension, such as those methods disclosed herein eliminate the risks associated with invasive monitoring, such as infection, bleeding, and procedural complications, making it particularly beneficial for high-risk patients.

[0056] In an aspect, the disclosed methods may be used to evaluate the significance of association of alCP with clinical phenotypes. These methods allow for provide valuable insights into the relationship between alCP and various clinical conditions while eliminating the need for surgical intervention and its subsequent downsides, including, patient discomfort, infection risk, andhealthcare costs. Tn some embodiments, the methods may be used to understand, determine, or predict how alCP correlates with clinical phenotypes, improving diagnostic precision and personalized treatment strategies.

[0057] In some aspects, the subject is a human. In some aspects, the subject is an adult.

[0058] One aspect of the disclosure includes the use of a non-transitory computer readable medium having instructions stored thereon, wherein execution of the instructions by a processor causes the processor to perform the functions and methods described herein. A controller in accordance with an exemplary embodiment of the present technology may comprise a data processor, a non- transitory computer-readable medium, and a data storage coupled to the data processor. The non- transitory computer- readable medium may comprise code, executable by the data processor, to perform the functions described herein. The data processor may store, for example, data for processing samples, sample data, or data for analyzing sample data. The computer-readable medium may comprise code, executable by the data processor to perform any suitable method.

[0059] The data processor may include any suitable data computation device or combination of such devices. An exemplary data processor may comprise one or more microprocessors working together to accomplish a desired function. The data processor may include a CPU that comprises at least one high-speed data processor adequate to execute program components for executing user and / or system-generated requests. The CPU may be a microprocessor such as AMD’s Athlon, Duron and / or Opteron; IBM and / or Motorola’s PowerPC; IBM’s and Sony’s Cell processor; Intel’s Celeron, Itanium, Pentium, Xeon, and / or XScale; Apple Ml, and / or the like processor(s).

[0060] The computer-readable medium and the data storage may be any suitable device or devices that can store electronic data. Examples of memories may comprise, for example, one or more memory chips, disk drives, etc. Such memories may operate using any suitable electrical, optical, and / or magnetic mode of operation.

[0061] The presently described technology and its advantages will be better understood by reference to the following examples. These examples are provided to describe specific implementations of the present technology. By providing these specific examples, it is not intended limit the scope and spirit of the present technology. It will be understood by those skilled in the art that the full scope of the presently described technology encompasses the subject matter defined by the claims appending this specification, and any alterations, modifications, or equivalents of those claims.

[0062] EXAMPLES

[0063] Study Setting and datasets

[0064] single-admission data from two distinct sources was used: 1) the publicly available MIMIC III Waveform Database Matched Subset which contains waveform records from bedside monitors for 10,282 patients admitted to intensive care units at the Beth Israel Deaconess Medical Center (Boston, MA) between 2001 and 2012, and 2) the MSH Bedmaster Matched Database, a database of waveform recordings for 50,894 patients admitted to the Mount Sinai Hospital (New York, NY) between 2018 and 2022 (FIG. 1A). The latter was derived from a deployment of the BedmasterTM software. This software is engineered to extract and store real-time patient data obtained from networked General Electric Healthcare bedside patient multi-parameter monitors (Excel Medical Electronics, Jupiter, FL).

[0065] Three cohorts from the two data sources described above were created: For training and internal validation of alCP, part of the MIMIC-III waveform dataset consisting of hospital stays with intracranial waveform recordings (MIMIC-ICP) was used. For external validation of alCP, a waveform dataset from Mount Sinai Hospital consisting of hospital stays with intracranial waveform recordings (MSH-ICP) was used. To map alCP to clinical associations, another, nonoverlapping subset from the MIMIC-III waveform dataset consisting of patients without intracranial waveform recordings was used (MIMIC-GENERAL) (FIG. 1A).

[0066] Preprocessing

[0067] Since intracranial pressure recordings in both data sources were recorded at 125 Hz, the extracranial waveforms were up-sampled or down-sampled (ABP, EKG, respiratory and PPG) to 125 Hz. Real-time arterial blood pressure (ABP) was obtained via a continuous arterial sampling from an arterial line. Respiratory waveforms were obtained by plethysmography. It was then filtered for quality control (FIG. 1A). Patient waveforms were included only if they had no missing data and had a non-zero mean and non-zero standard deviation across all waveforms. Next, all input variables to values between 0 and 1 were normalized before propagating them through the model. All the intracranial pressure waveforms collected were via external ventricular drain measurements in the intraventricular space. For intracranial pressure, boundaries of 2 mmHg to 45 mmHg were utilized and waveforms with standard deviations of zero were excluded, which may represent device malfunction. Subsequently, the intracranial waveforms were validated by two neurocritical care attendings. This procedure removed significant portions of missing orphysiologically implausible data that occurs during calibration and setup of the various lines. All patients with an intraventricular catheter had periods of elevated ICP > 15mm Hg. As a result, d weighted sampling of 10,000 1 -second segments from each patient in the training dataset was conducted to achieve class balance and to avoid over-representing patients with longer stays in the intensive care unit.

[0068] After filtering, the MIMIC-ICP dataset was partitioned into subsets, with 100 patients allocated to the training dataset, 20 patients to the validation dataset, and 37 patients to the test dataset. In the training dataset, this amounted to 1,000,000 waveforms containing 277 hours of data, with 200,000 waveforms and 55.5 h in the validation dataset, and 360,000 waveforms with 102.8 h. The MSH-ICP dataset had 56 patients after filtering (FIG. 1A) with a total of 560,000 waveforms and 155.6 hours of data. MIMIC-GENERAL had 1694 after filtering waveforms and HER data for missingness, containing 16,940,000 waveforms and 4,705.6 h of data.

[0069] A threshold of 15mm Hg was employed to define intracranial hypertension, as previously described in literature and to maximize the number of positive samples, given the relative low frequency (8%) in the dataset.

[0070] While various thresholds in the literature have been reported, the disclosed dataset had relatively lower prevalence at higher pressures. Performance of alCP at higher ICP thresholds of 20 mmHg and 22 mmHg analyzed and reported as shown in Table 1.

[0071] Table 1: ICP thresholds and corresponding prediction results.

[0072] Model Architecture

[0073] alCP utilizes a 5D-convolutional neural network which takes as input one second long segments of waveform data. These predictions arc aggregated to form patient level predictions. Each input consists of the 128 data points (approximately one second) of each of the five main perioperative waveforms (ABG, EKG leads II and V, respiratory, and PPG). The output for eachinput is a prediction at each second of whether the patient has TCP > 15mmof Hg. Implementation of the model architecture is visualized (FIG. IB). The loss function utilized a Tvcrsky Loss, which is a modified DICE score accounting for the relative class imbalance between positive and negative samples.

[0074] Briefly, the architecture follows a modified U-Net, a popular neural network design for semantic segmentation tasks, with a modified focus on handling input data with five channels rather than the traditional two or three. It is composed of two convolutional layers with batch normalization and rectified linear unit (ReLU) activation functions, aiming to capture non-linear temporal features within the input data. Skip connections to handle the high-dimensionality and large number of channels were added.

[0075] Slices within a given time series were labeled using the slice-level prediction model and then transformed into a sequence. The patient-level model aggregated these individual second predictions, generating a unified risk score reflecting the patient's frequency of dysregulation in intracranial pressure throughout their hospital stay. No database-related variables were used to generate the model architecture.

[0076] Model Development and Validation

[0077] The alCP model has approximately 10.4 million parameters and was trained on a single NVIDIA A100 GPU over the span of 3 days and 30 epochs. The segmentation performance was optimized using a modified Dice coefficient. The optimal model was selected and calibrated based on its performance on the validation dataset. Additionally, hyperparameter optimization, early stopping and an Adam optimizer within the PyTorch Lightning framework was employed.

[0078] Model Performance Metrics

[0079] Classification performance of time series and patient-level data using five key metrics: three threshold-dependent metrics: accuracy, sensitivity, specificity, and two thresholdindependent metrics: area-under-the-receiver-operator-curve (AUROC) and area-under-the- precision-recall curve (AUPRC) was assessed. The AUPRC curve was specifically reported due to high rates of class imbalance in the internal and external test dataset. Receiver operating curves (ROCs) and confusion matrices were evaluated the patient-level model performance in each classification task. The threshold was calibrated on the internal validation set, and then applied to the internal test set and external validation set.

[0080] Aa benchmark comparison against standard time series classification models was conducted. The benchmark included Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), and Temporal Convolutional Neural Networks (TCNs), which are state-of-the-art. The results of the benchmark are in Table 2.

[0081] Table 1. Benchmarking against other ML methods.0082]

[0083] Association with Phenotypes and Outcomes

[0084] To evaluate the association of alCP with relevant clinical outcomes in patients without intracranial pressure monitors, a testing set consisting of patients who had only extracranial waveforms (ABP, EKG, respiratory, PPG) in addition to clinical data from the electronic health record was used. alCP provided a patient- level risk score for 1694 patients. We also ran a phenome-wide association scan.

[0085] Statistical Analysis

[0086] The descriptive statistics for each cohort by mapping waveform IDs to electronic health records for available patients was calculated.

[0087] All experiments were seeded, to ensure reproducibility. The models underwent consistent training, validation, and testing on identical datasets as shown in FIG. 2. To quantify the uncertainty in experimental results, confidence intervals were computed using non-parametric bootstrapping 50 times. All evaluation and statistical metrics were computed utilizing the torchmetrics and statsmodels packages. Odds ratios were calculated by logistic regression and Fisher’s Exact Test.

[0088] To evaluate the significance of association of alCP with clinical phenotypes, an odds ratio with respect to a ten percentile increase in alCP was calculated. The unadjusted P-values and as well as the adjusted threshold for correlated phenotypes (P < 3.33x10-4) were reported. Thepatients were then stratified into low- and high-risk groups using the 75th percentile of predicted patient level risk score as the cutoff, and calculated odds ratios for prc-dctcrmincd outcomes. P- value was calculated using the Fisher’s Exact Test.

[0089] TRIPOD reporting guideline were filled out and included in the Supplement

[0090] Cohort Description

[0091] Average ages were 67.7 for the MIMIC-ICP. Females accounted for 40.8% of the MIMIC- 1CP cohort, and 60.5% identified as White. In the M1M1C-1CP cohort, 53.3% of patients were on Medicare, 23.3% on private insurance, and 20.0% on Medicaid. The MIMIC-ICP cohort exhibited an average intracranial blood pressure of 8.9mmHg, and intracranial hypertension values were observed for 9.7% of the total observation period in the MIMIC-ICP cohort. Neurological comorbidities were documented for all patients in both MIMIC-ICP, but only 19% of the MIMIC- ICP had documented cardiovascular comorbidities as defined by phecodes and as shown in Table 3.

[0092] Table 3. Patient Characteristics

[0093] Cardiovascular comorbidities in the study encompassed conditions such as ischemic heart disease, valvular disease, and arrhythmias. Additionally, neurological comorbidities were classified into categories including vascular etiologies (e.g., intracerebral hemorrhage), infectious etiologies (e.g., abscesses), and oncological issues (e.g., tumors). (FIG. 2). Patients with intracranial pressure monitors represent a range of different pathologies including traumatic brain injury (40.3%), cancer (25.0%), aneurysm (15.3%), and stroke (11.5%) (Table 4).

[0094] Table 4. Categorization of Intracranial Hypertension by underlying etiology in the training dataset

[0095] 56 patients were admitted to the Mount Sinai Hospital between 2020 and 2022 with intracranial monitors comprised of MSH-ICP. The average age was 54.6 (SD, 31.3) with 39.3% of patients identifying as female. 50.0% of patients identified as White, 16.1% as Black or African American, 17.8% as Hispanic, with the remaining 14.3% identifying as other. 28.6% of patientswere on Medicare, 42.8% on private insurance and 28.6% on Medicaid. This cohort demonstrated an average intracranial pressure of 16.7mmHg. Elevated ICP was observed for a total of 8%of the total observation period. All patients in this cohort had neurological and cardiovascular comorbidities.

[0096] For the clinical association cohort (MIMIC-GENERAL), the average age was 69.3 years and 41% of the patients were female. 71.0% of patients identified as White, 10.0% as Black or African American, 2.7% as Hispanic, and 16.3% as other. 55.4% were on Medicare, 30.3% on private insurance, and 10.7% on Medicaid. 52.8% of patients had an identifiable neurological comorbidity and 56.1% had an identified cardiovascular comorbidity.

[0097] alCP Performance

[0098] Patient-level performance of alCP using the testing set from MIMIC-ICP (Table 5) was assessed. The Receiver Operating Characteristic (ROC) curve was employed to measure the model’s ability to detect elevated intracranial pressure (ICP > 15mm Hg), with overall accuracy of 0.97, and an area-under-the-curve (AUC) of 0.91. The sensitivity of the model at a threshold of 0.5 was 0.87 and the specificity was 1.0. At higher cutoff points for elevated ICP (20 mmHg and 22 mmHg), the model had worse performance. This may be due to a reduced number of positive samples or different underlying physiologies. The results outperformed those previously reported (Table 5). It was also benchmarked against other deep learning architectures for time series data. Improvements in model architectures mostly lead to increased sensitivity with little to no difference in specificity.

[0099] Table 5. Performance of classification model for elevated ICP.Performance of patient level classification model on internal and external testing set, evaluated with accuracy (threshold: 0.5), sensitivity (threshold: 0.5), specificity (threshold: 0.5), area under the receiving-operator-curve, and area under the precision-recall curve for ICP elevation.

[0100] The performance on MSH-ICP, external validation test set was evaluated. The area-under- thc-curvc on the external validation test set was 0.80, with an accuracy of 0.72, a sensitivity of 0.63 and a specificity of 0.74. The Receiver Operating Curve and Precision-Recall Curves for internal and external validation testing sets showed discriminatory performance of alCP to identify patients with ICP values greater than 15mmHg (FIGs. 3A and 3B).

[0101] To evaluate the association of alCP with relevant clinical outcomes to pressure in patients without intracranial pressure monitors, MIMIC-GENERAL, a testing set consisting of patients that had extracranial waveforms (ABP, EKG, respiratory and PPG) and linked clinical data was used. alCP provided a patient level risk score for 1,694 patients. An odds ratio was calculated with respect to a 10 percent increase in alCP. A ten-percentile increase in alCP was associated with an increased likelihood of stroke (OR = 2.12, 95% CI, (1.27-1.13), P = 4.06 x 10’3), brain malignancy (OR = 1.68; 95% CI, (1.09-2.60), P = 1.93 x IO’2), subdural hemorrhage (OR = 1.66, 95% CI, (1.07-2.57), P = 2.38 x 10'2), intracerebral hemorrhage (OR = 1.18, 95% CI, (1.07-1.32), P = 1.17xl0'3). Moreover, when looking at patient procedures done over the course of admission, a ten percentile increase in alCP was associated with a percutaneous brain biopsy (OR = 1.58, 95% CI, (1.15-2.18), P - 4.58 x 10'3), and craniotomy and resection (OR = 1.43, 95% CI, (1.12-1.84), P - 4.10 x 10'3).

[0102] To see if the effect of alCP was larger in high-risk groups, the patients were stratified into low and high-risk groups-above 75th percentile was deemed high-risk. The inventors calculated odds ratios and P-values via the Fisher’s Exact Test for pre-determined outcomes. Patients in the top quartile demonstrated increased risk of subdural hemorrhage (OR = 24.2, P = 3.01 x IO-2), traumatic brain injury (OR = 6.04, P = 3.8 xlO"2), intracerebral hemorrhage (OR = 1.85, P = 1.32 x 10'3), and receiving a craniectomy (OR = 7.55, P = 1.58 x 10“2) and a percutaneous brain biopsy (OR = 5.03, P = 2.72 x IO'2).

[0103] Explainability behind elevated alCP

[0104] alCP and the corresponding input waveforms with two different patients across was visualized (FIGs. 4A and 4B). In the first example, a patient demonstrates different slightly elevated alCP. The EKG is normal, the arterial waveform has a flat systolic phase, and the plethysmography is normal. In the second example, a patient with significantly elevated Aicp demonstrates prolonged QT-interval, a high dicrotic notch on arterial waveform and bronchospasmon plethysmography. Pathological alCP correlates strongly with pathologies on other waveforms. However, more work needs to be done to highlight the underlying etiologies of intracranial hypertension and their manifestations in waveforms.

[0105] alCP versus ICP on the scale of 30 minutes for 3 patients in the testing dataset was also visualized. (FIGs. 5A - 5C).

[0106] ReferencesBaggiani, Marta, Francesca Graziano, Paola Rebora, Chiara Robba, Angelo Guglielmi, Stefania Galimberti, Carlo Giussani, Jose I. Suarez, Raimund Helbok, and Giuseppe Citerio. 2023. “Intracranial Pressure Monitoring Practice, Treatment, and Effect on Outcome in Aneurysmal Subarachnoid Hemorrhage.” Neurocritical Care 38 (3): 741-51.Baneke, Alex J., James Aubry, Ananth C. Viswanathan, and Gordon T. Plant. 2020. “The Role of Intracranial Pressure in Glaucoma and Therapeutic Implications.” Eye 34 (1): 178-91.Brasil, Sergio, Davi Jorge Fontoura Solla, Ricardo de Carvalho Nogueira, Manoel Jacobsen Teixeira, Luiz Marcelo Sa Malbouisson, and Wellingson da Silva Paiva. 2021. “A Novel Noninvasive Technique for Intracranial Pressure Waveform Monitoring in Critical Care.” Journal of Personalized Medicine 11 (12). https: / / doi.org / 10.3390 / jpml l l21302.Chen, Weiting, Xia Zhang, Xiuxiu Ye, and Pan Ying. 2023. “Diagnostic Accuracy of Optic Nerve Sheath Diameter on Ultrasound for the Detection of Increased Intracranial Pressure in Patients with Traumatic Brain Injury: A Systematic Review and Meta-analysis.” Biomedical Reports 19 (6). https: / / doi.org / 10.3892 / br.2023.1685.Corradi V, Swanson NR. Nonparametric Bootstrap Procedures for Predictive Inference Based on Recursive Estimation Schemes. SSRN; 2007.Chen, Zhenqin, Mengying Wang, Meiyu Zhang, Wei Huang, Hanjie Gu, and Jinshan Xu.2023. “Post-Processing Refined ECG Delineation Based on ID-UNet.” Biomedical Signal Processing and Control 79 (January): 104106.Chesnut, Randall M., Nancy Temkin, Nancy Carney, Sureyya Dikmen, Carlos Rondina, Walter Videtta, Gustavo Petroni, et al. 2012. “A Trial of Intracranial-Pressure Monitoring in Traumatic Brain Injury.” The New England Journal of Medicine 367 (26): 2471.Detlefsen, Nicki, Jiri Borovec, Justus Schock, Ananya Jha, Teddy Koker, Luca Di Liello, Daniel Stand, Changsheng Quan, Maxim Grechkin, and William Falcon. 2022. “TorchMetrics - Measuring Reproducibility in PyTorch.” Journal of Open Source Software 7 (70): 4101.Dubourg, Julie, Etienne Javouhey, Thomas Geeraerts, Mahmoud Messerer, and Behrouz Kassai. 2011. “Ultrasonography of Optic Nerve Sheath Diameter for Detection of Raised Intracranial Pressure: A Systematic Review and Meta-Analysis.” Intensive Care Medicine 37 (7): 1059-68.Fernando, Shannon M., Alexandre Tran, Wei Cheng, Bram Rochwerg, Monica Taljaard, Kwadwo Kyeremanteng, Shane W. English, et al. 2019. “Diagnosis of Elevated Intracranial Pressure in Critically Ill Adults: Systematic Review and Meta-Analysis.” BMJ 366 (July): 14225.Hawryluk, Gregory W. J., Giuseppe Citerio, Peter Hutchinson, Angelos Kolias, Geert Meyfroidt, Chiara Robba, Nino Stocchetti, and Randall Chesnut. 2022. “Intracranial Pressure: Current Perspectives on Physiology and Monitoring.” Intensive Care Medicine 48 (10): 1471-81.Hawryluk, Gregory W. J., Jessica L. Nielson, J. Russell Huie, Lara Zimmermann, Rajiv Saigal, Quan Ding, Ryan Hirschi, Frederick A. Zeiler, Adam R. Ferguson, and Geoffrey T. Manley. 2020. “Analysis of Normal High-Frequency Intracranial Pressure Values and Treatment Threshold in Neurocritical Care Patients.” JAMA Neurology 77 (9): 1150.Kamat, Pradip, Sachin Kunde, Miriam Vos, Atul Vats, Nitika Gupta, Thomas Heffron, Rene Romero, and James D. Fortenberry. 2012. “Invasive Intracranial Pressure Monitoring Is a Useful Adjunct in the Management of Severe Hepatic Encephalopathy Associated with Pediatric Acute Liver Failure.” Pediatric Critical Care Medicine: A Journal of the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies 13 (1): e33- 8.Lau, Vincent Issac, and Robert Thomas Arntfield. 2017. “Point-of-Care Transcranial Doppler by Intensivists.” Critical Ultrasound Journal 9 (1): 21.Lazaridis, Christos, Aswathy Ajith, Ali Mansour, David O. Okonkwo, Ramon Diaz-Arrastia, and Anoop Mayampurath. 2022. “Prediction of Intracranial Hypertension and Brain Tissue Hypoxia Utilizing High-Resolution Data from the BOOST-II Clinical Trial.” Neurotrauma Reports 3 (1): 473-78.Le Roux, Peter, David K. Menon, Giuseppe Citerio, Paul Vespa, Mary Kay Bader, Gretchen M. Brophy, Michael N. Diringer, et al. 2014. “Consensus Summary Statement of the International Multidisciplinary Consensus Conference on Multimodality Monitoring in Neurocritical Care.” Neurocritical Care 21 (2): 1-26.Lim, Leerang, and Hyung-Chul Lee. 2023. “Open Datasets in Perioperative Medicine: A Narrative Review.” Anesthesia and Pain Medicine 18 (3): 213-19.Manley, Geoffrey, Claude Hemphill, and Shirley Stiver. 2009. Intracranial Pressure and Brain Monitoring XIII: Mechanisms and Treatment. Springer Science & Business Media.Megjhani, Murad, Kalijah Terilli, Bennett Weinerman, Daniel Nametz, Soon Bin Kwon, Angela Velazquez, Shivani Ghoshal, et al. 2023. “A Deep Learning Framework for Deriving Noninvasive Intracranial Pressure Waveforms from Transcranial Doppler.” Annals of Neurology 94 (1): 196— 202.Muller, Sebastian Johannes, Elina Henkes, Matthew J. Gounis, Stephan Felber, Oliver Ganslandt, and Hans Henkes. 2023. “Non-Invasive Intracranial Pressure Monitoring.” Journal of Clinical Medicine Research 12 (6). https: / / doi.org / 10.3390 / jcml2062209.Nag, Deb Sanjay, Seelora Sahu, Amlan Swain, and Shashi Kant. 2019. “Intracranial Pressure Monitoring: Gold Standard and Recent Innovations.” World Journal of Clinical Cases 7 (13): 1535-53.Nair, Shiker S., Alina Guo, Joseph Boen, Ataes Aggarwal, Ojas Chahal, Arushi Tandon, Meer Patel, et al. 2023. “A Real-Time Deep Learning Approach for Inferring Intracranial Pressure from Routinely Measured Extracranial Waveforms in the Intensive Care Unit.” BioRxiv. https: / / doi.org / 10.1101 / 2023.05.16.23289747.Nattino, Giovanni, Lorenzo Gamberini, Obou Brissy, Greta Carrara, Randall Chesnut, Valentina Chiarini, Arturo Chieregato, et al. 2023. “Comparative Effectiveness of Intracranial Pressure Monitoring on 6-Month Outcomes of Critically Ill Patients With Traumatic Brain Injury.” JAMA Network Open 6 (9): e2334214.Robba, Chiara, Francesca Graziano, Paola Rebora, Francesca Elli, Carlo Giussani, Mauro Oddo, Geert Meyfroidt, et al. 2021. “Intracranial Pressure Monitoring in Patients with Acute Brain Injury in the Intensive Care Unit (SYNAPSE-ICU): An International, Prospective Observational Cohort Study.” Lancet Neurology 20 (7): 548-58.Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015. Springer International Publishing; 2015:234-241.Salehi SSM, Erdogmus D, Gholipour A. Tversky Loss Function for Image SegmentationUsing 3D Fully Convolutional Deep Networks. In: Machine Learning in Medical Imaging. Springer International Publishing; 2017:379-387.Seabold, Skipper, and Josef Perktold. 2010. “Statsmodels: Econometric and Statistical Modeling with Python.” In Proceedings of the 9th Python in Science Conference. SciPy. https: / / doi.org / 10.25080 / majora-92bfl922-011.Sepehrinezhad, Ali, Asadollah Zarifkar, Gholamreza Namvar, Ali Shahbazi, and Roger Williams. 2020. “Astrocyte Swelling in Hepatic Encephalopathy: Molecular Perspective of Cytotoxic Edema.” Metabolic Brain Disease 35 (4): 559-78.Tversky, A., and D. Kahneman. 1974. “Judgment under Uncertainty: Heuristics and Biases.” Science 185 (4157): 1124-31.Wei, Wei-Qi, Lisa A. Bastarache, Robert J. Carroll, Joy E. Mario, Travis J. Osterman, Eric R. Gamazon, Nancy J. Cox, Dan M. Roden, and Joshua C. Denny. 2017. “Evaluating Phecodes, Clinical Classification Software, and ICD-9-CM Codes for Phenome-Wide Association Studies in the Electronic Health Record.” PloS One 12 (7): e0175508.Wijdicks, Eelco F. M. 2022. “10 or 15 or 20 or 40 MmHg? What Is Increased Intracranial Pressure and Who Said So?” Neurocritical Care 36 (3): 1022-26.Yan, Jun, Junxia Meng, and Jianhu Zhao. 2021. “Bottom Detection from Backscatter Data of Conventional Side Scan Sonars through ID-UNet.” Remote Sensing 13 (5): 1024.

[0107] In summary, the inventors developed alCP to detect intracranial pressure abnormalities demonstrating both strong performance and clinical significance. alCP does not have any hardware requirements, and consequently, implementation of alCP as a tool for point-of-care bedside monitoring at the start of admission can occur on routine bedside monitors. Moreover, an alCP- generated risk score for each patient over the course of a patient’s hospital stay, could reduce detection times, and improve outcomes, especially given the time-sensitive nature of neurological diseases such as stroke and hemorrhage. Finally, the inventors anticipate that alCP has the potential to detect pathophysiological conditions like hepatic encephalopathy and glaucoma, which are associated with intracranial pressure changes but do not have clear neurovascular etiologies.

[0108] All features disclosed in the specification, including the claims, abstracts, and drawings, and all the steps in any method or process disclosed, may be combined in any combination, except combinations where at least some of such features and / or steps are mutually exclusive. Each feature disclosed in the specification, including the claims, abstract, and drawings, can be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly statedotherwise. Thus, unless expressly stated otherwise, each feature disclosed is one example only of a generic series of equivalent or similar features.

[0109] It will be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.

Claims

CLAIMS1. A computer-implemented method of predicting or detecting increased intracranial pressure in a subject, the method comprising: establishing, by using an alCP model, an ICP prediction model based on at least a plurality of training data sets that are related to a plurality of subjects, each of the plurality of training data sets including at least one extracranial waveform; receiving, via one or more processors, a first data set for the subject; determining, via one or more processors, an ICP value corresponding to the subject by analyzing the first data set using the ICP prediction model; comparing, via one or more processors, the ICP value to a first threshold value; and wherein the ICP value is equal to or greater than the first threshold value, generating a determination of ICP.

2. The method of claim 1, wherein the first data set comprises at least one extracranial waveform data from the subject.

3. The method of claim 1 or claim 2, wherein the extracranial waveform data comprises arterial blood pressure (ABG) waveforms, electrocardiogram (EKG) leads II and / or V waveforms, respiratory waveforms, plethysmography waveforms, SpCh waveforms, and / or Pho toplethy smogram (PPG) waveforms.

4. The method of any one of claims 1 to 3, wherein the first threshold value is at least about 15 mm of Hg.

5. A computer-implemented method of predicting or detecting increased intracranial pressure in a subject, the method comprising: receiving, via one or more processors, a first data set for the subject;determining, via one or more processors, an TCP value corresponding to the subject by analyzing the first data set using a trained machine learning model; comparing, via one or more processors, the TCP value to a first threshold value; and wherein the TCP value is equal to or greater than the first threshold value, generating a determination of TCP.

6. The method of claim 5, wherein the TCP value is at least about 15 mm of Hg.

7. The method of claim 5 or claim 6, wherein the first data set comprises extracranial waveform data.

8. The method of any one of claims 5 to 7, wherein the trained machine learning model has been trained using a plurality of training data sets that are related to a plurality of subjects, each of the plurality of training data sets including at least one extracranial waveform.

9. The method of claim 7 or claim 8, wherein the extracranial waveform data comprises arterial blood pressure (ABG), electrocardiogram (EKG) leads II and / or V, respiratory, plethysmography, SpCh, and / or PPG waveforms.

10. The method of any one of claims 1 to 9, wherein the method further comprises: receiving, via one or more processors, a second data set for the subject; determining, via the one or more processors, a risk score by analyzing the second data set using a risk prediction model; comparing, via the one or more processors, the risk score to a second threshold value; and when the risk score meets the second threshold value, generating a prediction of frequency of dysregulation in ICP.

11. The method of claim 10, wherein the risk prediction model is a slice-level prediction model.

12. The method of claim 10 or claim 11, wherein the method further comprises: determining, via one or more processors, a classification by comparing the risk score to a third threshold value.

13. The method of claim 12, wherein:(i) when the risk score is lower than the third threshold value, the subject is classified as low risk; or(ii) when the risk score is equal to or greater than the third threshold value, the subject is classified as high risk14. The method of claim 10, wherein the second data set is generated from the first data set.

15. The method of any one of claims 1 to 14, wherein the method further comprises evaluating the association of aCIP using digital clinical data for the subject.

16. The method of any one of claims 1 to 15, wherein the method is used to:(i) provide intracranial monitoring to subjects that have contraindications to invasive monitoring;(ii) screen subjects for intracranial hypertension;(iii) to evaluate the significance of association of alCP with clinical phenotypes; and / or(iv) detect intracranial pressure abnormalities.

17. The method of any one of claims 1 to 16, wherein the subject is a human.

18. The method of claim 17, wherein the subject is an adult.

19. A non-transitory computer readable medium having instructions stored thereon, wherein execution of the instructions by a processor causes the processor to: receive a first data set for the subject; determine an ICP value corresponding to the subject by analyzing the first data set for the subject using a trained machine learning model; and generate a determination of ICP by comparing the ICP value to a first threshold value.

20. The non-transitory computer readable medium of claim 19, wherein execution of the instructions by a processor further causes the processor to: receive a second data set for the subject; determine a risk score by analyzing the second data set for the subject; and generate a prediction of frequency of dysregulation in ICP by comparing the risk score to a second threshold value.

21. The non-transitory computer readable medium of claim 20, wherein execution of the instructions by a processor further causes the processor to: determine a classification by comparing the risk score to a third threshold value.

Images