Accurate prediction of neurologic changes in critically ill infants using pose ai

Abstract

Systems and methods for predicting a neurologic condition in subjects less than one year old using a movement index derived from pose estimation are provided. Video data of a body of a subject is captured over time using cameras. The video data is processed using a pose estimation model to detect and track anatomical landmarks of the subject in successive video frames. A movement index is computed for a defined time interval based on a statistical measure of variance in the position of the tracked anatomical landmarks across the interval within the video data, where the movement index is normalized to account for body size and camera distance. The movement index is inputted into a trained machine learning model to obtain, as output from the model, a prediction of the neurologic condition in the subject representing a clinically relevant neurological state.

Description

Attorney Docket No.: 104593-5055-WOACCURATE PREDICTION OF NEUROLOGIC CHANGES IN CRITICALLY ILL INFANTS USING POSE AlCROSS REFERENCE TO RELATED PATENT APPLICATIONS

[0001] This application claims priority to United States Provisional Patent Application No. 63 / 700,301, entitled “Accurate Prediction of Neurologic Changes in Critically Ill Infants using Pose Al,” filed September 27, 2024, which is hereby incorporated by reference.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0002] This invention was made with government support under Clinical and Translational Science Awards (CTSA) grant number UL1TR004419 awarded by the National Center for Advancing Translational Sciences and grant numbers S100D026880 and S100D030463 awarded by the Office of Research Infrastructure of the National Institutes of Health. The government has certain rights in the invention.TECHNICAL FIELD

[0003] The present invention relates generally to predicting neurologic conditions in infants using movement indexes derived from pose estimations.BACKGROUND

[0004] Infant alertness is considered the most sensitive piece of the neurologic exam, reflecting integrity throughout the central nervous system. Infant mental status changes can be due to encephalopathy, sedation, or other causes and are highly dynamic, necessitating continuous assessment.

[0005] Encephalopathy can be caused by a variety of diagnoses that require rapid identification and treatment (e.g., hypoxic, metabolic, and infectious etiologies). Lethargy, a sign of encephalopathy, is one of the most important indicators of neonatal sepsis. While encephalopathy is an example of pathology that impacts alertness, infant mental status can also be purposefully manipulated, most commonly with sedative medications. Titrating the appropriate level of sedation is challenging in any patient population, but more so in infants due to their inability to communicate and the extreme pharmacokinetic variability of sedativeDB2 / 650910081.2 1Attorney Docket No.: 104593-5055-WO medications. In infants, these medications have longer half-lives, high rates of tachyphylaxis, opioid antagonist metabolites, and more pronounced respiratory depression.

[0006] Mental status is routinely evaluated to make life-saving decisions in the neonatal intensive care unit (NICU). Despite its importance, the ability to continuously quantify and monitor infant alertness is limited.SUMMARY

[0007] Given the above background, there is a need in the art for improved systems and methods to quantify and monitor infant mental status and changes to mental status.

[0008] The present disclosure overcomes the above-identified drawbacks with a pose artificial intelligence (pose Al) approach to track infant movement that provides a continuous and relevant method of neurological monitoring. The present disclosure further overcomes the above-identified drawbacks by providing systems and methods for using infant movement to predict neurologic changes such as sedation and cerebral dysfunction.

[0009] One aspect disclosed herein provides a method for predicting a neurologic condition in a subject using a movement index derived from pose estimation, where the subject is a human subject less than one year old. In some embodiments, the method includes (a) capturing video data of the subject over a duration of time using one or more cameras positioned to observe the body of the subject. In some embodiments, the method further includes (b) processing the video data using a pose estimation artificial intelligence model to detect and track a plurality of anatomical landmarks of the subject in successive video frames in the video data. In some embodiments, the method further includes (c) computing a movement index for a defined time interval based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks across the interval within the video data, where the movement index is normalized to account for body size and camera distance. In some embodiments, the method further includes (d) inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, where the neurologic condition represents a clinically relevant neurological state.

[0010] Another aspect of the present disclosure provides a system for predicting a neurologic condition in a subject, the system comprising a camera configured to capture video data of a body of a subject over a duration of time, where the subject is a human subjectDB2 / 650910081.2 2Attorney Docket No.: 104593-5055-WO less than one year old; and a processor operatively coupled to the camera and configured to perform a method. In some embodiments, the method includes processing the video data using a pose estimation artificial intelligence model to detect and track a plurality of anatomical landmarks of the subject in successive video frames in the video data; computing a movement index for a defined time interval based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks across the interval within the video data, where the movement index is normalized to account for body size and camera distance; and inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, where the neurologic condition represents a clinically relevant neurological state.

[0011] Yet another aspect of the present disclosure provides a non-transitory computer- readable medium storing instructions that, when executed by one or more processors, cause the system to perform a method for predicting a neurologic condition. In some embodiments, the method includes receiving video data of a subject captured by a camera, where the subject is a human subject less than one year old; applying a pose estimation artificial intelligence model to the video data to detect and track a plurality of anatomical landmarks of the subject across a sequence of video frames; computing a movement index for a defined time interval based on a statistical variance in the positions of the anatomical landmarks, where the movement index is normalized for patient body size and camera distance; and inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, where the neurologic condition represents a clinically relevant neurological state.

[0012] The details of one or more embodiments are set forth in the accompanying drawings and the description below. Other features and advantages will be apparent from the description and drawings, and from the claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0013] For a better understanding of the aforementioned implementations of the invention as well as additional implementations thereof, reference should be made to the Description of Implementations below, in conjunction with the following drawings in which like reference numerals refer to corresponding parts throughout the figures.

[0014] FIGS. 1 A and IB is a block diagram of an example system for predicting a neurologic condition, according to some embodiments of the present disclosure.DB2 / 650910081.2 3Attorney Docket No.: 104593-5055-WO

[0015] FIGS. 2A, 2B, 2C, 2D and 2E are example flowcharts of a method for predicting a neurologic condition according to some embodiments of the present disclosure, in which optional steps are shown by dashed lines.

[0016] FIGS. 3A, 3B, and 3C collectively illustrate application of a pose estimation artificial intelligence model to critically ill infants, in accordance with some embodiments of the present disclosure. A large database of video-EEG data (N = 115 infants, 4,705 hours of video, 10.4 Tb) was built and stored on a HIPAA-compliant supercomputing cluster. A pose estimation Al model was then trained and tested on 2712 manually labelled video frames with DeepLabCut to predict infant anatomic landmarks. Finally, clinical utility of quantifying movement and using infant movement features to predict sedation and cerebral dysfunction was demonstrated. Abbreviation: DNN = deep neural network.

[0017] FIG. 4 illustrates a data filtering workflow for predicting a neurologic condition, in accordance with some embodiments of the present disclosure. The full data workflow including filters, sample sizes for each group, and train-test datasets comported with TRIPOD guidelines. Performance was evaluated on training data through k-fold repeated measures cross-validation. Performance was also evaluated on two held-out test sets, held-out infants and held-out frames from infants used in training.

[0018] FIGS. 5A, 5B, 5C, and 5D collectively illustrate performance of pose recognition with a pose estimation artificial intelligence model, in accordance with an embodiment of the present disclosure. A pose estimation artificial intelligence model comprising DeepLabCut on a ResNet-50 backbone was used. FIG. 5A illustrates that the L2 pixel error was comparable to literature and consistently low across all labeled non-occluded anatomic landmarks for training data (“Training,” median 3.2 pixels, N = 18,399), frames held out from training (“New Frames,” median 3.5 pixels, N = 797), and frames from infants held out from training (“New Babies,” median 4.6 pixels, N = 973). Thirteen labeled landmarks (0.06%) had a L2 error >50 pixels and their errors are shown with arrows. FIG. 5B illustrates that pose tracking had high area under ROC curves when predicting landmark occlusion. FIG. 5C illustrates the percentage of key points within a reference distance threshold (26.2 pixels, half the height of a head) was >95% for training and test data. FIG. 5D illustrates exemplary video frames from an infant with KCNQ2 epilepsy syndrome showing the predicted anatomic landmarks during a seizure.DB2 / 650910081.2 4Attorney Docket No.: 104593-5055-WO

[0019] FIGS. 6A and 6B collectively illustrate infant movement, predicted using pose Al, increased with age and decreased with sedative medications and EEG abnormalities, in accordance with an embodiment of the present disclosure. FIG. 6A illustrates movement variance within 1 -minute intervals (y-axis, log-scaled) increased with corrected age (x-axis) in patients with sedative medications or abnormal EEGs (dark gray, right-hand blocks per pair; N = 110,622 minutes, 53 infants) and in those without (black, left-hand blocks per pair; N = 29,093 minutes, 46 infants). FIG. 6B illustrates that among infants <44 weeks postmenstrual age, movement was lower with cerebral dysfunction (N = 21,541 minutes, 21 infants), phenobarbital (N = 1,488 minutes, 3 infants), both (N = 55,724 minutes, 24 infants), or those receiving sedative infusions (N = 5378 minutes, 3 infants) when compared to infants with normal EEG and neither medication group (N = 27,099 minutes, 35 infants). This was also statistically significant for all permutation tests (a permutation P-value of <10'3is labeled with ***). Abbreviation: PB = phenobarbital.

[0020] FIGS. 7A and 7B collectively illustrate machine learning models trained on infant movement to predict sedation and cerebral dysfunction, in accordance with an embodiment of the present disclosure. FIG. 7A illustrates XGBoost receiver operating characteristic curves for sedation (left panel) and cerebral dysfunction classifiers (right panel) trained on pose Al predictions. XGBoost ROC-AUCs are from subsets of the training dataset (median as solid line, 2.5thand 97.5thpercentiles as dashed lines), and on test datasets of held-out frames and held-out infants. The dashed black line is a reference, indicating the baseline performance of a classifier operating purely by chance (ROCAUC = 0.5). FIG. 7B illustrates feature importance from 100 repeats of five-fold cross-validation for both classifiers showing that anatomic landmarks from the limbs are more important than the nose and neck when predicting sedation and cerebral dysfunction.

[0021] FIGS. 8A, 8B, 8C, and 8D collectively illustrate pose Al prediction of sedation during therapeutic hypothermia for an infant with hypoxic ischemic encephalopathy, in accordance with an embodiment of the present disclosure. FIG. 8A illustrates sedation probability (y-axis) as a function of age in days (x-axis) for a full-term infant with therapeutic hypothermia. Sedation probability increased after hypothermia induction shown in FIG. 8C and after phenobarbital administration (regularly dashed lines). FIG. 8B illustrates the EEG had increased epileptiform discharges at two and three days old, so phenobarbital was administered to protect against seizures during rewarming. FIG. 8C illustrates temperature during hypothermia and rewarming. FIG. 8D illustrates exemplary frames during increasingDB2 / 650910081.2 5Attorney Docket No.: 104593-5055-WO levels of sedation, each annotated with a heat map depicting infant movement over the prior five minutes.

[0022] Reference will now be made in detail to implementations, examples of which are illustrated in the accompanying drawings. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that the present invention may be practiced without these specific details.DESCRIPTION OF IMPLEMENTATIONS

[0023] The implementations described herein provide various technical solutions to quantify and monitor infant mental status and changes to mental status.

[0024] Conventional methods for continuous quantification and monitoring of infant mental status and / or alertness commonly employ physical examinations. This method provides a single snapshot, is subjective, and can be delayed. Validated instruments, such as the NPASS for sedation or modified Sarnat for encephalopathy, mitigate exam subjectivity with high inter-rater reliability but are labor-intensive and not continuous.Electroencephalography (EEG) is a continuous record of neurologic activity. However, its implementation requires specialized staff and equipment not available at every NICU, and it carries the risk of pressure injuries with prolonged use. Applicability of EEG to alertness is also limited, as there is no association between EEG and neonatal depth of sedation. Furthermore, although continuous EEG can assess encephalopathy by indicating the degree of cerebral dysfunction, there is limited clinical adoption aside from evaluation for therapeutic hypothermia. This is likely owing to EEG’s high interrater variability and challenging interpretability among bedside providers. Overall, determining the level of an infant’s alertness remains a challenge, despite being fundamental to clinical care in the NICU.

[0025] The present disclosure addresses this unmet need with systems and methods for tracking infant movement using pose Al that provides a continuous and relevant method of neurological monitoring, which can further be used to predict neurologic changes such as sedation and cerebral dysfunction using machine learning models. In some aspects, the present disclosure provides methods and systems for predicting a neurologic condition in a subject (e.g., a human subject less than one year old) using a movement index derived from pose estimation. Video data of a subject is captured over time using one or more camerasDB2 / 650910081.2 6Attorney Docket No.: 104593-5055-WO positioned to observe the body of the subject. The video data is processed using a pose estimation artificial intelligence model to detect and track anatomical landmarks of the subject in successive video frames in the video data. A movement index is computed for a defined time interval based on a statistical measure of variance in the position of the tracked anatomical landmarks across the interval within the video data, where the movement index is normalized to account for body size and camera distance. The movement index is inputted into a trained machine learning model to obtain, as output from the model, a prediction of the neurologic condition in the subject representing a clinically relevant neurological state.

[0026] Advantageously, the presently disclosed methods and systems augment physical examinations in a scalable manner. By using a deep learning approach to track anatomic landmarks, the presently disclosed methods and systems can be used to continuously characterize changes in neurological phenotypes in the NICU. For instance, as described in Examples 1-6 below, a system and method in accordance with the present disclosure was successfully applied to a large dataset comprising 4,705 video-EEG hours for 115 infants to accurately predict sedation and cerebral dysfunction in critically ill infants. A deep learning pose algorithm in accordance with some embodiments accurately predicted anatomic landmarks in three evaluation sets (ROC-AUCs 0.83-0.94), showing feasibility of applying pose Al in an ICU. Classifiers trained on landmarks from pose Al exhibited high performance for sedation (ROC-AUCs 0.87-0.91) and cerebral dysfunction (ROC-AUCs 0.76-0.91), demonstrating that an EEG diagnosis can be predicted from video data alone. Taken together, deep learning with pose Al offers a scalable, minimally invasive method for neuro-telemetry in the NICU.

[0027] As used herein, the term “classification” refers to any number(s) or other characters(s) that are associated with a particular property of a sample or input (e.g., video data, one or more movement indices, or any portion or representation thereof). For example, in some embodiments, the term “classification” refers to prediction of a neurologic condition, including but not limited to sedation or cerebral dysfunction. The classification can be binary (e.g., positive or negative) or have more levels of classification (e.g., a scale from 1 to 10 or 0 to 1). The terms “cutoff’ and “threshold” can refer to predetermined numbers used in an operation. For example, a cutoff size can refer to a size above which fragments are excluded. A threshold value can be a value above or below which a particular classification applies. Either of these terms can be used in either of these contexts.DB2 / 650910081.2 7Attorney Docket No.: 104593-5055-WO

[0028] As used interchangeably herein, the term “classifier” or “model” refers to a machine learning model or algorithm.

[0029] In some embodiments, a model includes an unsupervised learning algorithm. One example of an unsupervised learning algorithm is cluster analysis. In some embodiments, a model includes supervised machine learning. Nonlimiting examples of supervised learning algorithms include, but are not limited to, logistic regression, neural networks, support vector machines, Naive Bayes algorithms, nearest neighbor algorithms, random forest algorithms, decision tree algorithms, boosted trees algorithms, multinomial logistic regression algorithms, linear models, linear regression, Gradient Boosting, mixture models, hidden Markov models, Gaussian NB algorithms, linear discriminant analysis, or any combinations thereof. In some embodiments, a model is a multinomial classifier algorithm. In some embodiments, a model is a 2-stage stochastic gradient descent (SGD) model. In some embodiments, a model is a deep neural network (e.g., a deep-and-wide sample-level model).

[0030] Neural networks. In some embodiments, the model is a neural network (e.g., a convolutional neural network and / or a residual neural network). Neural network algorithms, also known as artificial neural networks (ANNs), include convolutional and / or residual neural network algorithms (deep learning algorithms). In some embodiments, neural networks are machine learning algorithms that are trained to map an input dataset to an output dataset, where the neural network includes an interconnected group of nodes organized into multiple layers of nodes. For example, in some embodiments, the neural network architecture includes at least an input layer, one or more hidden layers, and an output layer. In some embodiments, the neural network includes any total number of layers, and any number of hidden layers, where the hidden layers function as trainable feature extractors that allow mapping of a set of input data to an output value or set of output values. In some embodiments, a deep learning algorithm comprises a neural network including a plurality of hidden layers, e.g., two or more hidden layers. In some instances, each layer of the neural network includes a number of nodes (or “neurons”). In some embodiments, a node receives input that comes either directly from the input data or the output of nodes in previous layers, and performs a specific operation, e.g., a summation operation. In some embodiments, a connection from an input to a node is associated with a parameter (c.g, a weight and / or weighting factor). In some embodiments, the node sums up the products of all pairs of inputs, xi, and their associated parameters. In some embodiments, the weighted sum is offset with a bias, b. In some embodiments, the output of a node or neuron is gated using aDB2 / 650910081.2 8Attorney Docket No.: 104593-5055-WO threshold or activation function, f, which, in some instances, is a linear or non-linear function. In some embodiments, the activation function is, for example, a rectified linear unit (ReLU) activation function, a Leaky ReLU activation function, or other function such as a saturating hyperbolic tangent, identity, binary step, logistic, arcTan, softsign, parametric rectified linear unit, exponential linear unit, softPlus, bent identity, softExponential, Sinusoid, Sine, Gaussian, or sigmoid function, or any combination thereof.

[0031] In some implementations, the weighting factors, bias values, and threshold values, or other computational parameters of the neural network, are “taught” or “learned” in a training phase using one or more sets of training data. For example, in some implementations, the parameters are trained using the input data from a training dataset and a gradient descent or backward propagation method so that the output value(s) that the ANN computes are consistent with the examples included in the training dataset. In some embodiments, the parameters are obtained from a back propagation neural network training process.

[0032] Any of a variety of neural networks are suitable for use in accordance with the present disclosure. Examples include, but are not limited to, feedforward neural networks, radial basis function networks, recurrent neural networks, residual neural networks, convolutional neural networks, residual convolutional neural networks, and the like, or any combination thereof. In some embodiments, the machine learning makes use of a pre-trained and / or transfer-learned ANN or deep learning architecture. In some implementations, convolutional and / or residual neural networks are used, in accordance with the present disclosure.

[0033] For instance, a deep neural network model includes an input layer, a plurality of individually parameterized (e.g., weighted) convolutional layers, and an output scorer. The parameters (e.g., weights) of each of the convolutional layers as well as the input layer contribute to the plurality of parameters (e.g., weights) associated with the deep neural network model. In some embodiments, at least 100 parameters, at least 1000 parameters, at least 2000 parameters or at least 5000 parameters are associated with the deep neural network model. As such, deep neural network models require a computer to be used because they cannot be mentally solved. In other words, given an input to the model, the model output needs to be determined using a computer rather than mentally in such embodiments. See, for example, Krizhevsky et al., 2012, “Imagenet classification with deep convolutional neural networks,” in Advances in Neural Information Processing Systems 2, Pereira, Burges, Bottou,DB2 / 650910081.2 9Attorney Docket No.: 104593-5055-WOWeinberger, eds., pp. 1097-1105, Curran Associates, Inc.; Zeiler, 2012 “ADADELTA: an adaptive learning rate method,” CoRR, vol. abs / 1212.5701; and Rumelhart etal., 1988, “Neurocomputing: Foundations of research,” ch. Learning Representations by Back- propagating Errors, pp. 696-699, Cambridge, MA, USA: MIT Press, each of which is hereby incorporated by reference.

[0034] Neural network algorithms, including convolutional neural network algorithms, suitable for use as models are disclosed in, for example, Vincent et al., 2010, “Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion,” J Mach Learn Res 11, pp. 3371-3408; Larochelle et al., 2009, “Exploring strategies for training deep neural networks,” J Mach Learn Res 10, pp. 1-40; and Hassoun, 1995, Fundamentals of Artificial Neural Networks, Massachusetts Institute of Technology, each of which is hereby incorporated by reference. Additional example neural networks suitable for use as models are disclosed in Duda et al., 2001, Pattern Classification, Second Edition, John Wiley & Sons, Inc., New York; and Hastie et al., 2001, The Elements of Statistical Learning, Springer-Verlag, New York, each of which is hereby incorporated by reference in its entirety. Additional example neural networks suitable for use as models are also described in Draghici, 2003, Data Analysis Tools for DNA Microarrays, Chapman & Hall / CRC; and Mount, 2001, Bioinformatics: sequence and genome analysis, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, each of which is hereby incorporated by reference in its entirety.

[0035] Support vector machines. In some embodiments, the model is a support vector machine (SVM). SVM algorithms suitable for use as models are described in, for example, Cristianini and Shawe-Taylor, 2000, “An Introduction to Support Vector Machines,” Cambridge University Press, Cambridge; Boser et al., 1992, “A training algorithm for optimal margin classifiers,” in Proceedings of the 5th Annual ACM Workshop on Computational Learning Theory, ACM Press, Pittsburgh, Pa., pp. 142-152; Vapnik, 1998, Statistical Learning Theory, Wiley, New York; Mount, 2001, Bioinformatics: sequence and genome analysis, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y.; Duda, Pattern Classification, Second Edition, 2001, John Wiley & Sons, Inc., pp. 259, 262-265; and Hastie, 2001, The Elements of Statistical Learning, Springer, New York; and Furey et al., 2000, Bioinformatics 16, 906-914, each of which is hereby incorporated by reference in its entirety. When used for classification, SVMs separate a given set of binary labeled data with a hyper-plane that is maximally distant from the labeled data. For certain cases in which noDB2 / 650910081.2 10Attorney Docket No.: 104593-5055-WO linear separation is possible, SVMs work in combination with the technique of 'kernels', which automatically realizes a non-linear mapping to a feature space. The hyper-plane found by the SVM in feature space corresponds, in some instances, to a non-linear decision boundary in the input space. In some embodiments, the plurality of parameters (e.g., weights) associated with the SVM define the hyper-plane. In some embodiments, the hyperplane is defined by at least 10, at least 20, at least 50, or at least 100 parameters and the SVM model requires a computer to calculate because it cannot be mentally solved.

[0036] Naive Bayes algorithms. In some embodiments, the model is a Naive Bayes algorithm. Naive Bayes models suitable for use as models are disclosed, for example, in Ng et al., 2002, “On discriminative vs. generative classifiers: A comparison of logistic regression and naive Bayes,” Advances in Neural Information Processing Systems, 14, which is hereby incorporated by reference. A Naive Bayes model is any model in a family of “probabilistic models” based on applying Bayes’ theorem with strong (naive) independence assumptions between the features. In some embodiments, they are coupled with Kernel density estimation. See, for example, Hastie et al. , 2001, The elements of statistical learning : data mining, inference, and prediction, eds. Tibshirani and Friedman, Springer, New York, which is hereby incorporated by reference.

[0037] Nearest neighbor algorithms. In some embodiments, a model is a nearest neighbor algorithm. In some implementations, nearest neighbor models are memory-based and include no model to be fit. For nearest neighbors, given a query point xo (a test subject), the k training points X(r), r, ... , k (here the training subjects) closest in distance to xo are identified and then the point xo is classified using the k nearest neighbors. In some embodiments, Euclidean distance in feature space is used to determine distance as= llx(0— x(o) lh Typically, when the nearest neighbor algorithm is used, the abundance data used to compute the linear discriminant is standardized to have mean zero and variance 1. In some embodiments, the nearest neighbor rule is refined to address issues of unequal class priors, differential misclassification costs, and feature selection. Many of these refinements involve some form of weighted voting for the neighbors. For more information on nearest neighbor analysis, see Duda, Pattern Classification, Second Edition, 2001, John Wiley & Sons, Inc; and Hastie, 2001, The Elements of Statistical Learning, Springer, New York, each of which is hereby incorporated by reference.DB2 / 650910081.2 11Attorney Docket No.: 104593-5055-WO

[0038] A k-nearest neighbor model is a non-parametric machine learning method in which the input consists of the k closest training examples in feature space. The output is a class membership. An object is classified by a plurality vote of its neighbors, with the object being assigned to the class most common among its k nearest neighbors (k is a positive integer, typically small). If k = 1, then the object is simply assigned to the class of that single nearest neighbor. See, Duda el al., 2001, Pattern Classification, Second Edition, John Wiley & Sons, which is hereby incorporated by reference. In some embodiments, the number of distance calculations needed to solve the k-nearest neighbor model is such that a computer is used to solve the model for a given input because it cannot be mentally performed.

[0039] Random forest, decision tree, and boosted tree algorithms. In some embodiments, the model is a decision tree. Decision trees suitable for use as models are described generally by Duda, 2001, Pattern Classification, John Wiley & Sons, Inc., New York, pp. 395-396, which is hereby incorporated by reference. Tree-based methods partition the feature space into a set of rectangles, and then fit a model (like a constant) in each one. In some embodiments, the decision tree is random forest regression. For example, one specific algorithm is a classification and regression tree (CART). Other specific decision tree algorithms include, but are not limited to, ID3, C4.5, MART, and Random Forests. CART, ID3, and C4.5 are described in Duda, 2001, Pattern Classification, John Wiley & Sons, Inc., New York, pp. 396-408 and pp. 411-412, which is hereby incorporated by reference. CART, MART, and C4.5 are described in Hastie et al, 2001, The Elements of Statistical Learning, Springer-Verlag, New York, Chapter 9, which is hereby incorporated by reference in its entirety. Random Forests are described in Breiman, 1999, “Random Forests— Random Features,” Technical Report 567, Statistics Department, U.C. Berkeley, September 1999, which is hereby incorporated by reference in its entirety. In some embodiments, the decision tree model includes at least 10, at least 20, at least 50, or at least 100 parameters (e.g., weights and / or decisions) and requires a computer to calculate because it cannot be mentally solved.

[0040] Regression. In some embodiments, the model uses a regression algorithm. In some embodiments, a regression algorithm is any type of regression. For example, in some embodiments, the regression algorithm is logistic regression. In some embodiments, the regression algorithm is logistic regression with lasso, L2 or elastic net regularization. In some embodiments, those extracted features that have a corresponding regression coefficient that fails to satisfy a threshold value are pruned (removed from) consideration. In someDB2 / 650910081.2 12Attorney Docket No.: 104593-5055-WO embodiments, a generalization of the logistic regression model that handles multicategory responses is used as the model. Logistic regression algorithms are disclosed in Agresti, An Introduction to Categorical Data Analysis, 1996, Chapter 5, pp. 103-144, John Wiley & Son, New York, which is hereby incorporated by reference. In some embodiments, the model makes use of a regression model disclosed in Hastie et al., 2001, The Elements of Statistical Learning, Springer-Verlag, New York. In some embodiments, the logistic regression model includes at least 10, at least 20, at least 50, at least 100, or at least 1000 parameters (e.g., weights) and requires a computer to calculate because it cannot be mentally solved.

[0041] Linear discriminant analysis algorithms. In some embodiments, linear discriminant analysis (LDA), normal discriminant analysis (ND A), or discriminant function analysis is a generalization of Fisher’s linear discriminant, a method used in statistics, pattern recognition, and machine learning to find a linear combination of features that characterizes or separates two or more classes of objects or events. In some embodiments, the resulting combination is used as the model (linear model) in some embodiments of the present disclosure.

[0042] Mixture model and Hidden Markov model. In some embodiments, the model is a mixture model, such as that described in McLachlan et al., Bioinformatics 18(3):413-422, 2002. In some embodiments, in particular, those embodiments including a temporal component, the model is a hidden Markov model such as described by Schliep et al., 2003, Bioinformatics 19(l):i255-i263.

[0043] Clustering. In some embodiments, the model is an unsupervised clustering model. In some embodiments, the model is a supervised clustering model. Clustering algorithms suitable for use as models are described, for example, at pages 211-256 of Duda and Hart, Pattern Classification and Scene Analysis, 1973, John Wiley & Sons, Inc., New York, (hereinafter "Duda 1973") which is hereby incorporated by reference in its entirety. As an illustrative example, in some embodiments, the clustering problem is described as one of finding natural groupings in a dataset. To identify natural groupings, two issues are addressed. First, a way to measure similarity (or dissimilarity) between two samples is determined. This metric (e.g., similarity measure) is used to ensure that the samples in one cluster are more like one another than they are to samples in other clusters. Second, a mechanism for partitioning the data into clusters using the similarity measure is determined. One way to begin a clustering investigation is to define a distance function and to compute the matrix of distances between all pairs of samples in the training set. If distance is a goodDB2 / 650910081.2 13Attorney Docket No.: 104593-5055-WO measure of similarity, then the distance between reference entities in the same cluster is significantly less than the distance between the reference entities in different clusters. However, in some implementations, clustering does not use a distance metric. For example, in some embodiments, a nonmetric similarity function s(x, x') is used to compare two vectors x and x'. In some such embodiments, s(x, x') is a symmetric function whose value is large when x and x' are somehow “similar.” Once a method for measuring “similarity” or “dissimilarity” between points in a dataset has been selected, clustering uses a criterion function that measures the clustering quality of any partition of the data. Partitions of the dataset that extremize the criterion function are used to cluster the data. Particular exemplary clustering techniques contemplated for use in the present disclosure include, but are not limited to, hierarchical clustering (agglomerative clustering using a nearest-neighbor algorithm, farthest-neighbor algorithm, the average linkage algorithm, the centroid algorithm, or the sum-of-squares algorithm), k-means clustering, fuzzy k-means clustering algorithm, and Jarvis-Patrick clustering. In some embodiments, the clustering includes unsupervised clustering (e.g., with no preconceived number of clusters and / or no predetermination of cluster assignments).

[0044] Ensembles of models and boosting. In some embodiments, an ensemble (two or more) of models is used. In some embodiments, a boosting technique such as AdaBoost is used in conjunction with many other types of learning algorithms to improve the performance of the model. In this approach, the output of any of the models disclosed herein, or their equivalents, is combined into a weighted sum that represents the final output of the boosted model. In some embodiments, the plurality of outputs from the models is combined using any measure of central tendency known in the art, including but not limited to a mean, median, mode, a weighted mean, weighted median, weighted mode, etc. In some embodiments, the plurality of outputs is combined using a voting method. In some embodiments, a respective model in the ensemble of models is weighted or unweighted.

[0045] As used herein, the term “parameter” refers to any coefficient or, similarly, any value of an internal or external element (e.g., a weight and / or a hyperparameter) in an algorithm, model, regressor, and / or classifier that can affect (e.g., modify, tailor, and / or adjust) one or more inputs, outputs, and / or functions in the algorithm, model, regressor and / or classifier. For example, in some embodiments, a parameter refers to any coefficient, weight, and / or hyperparameter that can be used to control, modify, tailor, and / or adjust the behavior, learning, and / or performance of an algorithm, model, regressor, and / or classifier. In someDB2 / 650910081.2 14Attorney Docket No.: 104593-5055-WO instances, a parameter is used to increase or decrease the influence of an input (e.g., a feature) to an algorithm, model, regressor, and / or classifier. As a nonlimiting example, in some embodiments, a parameter is used to increase or decrease the influence of a node (e.g., of a neural network), where the node includes one or more activation functions. Assignment of parameters to specific inputs, outputs, and / or functions is not limited to any one paradigm for a given algorithm, model, regressor, and / or classifier but can be used in any suitable algorithm, model, regressor, and / or classifier architecture for a desired performance. In some embodiments, a parameter has a fixed value. In some embodiments, a value of a parameter is manually and / or automatically adjustable. In some embodiments, a value of a parameter is modified by a validation and / or training process for an algorithm, model, regressor, and / or classifier (e.g., by error minimization and / or backpropagation methods). In some embodiments, an algorithm, model, regressor, and / or classifier of the present disclosure includes a plurality of parameters. In some embodiments, the plurality of parameters is n parameters, where: n > 2; n > 5; n > 10; n > 25; n > 40; n > 50; n > 75; n > 100; n > 125; n > 150; n > 200; n > 225; n > 250; n > 350; n > 500; n > 600; n > 750; n > 1,000; n > 2,000; n > 4,000; n > 5,000; n > 7,500; n > 10,000; n > 20,000; n > 40,000; n > 75,000; n > 100,000; n > 200,000; n > 500,000, n > 1 x 106, n > 5 x 106, or n > 1 x 107. As such, the algorithms, models, regressors, and / or classifiers of the present disclosure cannot be mentally performed. In some embodiments n is between 10,000 and 1 x 107, between 100,000 and 5 x 106, or between 500,000 and 1 x 106. In some embodiments, the algorithms, models, regressors, and / or classifier of the present disclosure operate in a k-dimensional space, where k is a positive integer of 5 or greater (e.g., 5, 6, 7, 8, 9, 10, etc. . As such, the algorithms, models, regressors, and / or classifiers of the present disclosure cannot be mentally performed.

[0046] As used herein, the term “untrained model” (e.g., “untrained classifier”) refers to a machine learning model or algorithm, such as a classifier or a neural network, that has not been trained on a target dataset. In some embodiments, “training a model” (e.g., “training a classifier”) refers to the process of training an untrained or partially trained model (e.g., “an untrained or partially trained classifier”). Moreover, it will be appreciated that the term “untrained model” does not exclude the possibility that transfer learning techniques are used in such training of the untrained or partially trained model. For instance, Fernandes et al., 2017, “Transfer Learning with Partial Observability Applied to Cervical Cancer Screening,” Pattern Recognition and Image Analysis: 8thIberian Conference Proceedings, 243-250, which is hereby incorporated by reference, provides non-limiting examples of such transferDB2 / 650910081.2 15Attorney Docket No.: 104593-5055-WO learning. In instances where transfer learning is used, the untrained model described above is provided with additional data over and beyond that of the primary training dataset. Typically, this additional data is in the form of parameters (e.g., coefficients, weights, and / or hyperparameters) that were learned from another, auxiliary training dataset. Moreover, while a description of a single auxiliary training dataset has been disclosed, it will be appreciated that there is no limit on the number of auxiliary training datasets that can be used to complement the primary training dataset in training the untrained model in the present disclosure. For instance, in some embodiments, two or more auxiliary training datasets, three or more auxiliary training datasets, four or more auxiliary training datasets or five or more auxiliary training datasets are used to complement the primary training dataset through transfer learning, where each such auxiliary dataset is different than the primary training dataset. Any manner of transfer learning is used, in some such embodiments. For instance, consider the case where there is a first auxiliary training dataset and a second auxiliary training dataset in addition to the primary training dataset. In such a case, the parameters learned from the first auxiliary training dataset (by application of a first model to the first auxiliary training dataset) are applied to the second auxiliary training dataset using transfer learning techniques (e.g., a second model that is the same or different from the first model), which in turn results in a trained intermediate model whose parameters are then applied to the primary training dataset and this, in conjunction with the primary training dataset itself, is applied to the untrained model. Alternatively, in another example embodiment, a first set of parameters learned from the first auxiliary training dataset (by application of a first model to the first auxiliary training dataset) and a second set of parameters learned from the second auxiliary training dataset (by application of a second model that is the same or different from the first model to the second auxiliary training dataset) are each individually applied to a separate instance of the primary training dataset (e.g., by separate independent matrix multiplications) and both such applications of the parameters to separate instances of the primary training dataset in conjunction with the primary training dataset itself (or some reduced form of the primary training dataset such as principal components or regression coefficients learned from the primary training set) are then applied to the untrained model in order to train the untrained model.

[0047] Several aspects are described herein with reference to example applications for illustration. It should be understood that numerous specific details, relationships, and methods are set forth to provide a full understanding of the features described herein. OneDB2 / 650910081.2 16Attorney Docket No.: 104593-5055-WO having ordinary skill in the relevant art, however, will readily recognize that the features described herein can be practiced without one or more of the specific details or with other methods. The features described herein are not limited by the illustrated ordering of acts or events, as some acts can occur in different orders and / or concurrently with other acts or events. Furthermore, not all illustrated acts or events are required to implement a methodology in accordance with the features described herein.

[0048] Reference is made herein to embodiments, examples of which are illustrated in the accompanying drawings. In the present disclosure, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be apparent to one of ordinary skill in the art that the present disclosure can be practiced without these specific details. In other instances, well-known methods, procedures, components, circuits, and networks have not been described in detail so as not to unnecessarily obscure aspects of the embodiments.

[0049] Details of implementations are now described in relation to the Figures.

[0050] Example System Embodiments

[0051] An exemplary system for predicting a neurologic condition is now described in conjunction with FIGS. 1 A-B. Advantageously, the example system illustrated in FIGS. 1 A- B improves upon conventional methods for predicting neurologic conditions using neurological monitoring.

[0052] FIGS. 1 A-B is a block diagram illustrating a system in accordance with some implementations. The device 100 in some implementations includes one or more processing units CPU(s) 74 (also referred to as processors), one or more network interfaces 16, a user interface 78, e.g., including a display 82 and / or an input 80 (e.g., a mouse, touchpad, keyboard, etc.), memory (non-persistent and / or persistent) 92, and one or more communication buses 13 for interconnecting these components. The one or more communication buses 13 optionally include circuitry (sometimes called a chipset) that interconnects and controls communications between system components. The memory 92 typically includes high-speed random access memory, such as DRAM, SRAM, DDR RAM, ROM, EEPROM, flash memory, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, magnetic disk storage devices, optical disk storage devices, flash memory devices, or other non-volatile solid state storage devices, or any combination thereof. The memoryDB2 / 650910081.2 17Attorney Docket No.: 104593-5055-WO optionally includes one or more storage devices remotely located from the CPU(s) 74. The memory comprises a non-transitory computer readable storage medium. In some implementations, the memory 92 or alternatively a non-transitory computer readable storage medium stores the following programs, modules and data structures, or a subset thereof:• an optional operating system 116, which includes procedures for handling various basic system services and for performing hardware dependent tasks;• an optional network communication module (or instructions) 118 for connecting the system 100 with other devices and / or a communication network 105;• a continuous neurological monitoring module 120 for capturing video data of the subject over a duration of time using one or more cameras positioned to observe the body of the subject;• video data 130 of the subject, the video data optionally comprising a plurality of video frames 132 (e.g., 132-1, 132-2,. . . 132-M), where M is an integer of 2 or greater;• a pose estimation model 140, optionally for processing the video data 130 to detect and track a plurality of anatomical landmarks 140 (e.g., 140-1-1, 140-1-2,... 140-1-N, 140-2-1, 140-2-2,. . . 140-2-N) of the subject in successive video frames 132 in the video data 130, where N is an integer of 2 or greater;• a movement index data structure 150, optionally comprising a movement index 154 (e.g., 154-1-1,... 154-1 -S) for a defined time interval 152 (e.g., 152-1,. .. 152-T) based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks 140 across the interval within the video data 130, wherein the movement index is normalized to account for body size and camera distance; and• a machine learning model 160, optionally for obtaining as output, responsive to inputting the movement index 154 into the model, a prediction of the neurologic condition in the subject.

[0053] Although FIGS. 1 A-B depicts a “system 100,” the figure is intended more as a functional description of the various features that may be present in computer systems than as a structural schematic of the implementations described herein. In practice, and as recognized by those of ordinary skill in the art, items shown separately could be combined and some items could be separated. Moreover, some or all of the data and modules depicted in memory in FIGS. 1 A-B may be in persistent memory and / or nonpersistent memory. The above identified modules, data, or programs (e.g., sets of instructions) need not be implemented as separate software programs, procedures, datasets, or modules, and thusDB2 / 650910081.2 18Attorney Docket No.: 104593-5055-WO various subsets of these modules and data may be combined or otherwise re-arranged in various implementations.

[0054] In some implementations, non-persistent memory optionally stores a subset of the modules and data structures identified above. Furthermore, in some embodiments, the memory stores additional modules and data structures not described above. In some embodiments, one or more of the above-identified elements is stored in a computer system, other than that of system 100, that is addressable by system 100 so that system 100 may retrieve all or a portion of such data when needed.

[0055] For purposes of illustration in FIGS. 1A-B, system 100 is represented as a single computer that includes all of the functionality for discovering or confirming at least one molecular or clinical feature is associated with a disease state or therapeutic response. However, while a single machine is illustrated, the term “system” shall also be taken to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein. For example, in some embodiments, system 100 includes one or more computers. In some embodiments, the functionality for predicting a neurologic condition is spread across any number of networked computers and / or resides on each of several networked computers and / or is hosted on one or more virtual machines at a remote location accessible across the communications network 105.

[0056] The system may operate in the capacity of a server or a client machine in clientserver network environment, as a peer machine in a peer-to-peer (or distributed) network environment, or as a server or a client machine in a cloud computing infrastructure or environment. The system may be a personal computer (PC), a tablet PC, a set-top box (STB), a Personal Digital Assistant (PDA), a cellular telephone, a web appliance, a server, a network router, a switch or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) that specify actions to be taken by that machine. In another implementation, the system comprises a virtual machine that includes a module for executing instructions for performing any one or more of the methodologies disclosed herein. In computing, a virtual machine (VM) is an emulation of a computer system that is based on computer architectures and provides functionality of a physical computer. Some such implementations may involve specialized hardware, software, or a combination of hardware and software.DB2 / 650910081.2 19Attorney Docket No.: 104593-5055-WO

[0057] One of skill in the art will appreciate that any of a wide array of different computer topologies are used for the application and all such topologies are within the scope of the present disclosure.

[0058] Example Methods

[0059] Now that details of a system 100 for predicting a neurologic condition has been disclosed, details regarding processes and features of the system, in accordance with various embodiments of the present disclosure, are disclosed below. Specifically, FIGS. 2A-E are flowcharts that illustrate the method 200 for predicting a neurologic condition, according to some embodiments of the subject computing system. In some embodiments, the method is carried out by one or more programs of the subject computer system described herein.

[0060] In accordance with FIGS. 2A-E, the present disclosure relates to systems, methods, and computer-readable media for predicting a neurologic condition of a subject using a movement index derived from pose estimation artificial intelligence. Referring to FIGS. 2A-E, one aspect of the present disclosure provides a method 200 for predicting a neurologic condition in a subject using a movement index derived from pose estimation.

[0061] Referring to Block 202, in some embodiments, the subject is a human subject less than one year old. Referring to Block 204, in some embodiments, the subject has a postmenstrual age of 64 weeks or less. Referring to Block 206, in some embodiments, the subject has a postmenstrual age of 48 weeks or less. Postmenstrual age (PMA) refers to the chronological age of a neonate or infant measured from the first day of the mother’s last menstrual period (LMP) to a specific point in time following birth. In clinical and biomedical contexts, particularly in neonatal and pediatric care, PMA is used to assess developmental and physiological maturity, integrating both gestational age at birth and the time elapsed postnatally. For the purposes of this disclosure, PMA is expressed in weeks and is calculated as the sum of the gestational age (e.g., the duration in weeks from the first day of the maternal LMP to the date of birth) and the postnatal age (e.g., the duration in weeks from the date of birth to the time of assessment).

[0062] In some embodiments, the subject has a postmenstrual age (PMA) of at least 20 weeks, at least 24 weeks, at least 28 weeks, at least 32 weeks, at least 36 weeks, at least 40 weeks, at least 44 weeks, at least 48 weeks, at least 56 weeks, at least 64 weeks, or at least 72 weeks. In some embodiments, the subject has a PMA of no more than 80 weeks, no more than 72 weeks, no more than 64 weeks, no more than 56 weeks, no more than 48 weeks, noDB2 / 650910081.2 20Attorney Docket No.: 104593-5055-WO more than 44 weeks, no more than 36 weeks, no more than 28 weeks, or no more than 24 weeks. In some embodiments, the subject has a PMA of from 20 weeks to 44 weeks, from 23 weeks to 41 weeks, from 24 weeks to 60 weeks, from 36 weeks to 72 weeks, or from 44 weeks to 80 weeks. In some embodiments, the subject has a PMA that falls within another range starting no lower than 20 weeks and ending no higher than 80 weeks.

[0063] As described above, infant alertness is considered a highly sensitive piece of the neurologic exam, reflecting integrity throughout the central nervous system. Infant mental status changes can be due to encephalopathy, sedation, or other causes and are highly dynamic, necessitating continuous assessment. Encephalopathy can be caused by a variety of diagnoses which require rapid identification and treatment (e.g., hypoxic, metabolic, and infectious etiologies). Lethargy, a sign of encephalopathy, is one of the most important indicators of neonatal sepsis. Other causes for decreased infant alertness, include purposeful manipulations to infant mental status, such as with sedative medications.

[0064] Referring to Block 208, in some embodiments, the subject has been administered with a sedative medication and / or an anti-seizure medication. In some embodiments, the sedative medication and / or anti-seizure medication comprises levetiracetam, phenobarbital, or pentobarbital. In some embodiments, the sedative medication and / or anti-seizure medication comprises one or more of phenobarbital, pentobarbital, levetiracetam, fosphenytoin, midazolam, lorazepam, and diazepam.

[0065] In some embodiments, the subject has a neurologic condition. In some embodiments, the neurologic condition comprises sedation, cerebral dysfunction, lethargy, and / or decreased alertness. In some embodiments, the subject comprises a clinically relevant neurological state. In some embodiments, the clinically relevant neurological state comprises sedation management in the settings of one or more of therapeutic hypothermia, pulmonary hypertension, and / or unrepaired Tetralogy of Fallot. In some embodiments, the clinically relevant neurological state comprises one or more of encephalopathy, hypoglycemia, sepsis, hyperammonenia, neonatal stroke, seizure, and / or adverse effect of medication.

[0066] Referring to Block 210, in some embodiments, methods disclosed herein further comprise (a) capturing video data 130 of the subject over a duration of time using one or more cameras positioned to observe the body of the subject.

[0067] In some embodiments, the one or more cameras comprise at least 1, at least 2, at least 3, at least 4, or at least 5 cameras positioned to observe the body of the subject. In someDB2 / 650910081.2 21Attorney Docket No.: 104593-5055-WO embodiments, the one or more cameras comprise no more than 10, no more than 5, no more than 3, or no more than 2 cameras positioned to observe the body of the subject. In some embodiments, the one or more cameras consist of from 1 to 4, from 2 to 7, or from 5 to 10 cameras positioned to observe the body of the subject. In some embodiments, methods disclosed herein utilize one camera per subject, as illustrated in FIG. 3 A. In some embodiments, methods disclosed herein utilize multiple cameras per subject, for instance, to observe different portions or angles of the body of the subject.

[0068] In some embodiments, the one or more cameras are positioned to capture at least two anatomical landmarks, in the plurality of anatomical landmarks, on the body of the subject. In some embodiments, the one or more cameras are positioned to capture at least 1, at least 2, at least 3, at least 4, at least 5, at least 8, at least 10, at least 20, or at least 30 anatomical landmarks. In some embodiments, the one or more cameras are positioned to capture no more than 50, no more than 30, no more than 20, no more than 15, no more than 10, no more than 8, no more than 5, or no more than 3 anatomical landmarks. In some embodiments, the one or more cameras are positioned to capture from 1 to 5, from 2 to 8, from 1 to 10, from 5 to 20, from 12 to 30, or from 25 to 50 anatomical landmarks. In some embodiments, the one or more cameras are positioned to capture another range of anatomical landmarks starting no lower than 1 anatomical landmark and ending no higher than 50 anatomical landmarks.

[0069] In some embodiments, the one or more cameras are positioned to capture all of the body of the subject. In some embodiments, the one or more cameras are positioned to capture at least a portion of the body of the subject (e.g., where a portion of the body of the subject is out of frame of the one or more cameras).

[0070] In some embodiments, the duration of time of video data comprises at least 1 minute, at least 10 minutes, or at least 1 hour of video data of the subject. In some embodiments, the duration of time comprises at least 30 seconds, at least 1 minute, at least 5 minutes, at least 10 minutes, at least 30 minutes, at least 1 hour, at least 2 hours, at least 4 hours, at least 6 hours, at least 12 hours, at least 24 hours, at least 2 days, or at least 1 week of video data of the subject. In some embodiments, the duration of time comprises no more than 2 weeks, no more than 1 week, no more than 2 days, no more than 1 day, no more than 12 hours, no more than 6 hours, no more than 1 hour, no more than 30 minutes, or no more than 10 minutes of video data of the subject. In some embodiments, the duration of time consists of from 30 seconds to 2 hours, from 1 hour to 6 hours, from 1 hour to 12 hours, from 6 hoursDB2 / 650910081.2 22Attorney Docket No.: 104593-5055-WO to 1 day, from 12 hours to 2 days, from 1 day to 1 week, or from 2 days to 2 weeks of video data of the subject. In some embodiments, the duration of time falls within another range starting no lower than 30 seconds and ending no higher than 2 weeks.

[0071] Referring to Block 212, in some embodiments, methods disclosed herein further comprise (b) processing the video data 130 using a pose estimation artificial intelligence model 140 to detect and track a plurality of anatomical landmarks 140 of the subject in successive video frames 132 in the video data.

[0072] As used herein, an anatomical landmark refers to a body part of the subject that can be detected and used as a landmark or reference point, for instance, using a model such as a pose estimation Al model. Referring to Block 214, in some embodiments, the plurality of anatomical landmarks comprise one or more anatomical landmarks selected from the group consisting of: nose, neck, shoulders, elbows, hands, hips, knees, and feet. In some embodiments, the plurality of anatomical landmarks comprise a subset selected from the group consisting of: nose, neck, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hand, right hand, left hip, right hip, left knee, right knee, left ankle, right ankle, left foot, and right foot. Referring to Block 216, in some embodiments, the plurality of anatomical landmarks comprises at least shoulders and feet. In some embodiments, anatomical landmarks include any detectable body part, as will be apparent to one skilled in the art.

[0073] Referring to Block 218, in some embodiments, the plurality of anatomical landmarks consists of from 3 to 8 anatomical landmarks. In some embodiments, the plurality of anatomical landmarks comprises at least 1, at least 2, at least 3, at least 4, at least 5, at least 8, at least 10, at least 20, or at least 30 anatomical landmarks. In some embodiments, the plurality of anatomical landmarks comprises no more than 50, no more than 30, no more than 20, no more than 15, no more than 10, no more than 8, no more than 5, or no more than 3 anatomical landmarks. In some embodiments, the plurality of anatomical landmarks consists of from 1 to 5, from 2 to 8, from 1 to 10, from 5 to 20, from 12 to 30, or from 25 to 50 anatomical landmarks. In some embodiments, the plurality of anatomical landmarks comprises another range of anatomical landmarks starting no lower than 1 anatomical landmark and ending no higher than 50 anatomical landmarks.

[0074] In some embodiments, for one or more video frames of the successive video frames, one or more anatomical landmarks in the plurality of anatomical landmarks areDB2 / 650910081.2 23Attorney Docket No.: 104593-5055-WO occluded or partially occluded. In some embodiments, occlusion of anatomical landmarks occurs due to all or a portion of the infant subject’s body positioned or moved out of frame. Alternately or additionally, in some embodiments, occlusion of anatomical landmarks occurs due to swaddling of the subject. In some embodiments, methods disclosed herein further includes using the pose estimation Al model to generate a prediction that one or more anatomical landmarks are occluded.

[0075] Referring to Block 220, in some embodiments, the pose estimation artificial intelligence model comprises a convolutional neural network or a deep convolutional neural network. In some embodiments, the pose estimation artificial intelligence model comprises DeepLabCut. Alternatively or additionally, in some embodiments, the pose estimation artificial intelligence model is generated, at least in part, by using transfer learning. In some embodiments, the pose estimation artificial intelligence model comprises one or more parameters, in a plurality of parameters, that reflect training from a convolutional neural network or deep convolutional neural network. In some embodiments, the one or more parameters in the plurality of parameters reflect training from ResNet. In some embodiments, the pose estimation artificial intelligence model comprises ViTPose, which leverages Vision Transformers (ViT) for pose estimation by capturing global dependencies in images. In some embodiments, the pose estimation artificial intelligence model includes, but is not limited to, any one or more of: DeepLabCut, ViTPose, OpenPose, AlphaPose, HRNet (High-Resolution Network), PoseNet, MediaPipe Pose, LEAP, SLEAP, Detectron2, DensePose, SimpleBaseline, DarkPose, SPM (Single-Person Mesh), Lite-HRNet, MMPose, TPNet (Temporal Pose Network), EfficientPose, PRTR (Pose Regression Transformer), and / or Keypoint R-CNN.

[0076] In some embodiments, the pose estimation artificial intelligence model comprises any one or more of the models disclosed above.

[0077] Referring to Block 222, in some embodiments, the pose estimation artificial intelligence model generates, for each respective video frame in the successive video frames, a corresponding vectorized pose for the subject. In some embodiments, the corresponding vectorized pose comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, corresponding X and Y coordinates for the respective anatomical landmark in the respective video frame. Referring to Block 224, in some embodiments, the corresponding vectorized pose further comprises, for each respective anatomical landmark inDB2 / 650910081.2 24Attorney Docket No.: 104593-5055-WO the plurality of anatomical landmarks, a corresponding likelihood for the respective anatomical landmark in the respective video frame.

[0078] Referring to Block 226, in some embodiments, methods disclosed herein further comprise (c) computing a movement index 154 for a defined time interval 152 based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks 140 across the interval within the video data 130. In some embodiments, the movement index is normalized. In some embodiments, normalization of the movement index comprises normalization to account for body size and / or camera distance.

[0079] Referring to Block 228, in some embodiments, methods disclosed herein further comprise determining the movement index using, for each anatomical landmark in the plurality of anatomical landmarks, X and Y coordinates of the respective anatomical landmark across a subset of successive video frames during the defined time interval. Referring to Block 230, in some embodiments, the movement index comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a variance of movement derived from the X and Y coordinates of the respective anatomical landmark across the subset of successive video frames during the defined time interval.

[0080] In some embodiments, the movement index is determined using another statistical measure for the X and Y coordinates. In some embodiments, the statistical measure comprises a standard deviation, a standard error, a range, a mean, a median, and / or a mode. In some embodiments, any suitable statistical measure for determining movement is contemplated for use in the present disclosure, as will be apparent to one skilled in the art.

[0081] Referring to Block 232, in some embodiments, the defined time interval comprises at least 30 seconds, at least 1 minute, or at least 2 minutes. In some embodiments, the defined time interval is from 30 seconds to 2 minutes.

[0082] In some embodiments, the defined time interval comprises at least 10 seconds, at least 20 seconds, at least 30 seconds, at least 1 minute, at least 2 minutes, or at least 5 minutes. In some embodiments, the defined time interval comprises no more than 10 minutes, no more than 5 minutes, no more than 2 minutes, no more than 1 minute, or no more than 30 seconds. In some embodiments, the defined time interval consists of from 10 seconds to 30 seconds, from 20 seconds to 1 minute, from 30 seconds to 2 minutes, from 1 minute to 5 minutes, or from 5 minutes to 10 minutes. In some embodiments, the definedDB2 / 650910081.2 25Attorney Docket No.: 104593-5055-WO time interval falls within another range starting no lower than 10 seconds and ending no higher than 10 minutes.

[0083] In some embodiments, the defined time interval corresponds to a respective subset of successive video frames, or a portion thereof, in the video data captured during the respective portion of the duration of time (e.g., the subset of successive video frames captured during the respective time interval is used for infant pose estimation and / or further analysis).

[0084] In some embodiments, the normalization of the movement index uses the size of one or more body parts of the subject (e.g., anatomical landmarks). In some embodiments, the normalization uses the size of the subject’s head as a normalization factor. In some embodiments, the normalization of the movement index uses a number of pixels or a ratio of pixels relative to all or a portion of the subject’s body. In some embodiments, the normalization uses a number of pixels corresponding to a size of one or more body parts of the subject, or a portion thereof (e.g., an infant’s head or a portion thereof).

[0085] In some embodiments, the normalization of the movement index accounts for subject body size and / or camera distance by using a distance between two or more anatomical landmarks of the subject. Referring to Block 234, in some embodiments, the normalization of the movement index accounts for subject body size and camera distance by using the distance between a nose landmark and a neck landmark of the subject.

[0086] Referring to Block 236, in some embodiments, computing the movement index further comprises computing, for each respective defined time interval in a plurality of defined time intervals, a corresponding movement index in the position of the tracked plurality of anatomical landmarks across the defined time interval within the video data. In some embodiments, as described above, the corresponding movement index of each respective defined time interval is normalized to account for body size and camera distance. In some embodiments, referring again to Block 236, computing the movement index further includes determining a measure of central tendency for a plurality of movement indices across the plurality of defined intervals.

[0087] Referring to Block 238, in some embodiments, each respective defined time interval in the plurality of defined time intervals (i) comprises a respective portion of the duration of time for capturing video data and (ii) corresponds to a respective subset of successive video frames in the video data captured during the respective portion of theDB2 / 650910081.2 26Attorney Docket No.: 104593-5055-WO duration of time. In some embodiments, each defined time interval in the plurality of defined time intervals comprises an amount of time as disclosed above. In some embodiments, each respective defined time interval in the plurality of defined time intervals comprises an equal amount of time. In some embodiments, two or more defined time intervals in the plurality of defined time intervals comprises different amounts of time.

[0088] In some embodiments, one or more time intervals in the duration of time for capturing video data are removed from the video data. In some embodiments, one or more video frames in a respective subset of successive video frames are removed from the video data (e.g., for frames including occluded anatomical landmarks or where all or a portion of the infant subject is positioned out of frame).

[0089] Referring to Block 240, in some embodiments, the measure of central tendency for the movement index comprises a median of a plurality of movement variances, wherein each respective movement variant in the plurality of movement variances is determined within a 1 -minute time interval for each respective anatomical landmark in the plurality of anatomical landmarks. For instance, as described in Examples 1-6 below, in some embodiments, the movement index comprises movement variances calculated within 1- minute intervals per anatomic landmark, for which a median movement variance was obtained.

[0090] In some embodiments, the measure of central tendency comprises a mean, a median, a mode, a weighted mean, a weighted median, a weighted mode, a range, a distribution, and / or any other suitable statistical measure, as will be apparent to one skilled in the art.

[0091] Referring to Block 241, in some embodiments, methods disclosed herein further comprise (d) inputting the movement index 154 to a trained machine learning model 160, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, wherein the neurologic condition represents a clinically relevant neurological state.

[0092] Referring to Block 242, in some embodiments, the trained machine learning model comprises a logistic regression model, a support vector machine, or XGBoost. In some embodiments, the trained machine learning model comprises any one or more of the models disclosed above.

[0093] In some embodiments, the trained machine learning model comprises a plurality of parameters that reflects a plurality of training data for a plurality of subjects, the pluralityDB2 / 650910081.2 27Attorney Docket No.: 104593-5055-WO of training data comprising, for each respective subject in the plurality of subjects, (i) for each respective defined time interval in a plurality of defined time intervals, a corresponding variance in the position of a tracked plurality of anatomical landmarks across the defined time interval within the video data, and (ii) a respective indication of a neurologic condition in the respective subject. In some embodiments, the movement index is normalized (e.g., to account for body size and / or camera distance).

[0094] In some embodiments, the respective indication of the neurologic condition is determined using electroencephalography. Alternatively or additionally, in some embodiments, the respective indication of the neurologic condition is obtained using a diagnosis from a clinician or medical practitioner, such as an epileptologist.

[0095] Referring to Block 244, in some embodiments, the trained machine learning model comprises a plurality of parameters. In some embodiments, the plurality of parameters comprises at least 10, at least 1000, at least 10,000, or at least 1 x 106parameters.

[0096] In some embodiments, the plurality of parameters includes at least 10, at least 50, at least 100, at least 500, at least 1000, at least 2000, at least 5000, at least 10,000, at least 20,000, at least 50,000, at least 100,000, at least 200,000, at least 500,000, at least 1 million, at least 2 million, at least 3 million, at least 4 million, at least 5 million, at least 10 million, at least 50 million, or at least 100 million parameters. In some embodiments, the plurality of parameters includes no more than 1 billion, no more than 100 million, no more than 50 million, no more than 10 million, no more than 5 million, no more than 4 million, no more than 1 million, no more than 500,000, no more than 100,000, no more than 50,000, no more than 10,000, no more than 5000, no more than 1000, or no more than 500 parameters. In some embodiments, the plurality of parameters consists of from 10 to 5000, from 500 to 10,000, from 10,000 to 500,000, from 20,000 to 1 million, from 1 million to 10 million, from 2 million to 50 million, from 10 million to 100 million, or from 100 million to 1 billion parameters. In some embodiments, the plurality of parameters falls within another range starting no lower than 10 parameters and ending no higher than 1 billion parameters.

[0097] As described above, referring to Block 246, in some embodiments, the neurologic condition comprises sedation or cerebral dysfunction. Referring to Block 248, in some embodiments, the clinically relevant neurological state comprises sedation management in the settings of therapeutic hypothermia, pulmonary hypertension, and / or unrepaired Tetralogy of Fallot. Referring to Block 250, in some embodiments, the clinically relevant neurologicalDB2 / 650910081.2 28Attorney Docket No.: 104593-5055-WO state comprises encephalopathy, hypoglycemia, sepsis, hyperammonenia, neonatal stroke, seizure, or adverse effect of medication. Referring to Block 252, in some embodiments, the encephalopathy comprises hypoxic ischemic encephalopathy.

[0098] Alternatively or additionally, in some embodiments, the clinically relevant neurological state comprises sepsis, seizures before loading ASMs, and / or withdrawal.

[0099] Referring to Block 253, in some embodiments, methods disclosed herein further comprise: performing continuous neurological monitoring of the subject by repeating the (a) capturing, (b) processing, (c) computing, and (d) inputting over a period of time, thereby obtaining an updated prediction of the neurologic condition in the subject.

[0100] In some embodiments, the period of time comprises at least 1 hour, at least 6 hours, at least 12 hours, or at least 24 hours. In some embodiments, the period of time comprises at least 5 minutes, at least 10 minutes, at least 30 minutes, at least 1 hour, at least 2 hours, at least 4 hours, at least 6 hours, at least 12 hours, at least 24 hours, at least 2 days, or at least 1 week. In some embodiments, the period of time comprises no more than 2 weeks, no more than 1 week, no more than 2 days, no more than 1 day, no more than 12 hours, no more than 6 hours, no more than 1 hour, no more than 30 minutes, or no more than 10 minutes. In some embodiments, the period of time consists of from 5 minutes to 2 hours, from 1 hour to 6 hours, from 1 hour to 12 hours, from 6 hours to 1 day, from 12 hours to 2 days, from 1 day to 1 week, or from 2 days to 2 weeks. In some embodiments, the period of time falls within another range starting no lower than 5 minutes and ending no higher than 2 weeks.

[0101] In some embodiments, the method is repeated at least 1, at least 2, at least 3, at least 5, at least 10, at least 20, at least 50, at least 100, or at least 500 times. In some embodiments, the method is repeated no more than 1000, no more than 500, no more than 100, no more than 50, no more than 10, or no more than 5 times. In some embodiments, the method is repeated from 1 to 10, from 8 to 20, from 10 to 100, or from 100 to 1000 times. In some embodiments, the method comprises another range of repeats starting no lower than 1 repeat and ending no higher than 1000 repeats.

[0102] Referring to Block 254, in some embodiments, methods disclosed herein further comprise providing a report comprising at least the prediction of the neurologic condition in the subject and / or the clinically relevant neurological state. In some embodiments, the report further comprises visual or textual annotations indicating the portions of the input data thatDB2 / 650910081.2 29Attorney Docket No.: 104593-5055-WO contributed most significantly to the classifier’s output, thereby enabling interpretability and clinical decision support. In some embodiments, the report is formatted for integration with an electronic health record (EHR) system and / or includes user interface elements or structured fields to allow clinician review, comment, and validation. In some embodiments, the report includes recommendations for follow-up diagnostics or therapeutic interventions, which are generated based at least in part on the machine learning model’s output and / or predefined clinical protocols.

[0103] In some embodiments, methods disclosed herein further comprise transmitting the report. In some embodiments, systems disclosed herein are configured to transmit the generated report, including at least the prediction of the neurologic condition in the subject and / or the clinically relevant neurological state, to one or more designated medical providers or clinical systems. In some embodiments, transmission is performed via network communication module described with reference to FIGS. 1 A-B. In some embodiments, the report is automatically uploaded to an electronic health record (EHR) system associated with the subject’s medical profile, where it becomes accessible to authorized clinicians for review and clinical decision-making. In some embodiments, the report is delivered in real-time to a clinician’s workstation, mobile device, or centralized monitoring system, enabling timely intervention and response. In some embodiments, the system includes notification mechanisms, such as automated alerts or push notifications, triggered when a prediction by the machine learning model exceeds predefined thresholds or indicate a critical condition.

[0104] Referring to Block 256, in some embodiments, methods disclosed herein further comprise diagnosing the subject with the clinically relevant neurological state.

[0105] In some embodiments, methods disclosed herein further comprise detecting one or more abnormalities in a mental state or neurologic condition of the subject. In some embodiments, methods disclosed herein further comprise detecting tolerance to one or more medications (e.g., sedative medications and / or anti-seizure medications).

[0106] Referring to Block 258, in some embodiments, methods disclosed herein further comprise adjusting a dosage or schedule of a medication based on the prediction of the neurologic condition.

[0107] In some embodiments, adjusting a medication includes modifying one or more treatment parameters including, but not limited to, dose amount, dosing frequency, timing of administration, delivery route, and / or duration of therapy. In some embodiments, a currentDB2 / 650910081.2 30Attorney Docket No.: 104593-5055-WO dosing regimen is increased in amount, increased in frequency, decreased in amount, decreased in frequency, delayed, and / or ceased. In some embodiments, adjusting a medication further considers additional factors such as weight, age, organ function, drug levels, or presence of adverse events. In some embodiments, the adjusted dosage or schedule is implemented automatically or provided as a recommendation for clinician approval.

[0108] Referring to Block 260, in some embodiments, methods disclosed herein further comprise determining a state of oversedation based on the prediction of the neurologic condition, and, responsive to determining the state of oversedation, reducing an amount of a medication administered to the subject.

[0109] Referring to Block 262, in some embodiments, methods disclosed herein further comprise determining a state of undersedation based on the prediction of the neurologic condition, and, responsive to determining the state of undersedation, increasing an amount of a medication administered to the subject.

[0110] Referring to Block 264, in some embodiments, the medication is a sedative medication or an anti-seizure medication.

[0111] In some embodiments, the medication comprises any of the sedative and / or antiseizure medications disclosed above (e.g., phenobarbital, pentobarbital, levetiracetam, fosphenytoin, midazolam, lorazepam, and / or diazepam). In some embodiments, the medication comprises levetiracetam, phenobarbital, or pentobarbital.

[0112] Referring to Block 265, in some embodiments, methods disclosed herein further comprise validating the neurologic condition, where the neurologic condition is cerebral dysfunction, using electroencephalography for the subject. As described above, in some embodiments, the validation of the neurologic condition is obtained using a diagnosis from a clinician or medical practitioner, such as an epileptologist.

[0113] Referring to Block 266, in some embodiments, methods disclosed herein further comprise displaying a real-time graphical representation of the movement index over time for use by clinical personnel.

[0114] Advantageously, the systems and methods disclosed herein e.g., using a pose estimation artificial intelligence model and / or a trained machine learning model or classifier) can accurately track infant movement and predict sedation and cerebral dysfunction in critically ill infants in their “natural” clinical setting. In some embodiments, the disclosed systems and methods allow for neurologic monitoring akin to cardiorespiratory telemetry. InDB2 / 650910081.2 31Attorney Docket No.: 104593-5055-WO an example implementation, one or more cameras are facing an infant’s incubator to record and store video data. Pose Al (e.g., including the pose estimation artificial intelligence model and / or the trained machine learning model or classifier) then provides report in the form of a neuro-telemetry strip with anatomic landmark position and algorithmic predictions including level of sedation and cerebral dysfunction. If the algorithm detects abnormalities or there is a clinical concern (e.g., tolerance to sedative medications), then the stored video footage, pose tracking, and relevant predictions can be reviewed. Beneficially, this system has high interpretability for bedside providers because reviewing videos is intuitive. Such interpretability is not the case with other conventional infant motion sensing technology (e.g., wearables, mattresses, radar, or waveform artifacts). In some embodiments, the methods and systems disclosed herein can be applied to other clinical conditions, including but not limited to arrhythmias and apnea-bradycardia events.

[0115] Advantageously, there is broad applicability of the presently disclosed systems and methods because infants are routinely sedated. Common reasons for sedation are mechanical ventilation, bedside procedures, imaging, therapeutic hypothermia, extracorporeal membrane oxygenation, and clinical pathology such as pulmonary hypertension and unrepaired Tetralogy of Fallot. In some cases, oversedation leads to prolonged mechanical ventilation, brain injury, and drug withdrawal syndromes. In other cases, undersedation is associated with pain, pulmonary hypertensive crises, and adverse events including repeat imaging, catheter displacement, and unplanned extubation. Pose tracking has potential to address these sedation complications through more precise titration (e.g., by adjusting the dosage or schedule of a medication).

[0116] Infants are also frequently evaluated for encephalopathy. Cerebral dysfunction is an EEG finding that can reflect encephalopathy, which can be iatrogenic or due to an unknown clinical pathology. For example, cerebral dysfunction can be due to medications, hypoglycemia, sepsis, hyperammonemia, and / or permanent injury after neonatal stroke or seizures. Pose tracking allows for prediction of cerebral dysfunction in infants, thereby providing a practical application by facilitating the identification and monitoring of encephalopathy.

[0117] Other advantages include the use of explainable features (e.g., movement indices of anatomic landmarks) and models (e.g., a pose estimation Al model and / or a machine learning model for prediction) trained on a large, diverse patient population (e.g., 25.2% white non-Hispanic, 46% female in Examples 1-6), addressing important concerns about biasDB2 / 650910081.2 32Attorney Docket No.: 104593-5055-WO and interpretability with deploying Al. Challenges in conventional methodology include that pose Al trained on adults has worse performance on infants, which the presently disclosed systems and methods overcome by providing models trained on patients less than one year old. As shown in Examples 1-6, pose Al and neurologic predictions had high accuracy, including on held-out test datasets. This is in contrast to prior studies, which were small (N<30) and did not associate the learned infant pose with neurologic changes in the NICU. One study correlated pose with infant neuromotor risk, but this study was also small (19 infants, 9.7 hours of video) and placed infants in a bespoke setting outside the NICU. Moreover, Examples 1-6 illustrate that variance in movement is related to neurologic status, analogous to the relationship between heart rate variability and autonomic status. Another challenge when generating predictions for infant pose and movement include infant occlusion due to swaddling. Advantageously, the presently disclosed systems and methods provide accurate movement prediction using only 47% of video data. The presently disclosed systems and methods thus provide further practical applications by facilitating continuous pose Al monitoring in the NICU in a minimally invasive, interpretable, and scalable manner.

[0118] Yet another aspect of the present disclosure provides a system for predicting a neurologic condition in a subject, the system comprising: a camera configured to capture video data of a body of a subject over a duration of time, where the subject is a human subject less than one year old; and a processor operatively coupled to the camera and configured to perform a method comprising: processing the video data using a pose estimation artificial intelligence model to detect and track a plurality of anatomical landmarks of the subject in successive video frames in the video data; computing a movement index for a defined time interval based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks across the interval within the video data, where the movement index is normalized to account for body size and camera distance; and inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, where the neurologic condition represents a clinically relevant neurological state.

[0119] In some embodiments, the system comprises one or more cameras, as described above. In some embodiments, the method is performed using system 100, as described above.

[0120] In some embodiments, the camera is positioned to capture at least two anatomical landmarks, in the plurality of anatomical landmarks, on the body of the subject, where theDB2 / 650910081.2 33Attorney Docket No.: 104593-5055-WO plurality of anatomical landmarks are selected from the group consisting of: nose, neck, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hand, right hand, left hip, right hip, left knee, right knee, left ankle, right ankle, left foot, and right foot. In some embodiments, the camera is positioned to capture at least one or both shoulders and one or both feet of the subject.

[0121] In some embodiments, the neurologic condition comprises sedation or cerebral dysfunction. In some embodiments, the clinically relevant neurological state comprises sedation management in the settings of therapeutic hypothermia, pulmonary hypertension, and / or unrepaired Tetralogy of Fallot. In some embodiments, the clinically relevant neurological state comprises encephalopathy, hypoglycemia, sepsis, hyperammonenia, neonatal stroke, seizure, and / or adverse effect of medication. In some embodiments, the encephalopathy comprises hypoxic ischemic encephalopathy.

[0122] In some embodiments, the subject has a postmenstrual age of 64 weeks or less. In some embodiments, the subject has a postmenstrual age of 48 weeks or less. In some embodiments, the subject has been administered with a sedative mediation or an anti-seizure medication. In some embodiments, the anti-seizure medication comprises levetiracetam, phenobarbital, or pentobarbital.

[0123] In some embodiments, the duration of time of video data comprises at least 1 minute, at least 10 minutes, or at least 1 hour of video data of the subject. In some embodiments, the plurality of anatomical landmarks comprise a subset selected from the group consisting of: nose, neck, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hand, right hand, left hip, right hip, left knee, right knee, left ankle, right ankle, left foot, and right foot. In some embodiments, the plurality of anatomical landmarks comprise one or more anatomical landmarks selected from the group consisting of: nose, neck, shoulders, elbows, hands, hips, knees, and feet. In some embodiments, the plurality of anatomical landmarks comprises at least shoulders and feet.

[0124] In some embodiments, for one or more video frames of the successive video frames, one or more anatomical landmarks in the plurality of anatomical landmarks are occluded. In some embodiments, the plurality of anatomical landmarks comprises from 3 to 8 anatomical landmarks.DB2 / 650910081.2 34Attorney Docket No.: 104593-5055-WO

[0125] In some embodiments, the pose estimation artificial intelligence model comprises a convolutional neural network or a deep convolutional neural network. In some embodiments, the pose estimation artificial intelligence model comprises DeepLabCut.

[0126] In some embodiments, the pose estimation artificial intelligence model generates, for each respective video frame in the successive video frames, a corresponding vectorized pose for the subject, and the corresponding vectorized pose comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, corresponding X and Y coordinates for the respective anatomical landmark in the respective video frame. In some embodiments, the corresponding vectorized pose further comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a corresponding likelihood for the respective anatomical landmark in the respective video frame.

[0127] In some embodiments, the method further includes determining the movement index using, for each anatomical landmark in the plurality of anatomical landmarks, X and Y coordinates of the respective anatomical landmark across a subset of successive video frames during the defined time interval. In some embodiments, the movement index comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a variance of movement derived from the X and Y coordinates of the respective anatomical landmark across the subset of successive video frames during the defined time interval.

[0128] In some embodiments, the defined time interval comprises at least 30 seconds, at least 1 minute, or at least 2 minutes. In some embodiments, the defined time interval is from 30 seconds to 2 minutes.

[0129] In some embodiments, computing the movement index further comprises: computing, for each respective defined time interval in a plurality of defined time intervals, a corresponding movement index in the position of the tracked plurality of anatomical landmarks across the defined time interval within the video data, where the movement index is normalized to account for body size and camera distance; and determining a measure of central tendency for the movement index across the plurality of defined intervals. In some embodiments, each respective defined time interval in the plurality of defined time intervals (i) comprises a respective portion of the duration of time for capturing video data and (ii) corresponds to a respective subset of successive video frames in the video data captured during the respective portion of the duration of time. In some embodiments, each respective defined time interval in the plurality of defined time intervals comprises an equal amount ofDB2 / 650910081.2 35Attorney Docket No.: 104593-5055-WO time. In some embodiments, the measure of central tendency for the movement index comprises a median of a plurality of movement variances, where each respective movement variant in the plurality of movement variances is determined within a 1 -minute time interval for each respective anatomical landmark in the plurality of anatomical landmarks.

[0130] In some embodiments, the normalization of the movement index accounts for subject body size and camera distance by using the distance between a nose landmark and a neck landmark of the subject.

[0131] In some embodiments, the trained machine learning model comprises a logistic regression model, a support vector machine, or XGBoost. In some embodiments, the trained machine learning model comprises a plurality of parameters that reflects a plurality of training data for a plurality of subjects, the plurality of training data comprising: for each respective subject in the plurality of subjects, (i) for each respective defined time interval in a plurality of defined time intervals, a corresponding variance in the position of a tracked plurality of anatomical landmarks across the defined time interval within the video data, where the movement index is normalized to account for body size and camera distance, and (ii) a respective indication of a neurologic condition in the respective subject. In some embodiments, the respective indication of the neurologic condition is determined using electroencephalography. In some embodiments, the plurality of parameters comprises at least 10, at least 1000, at least 10,000, or at least 1 x 106parameters.

[0132] In some embodiments, the method further comprises: performing continuous neurological monitoring of the subject by repeating the (a) capturing, (b) processing, (c) computing, and (d) inputting over a period of time, thereby obtaining an updated prediction of the neurologic condition in the subject. In some embodiments, the period of time comprises at least 1 hour, at least 6 hours, at least 12 hours, or at least 24 hours.

[0133] In some embodiments, the method further comprises providing a report comprising at least the prediction of the neurologic condition in the subject and the clinically relevant neurological state.

[0134] Yet another aspect of the present disclosure provides a non-transitory computer- readable medium storing instructions that, when executed by one or more processors, cause the system to perform a method for predicting a neurologic condition, the method comprising: receiving video data of a subject captured by a camera, where the subject is a human subject less than one year old; applying a pose estimation artificial intelligence modelDB2 / 650910081.2 36Attorney Docket No.: 104593-5055-WO to the video data to detect and track a plurality of anatomical landmarks of the subject across a sequence of video frames; computing a movement index for a defined time interval based on a statistical variance in the positions of the anatomical landmarks, where the movement index is normalized for patient body size and camera distance; and inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, wherein the neurologic condition represents a clinically relevant neurological state.

[0135] Yet another aspect of the present disclosure provides a system for predicting a neurologic condition in a subject, the system comprising: one or more cameras configured to capture video data of a body of a subject over a duration of time, where the subject is a human subject less than one year old; and a processor operatively coupled to the camera and configured to perform any of the methods disclosed herein.

[0136] Yet another aspect of the present disclosure provides a non-transitory computer- readable medium storing instructions that, when executed by one or more processors, cause the system to perform any of the methods disclosed herein.

[0137] EXAMPLES

[0138] Example 1 — Workflow for application of pose Al to critically ill infants.

[0139] Pose Al was applied to a large dataset including 4,705 video-EEG hours for 115 infants and used to accurately predict sedation and cerebral dysfunction in critically ill infants. FIGS. 3 A-C collectively illustrate an example schematic for the application of pose Al to critically ill infants, in accordance with some embodiments of the present disclosure.

[0140] Referring to FIG. 3 A, video data was captured for infants over time using one or more cameras positioned to observe the bodies of the infants. A large database of video-EEG data was built using this video data (N = 115 infants, 4,705 hours of video, 10.4 Tb) and stored on a HIPAA-compliant supercomputing cluster. Referring to FIG. 3B, the video data was processed using a pose Al model to detect and track a plurality of anatomical landmarks of the infants in successive video frames in the video data. In particular, a pose Al algorithm was trained and tested on 2712 manually labelled video frames with a deep neural network (DNN) (e.g., DeepLabCut) to predict infant anatomic landmarks. Referring to FIG. 3C, movement indices for defined time intervals were computed based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks across the intervalsDB2 / 650910081.2 37Attorney Docket No.: 104593-5055-WO within the video data and a machine learning model was used to determine one or more clinically relevant neurological states for the infants. In particular, the clinical utility of quantifying movement was demonstrated by using infant movement features to predict sedation and cerebral dysfunction.

[0141] Examples 2-6 describe specific implementations and results for each of the processes illustrated by FIGS. 3A-C in further detail.

[0142] Example 2 — Generating database of infant video data for pose Al.

[0143] Cohort characteristics. Video-EEG and clinical data were collected from February 2021 to December 2022, during which data from 115 individuals met inclusion criteria. To train algorithms generalizable across race, sex, and gestational age, broad inclusion criteria were used and data was obtained from a racially and ethnically diverse patient population. Data from all infants who had video-EEG data and chronologic age <1 year at the start of recording were included. All video-EEG and electronic health record data from an Epilepsy Monitoring Unit were transferred over an encrypted connection directly to a HIPAA-compliant supercomputing cluster, where they were stored. The full data workflow, which follows Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines, is shown in FIG. 4.

[0144] Inclusion criteria are provided in Table 1. Caregiver-reported race and ethnicity were obtained retrospectively from the electronic health record. Consistent with the routine use of video-EEG in therapeutic hypothermia, most infants were <1 month old (67%) and the most common underlying suspected / known pathology was hypoxic ischemic encephalopathy (21%). Both levetiracetam and phenobarbital were the most common first-line antiseizure medications (ASMs).DB2 / 650910081.2 38Attorney Docket No.: 104593-5055-WO

[0145] Table 1. Clinical characteristics of all patients included in the study.*A11 medications acting on the central nervous system (CNS) that were used at any point while the patient was undergoing the video-EEG study. f ASMs = antiseizure medications f f Other ASMs were lacosamide, oxcarbazepine, topiramate, and vigabatrin

[0146] Clinical outcomes. The database was further used to evaluate the utility and predictive capability of pose Al on two clinical outcomes, sedation and cerebral dysfunction, as described in Examples 4-6, below. Phenobarbital and sedative infusions (e.g., midazolam, dexmedetomidine and fentanyl) were classified as sedating. Levetiracetam and less frequently used antiseizure medications (ASMs), such as phenytoin and oxcarbazepine, were not considered sedating for these Examples. The observed ASM combinations areDB2 / 650910081.2 39Attorney Docket No.: 104593-5055-WO enumerated in Table 1. Cerebral dysfunction or encephalopathy was diagnosed from EEG readings by an epileptologist separately for each calendar day. The diagnosis of cerebral dysfunction and encephalopathy was based on focal or generalized abnormalities of background EEG activity, such as reactivity, synchrony, discontinuities, and slowing.

[0147] Example 3 - Training and evaluation of pose Al for infant pose recognition.

[0148] Pose Al development and training. Conventional methods for determining human pose from video data are generally trained on adults and have poor performance on infants, likely because infant body proportions are different from adults. A pose estimation artificial intelligence model (DeepLabCut) was selected was selected for its generalizability to many animals, including humans, and robots. DeepLabCut uses transfer learning to leverage ResNet, which is trained on >1 million images, for pose tracking.

[0149] To train DeepLabCut to recognize poses of subjects less than 1 year old (e.g., infants), 25 frames per infant were randomly sampled from the database described in Example 2 and these frames were labeled up to 14 landmarks per frame. Frames were excluded if all body parts were occluded, if the infant was out of frame, or if subsets of the video were incompatible with DeepLabCut, leaving 2712 frames from 109 infants for model training and evaluation. To test generalizability and prevent data leakage, 10% of infants (N = 239 frames, 10 infants) were randomly excluded from training and 5% of frames (N = 122 frames) from infants used in training were randomly excluded from training. DeepLabCut was trained on 2,351 frames from 99 infants. Performance was evaluated within using the set of held-out frames from the infants used for training, and the set of infants not used for training, with repeated measures k-fold cross-validation. The trained model included a plurality of parameters (e.g., model weights) by which any infant video could be converted to vectorized pose, which consisted of X and Y coordinates and a likelihood for each anatomic landmark in each frame.

[0150] Three evaluation metrics were used to assess the performance of the infant DeepLabCut model. First, L2 pixel errors were generated by measuring the Euclidean distance between predicted body part coordinates and their manually labeled anatomic landmarks. Second, area under receiver operating characteristic (AUROC) curves were generated based on the model’s ability to correctly predict whether an anatomic landmark was occluded using DeepLabCut’ s built-in confidence score. Finally, a standard pose recognition evaluation metric, percentage of correct key points (PCK), was used. The PCK isDB2 / 650910081.2 40Attorney Docket No.: 104593-5055-WO the percent of body part coordinates within 26.2 pixels of a landmark’s known position, where 26.2 pixels is on average half the height of an infant’s head in the plurality of video frames in the database described in Example 2.

[0151] Pose Al performance evaluation. A low L2 pixel error (median L2train = 3.2,L2test-new-frames = 3.5, L2test-new-infants = 4.6 pixels) comparable to prior work was observed. See, e.g., Mathis etal., “DeepLabCut: markerless pose estimation of user-defined body parts with deep learning,” Nat. Neurosci. 21, 1281-1289 (2018), which is hereby incorporated herein by reference in its entirety. This was consistent across all anatomic landmarks, as illustrated in FIG. 5 A, including all labeled non-occluded anatomic landmarks for training data (“Training,” median 3.2 pixels, N = 18,399), frames held out from training (“New Frames,” median 3.5 pixels, N = 797), and frames from infants held out from training (“New Babies,” median 4.6 pixels, N = 973). Thirteen labeled landmarks (0.06%) had an L2 error of >50 pixels and their errors are shown with arrows.

[0152] Further illustrating excellent model performance, high ROC-AUCs >0.83 for landmark occlusion were observed, as illustrated in FIG. 5B and >95% of key point predictions (PCK) were observed within a reference distance threshold of 26.2 pixels, or half the size of the head, for training and test data, as illustrated in FIG. 5C. Finally, exemplary video frames during a seizure in a neonate with Y WO- -related epilepsy syndrome showed predicted landmarks (circles) overlapping expected locations, as illustrated in FIG. 5D.

[0153] The trained pose recognition algorithm was then used to generate pose X and Y coordinates for all infant video data from all patients (N = 4,705 hours from N = 115 infants). This included videos from six additional infants not initially available for training or evaluation, bringing the total from 109 to 115 infants for all downstream analyses.

[0154] Example 4 — Association between infant movement, predicted by pose Al, and neurologic changes.

[0155] To evaluate the utility of pose recognition in a typical intensive care unit, movement variance was used as an intuitive metric of infant movement that would reflect major clinical features including postmenstrual age (PMA), cerebral dysfunction, and use of sedative medications, as obtained using the clinical outcomes of the infant database described in Example 2.

[0156] Statistical analysis. Movement variance was determined by first developing a summary metric of infant neurologic status. To quantify baseline movement, variance forDB2 / 650910081.2 41Attorney Docket No.: 104593-5055-WO each X and Y position was calculated for each body part per minute. Variance was divided by median nose-to-neck distance within each interval to adjust for infant size and camera position. Variance was chosen because it summarized frequency and amplitude of motion while maintaining robustness to outliers. In some embodiments, kinematic calculations can be used; however, kinematics can be challenging to calculate from two-dimensional videos across the wide variety of camera positions and combination of occluded anatomic landmarks seen in video-EEG data. Variance was also chosen for its demonstrated clinical relevance with infant mental status: a metric related to variance in infant pose, standard deviation derived from ECG-motion artifacts, was previously shown to predict lethargy in late onset sepsis. Movement variance was calculated within 1 -minute intervals per anatomic landmark and then the median was taken across all landmarks. One-minute movement intervals were then compared between different groups (e.g., sedation vs. no sedation) using a resampling procedure to account for repeated measures (e.g., permutation testing). A p-value less than 0.05 was considered statistically significant.

[0157] Next, whether movement decreased in infants receiving sedative medications and in those with cerebral dysfunction was evaluated. These hypothesis tests were restricted to neonates with PMA <44 weeks because the relationship between age and movement was non-linear and possibly dependent on underlying neuropathology. The proportion of videos with sufficient information to calculate variance per 1 -minute intervals for at least 7 body parts was 47% (range 0-100% per infant).

[0158] Movement increased with corrected age as illustrated in FIGS. 6A-B in infants with EEG abnormalities or sedative medications (47-fold increase between lowest and highest age group, permutation P = 1 x 10'4, 10,000 permutations) and in those without (15- fold increase between lowest and highest age groups, permutation P = 2.4 x 1 O’3). Moreover, decreased movement was observed in infants with cerebral dysfunction (FIGS. 6A-B, P = 1 x 1 O’4), phenobarbital (P = 3.6 x 1 O’3), both (P = 1 x 1 O’4), and those receiving sedative infusions (P = 1 x 1 O’4). Taken together, the data show that movement increased with age and decreased with sedative medications and with cerebral dysfunction, decreasing more so with their combined effects.

[0159] Example 5 — Pose Al predicts sedation and cerebral dysfunction.

[0160] The movement variance used in the prior section was used to validate pose Al for clinical applicability but did not account for the relative importance of different anatomicDB2 / 650910081.2 42Attorney Docket No.: 104593-5055-WO landmarks or their relationship. To overcome this limitation, prediction models for sedation and cerebral dysfunction were developed, as shown in the example schematic illustrated in in FIG. 4.

[0161] Machine learning. The 1 -minute movement intervals described in Example 4, above, were used to develop classifiers for sedation and cerebral dysfunction. To improve the quality of input data and thereby improve machine learning performance, the following filters were implemented as shown in FIG. 4: postmenstrual age (PM A) <44 weeks to increase homogeneity, >7 body parts visible per 1 -minute interval to ensure sufficient visibility, variance calculations based on >30 seconds per 1-minute interval to ensure stable variance estimates for each anatomic landmark, and >60 minutes of usable footage per infant to ensure sufficient sample size per infant. The filtered dataset used for developing sedation and cerebral dysfunction classifiers thus comprised 118,823 minutes from 63 infants (ranging from 65 to 14,409 minutes per infant).

[0162] Three models were trained to predict sedation, including logistic regression, support vector machines, and XGBoost. XGBoost hyperparameters were fine-tuned via 5,000 simulations of random search and hyperparameters were identified through repeated k- fold cross-validation. For each iteration of random search, the values for 12 hyperparameters were randomly assigned from a prespecified range and performance was calculated as the mean Fl score from three repeats of five-fold cross-validation. A macro-averaged Fl score was used to account for potential class imbalances.

[0163] A final classifier (e.g., a trained machine learning model) was selected based on multiple metrics, including Fl score, accuracy, AUROC, and precision-recall receiver operating characteristic (ROC). XGBoost had substantially better performance compared to logistic regression and support vector machines across all metrics for sedation prediction, as illustrated in Table 2, and it was selected for all downstream applications.

[0164] Table 2. Performance of sedation classifiers on training data obtained via five-fold cross-validation.DB2 / 650910081.2 43Attorney Docket No.: 104593-5055-WO

[0165] Results. Next, the selected classifier was evaluated on training data through repeated measures k-fold cross-validation and on two randomly selected test sets, held-out infants not used in training, and held-out minutes from infants used in training. To better understand which anatomic landmarks were contributing to classification performance, feature importance was calculated using the XGBoost gain function. To generate a resampled distribution for gain, which is a deterministic function of the XGBoost classifier, gain was recalculated through random subsampling of 80% of the data and the recalculation was repeated 500 times. As illustrated in FIG. 7A, left panel, when predicting sedation, the selected classifier had high performance in training data (median ROC-AUC = 0.90), held- out minutes (ROC-AUC = 0.91), and the set of infants not used for training (AUROC = 0.87). Similarly high performance was observed for a machine learning model trained to predict cerebral dysfunction, when evaluating training data (median AUROC = 0.91), held- out minutes (AUROC = 0.90), and held-out infants (AUROC = 0.76), illustrated in FIG. 7A, right panel. Both classifiers were interrogated to determine feature importance and revealed that feet and shoulders were consistently the most important features, illustrated in FIG. 7B.

[0166] Example 6 — Sedation prediction in an infant undergoing therapeutic hypothermia.

[0167] To illustrate the utility of predicting sedation in an individual case, sedation probability was plotted as a function of time in an infant born at gestational age 37+1 weeks with Apgar scores 1, 4, and 5, at 1, 5, and 10 minutes, who required therapeutic hypothermia for probable hypoxic ischemic encephalopathy, as illustrated in FIGS. 8A-C. The newborn received phenobarbital before rewarming due to increasing epileptiform discharges observed on EEG. Sedation probability increased first with hypothermia induction and again after phenobarbital administration (P(SedcUion m^Mm = 0.23 from 1476 minutes before phenobarbital versus 0.59 from 2268 minutes after, Wilcoxon-rank sum test P<10'10). Exemplary video frames with heatmaps of infant motion during the preceding five minutes visually capture this decrease in movement, illustrated in FIG. 8D.

[0168] REFERENCES1. Khan, O. A., Garcia-Sosa, R., Hageman, J. R., Msall, M. & Kelley, K. R. Core Concepts: Neonatal Neurological Examination. Neoreviews 15, e316-e324 (2014).DB2 / 650910081.2 44Attorney Docket No.: 104593-5055-WO2. Volpe, J. J., Inder, T. E., Darras, B. T., de Vries, L. S., du Plessis, A. J., Neil, J. J. & Perlman, J. M. Volpe’s Neurology of the Newborn. Volpe’s Neurol. Newborn 1-1224 (2018). doi: 10.1016 / C2010-0-68825-0.3. Russ, J. B., Simmons, R. & Glass, H. C. Neonatal Encephalopathy: Beyond Hypoxic- Ischemic Encephalopathy. Neoreviews 22, el48-el62 (2021).4. Verstraete, E. H., Blot, K., Mahieu, L., Vogelaers, D. & Blot, S. Prediction models for neonatal health care-associated sepsis: a meta-analysis. Pediatrics 135, el002-el014 (2015).5. Mayock, D. E. & Gleason, C. A. Pain and Sedation in the NICU. Neoreviews 14, e22-e31 (2013).6. Donato, J., Rao, K. & Lewis, T. Pharmacology of Common Analgesic and Sedative Drugs Used in the Neonatal Intensive Care Unit. Clin. Perinatol. 46, 673-692 (2019).7. Morgan, M. E., Kukora, S., Nemshak, M. & Shuman, C. J. Neonatal Pain, Agitation, and Sedation Scale’s use, reliability, and validity: a systematic review. J. Perinatol. 2020 4012 40, 1753-1763 (2020).8. Sarnat, H. B. & Samat, M. S. Neonatal encephalopathy following fetal distress. A clinical and electroencephalographic study. Arch. Neurol. 33, 696-705 (1976).9. Shankaran, S., Laptook, A. R., Ehrenkranz, R. A., Tyson, J. E., McDonald, S. A., Donovan, E. F., Fanaroff, A. A., Poole, W. K., Wright, L. L., Higgins, R. D., Finer, N. N., Carlo, W. A., Duara, S., Oh, W., Cotten, C. M., Stevenson, D. K., Stoll, B. J., Lemons, J. A., Guillet, R. & Jobe, A. H. Whole-body hypothermia for neonates with hypoxic-ischemic encephalopathy. N. Engl. J. Med. 353, 1574-1584 (2005).10. Pavageau, L., Sanchez, P. J., Steven Brown, L. & Chalak, L. F. Inter-rater reliability of the modified Samat examination in preterm infants at 32-36 weeks’ gestation. Pediatr. Res. 87, 697-702 (2020).11. Abend, N. S., Topjian, A. A. & Williams, S. How much does it cost to identify a critically ill child experiencing electrographic seizures? J. Clin. Neurophysiol. 32, 257-264 (2015).12. Moyer, L. B., Lauderbaugh, D. L., Worten, K., Carter, C., Holub, P., Santos Manrique, R. A., Bergman, J. H., Dilloway, M. A., Hamid, M. & Glenn, L. Highstage Device-related Pressure Injury Reduction in a Neonatal Intensive Care Unit: A Quality Improvement Project. Pediatr. Qual. Saf. 7, E554 (2022).DB2 / 650910081.2 45Attorney Docket No.: 104593-5055-WO13. Schultz, B., Schultz, M., Boehne, M. & Dennhardt, N. EEG monitoring during anesthesia in children aged 0 to 18 months: amplitude-integrated EEG and age effects. BMC Pediatr. 22, (2022).14. Shellhaas, R. A., Chang, T., Tsuchida, T., Scher, M. S., Riviello, J. J., Abend, N. S., Nguyen, S., Wusthoff, C. J. & Clancy, R. R. The American clinical neurophysiology society’s guideline on continuous electroencephalography monitoring in neonates. J. Clin. Neurophysiol. 28, 611-617 (2011).15. Massey, S. L., Shou, H., Clancy, R., DiGiovine, M., Fitzgerald, M. P., Fung, F. W., Farrar, J. & Abend, N. S. Interrater and Intrarater Agreement in Neonatal Electroencephalogram Background Scoring. J. Clin. Neurophysiol. 36, 1-8 (2019).16. Mathis, A., Mamidanna, P., Cury, K. M., Abe, T., Murthy, V. N., Mathis, M. W. & Bethge, M. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 21, 1281-1289 (2018).17. Cao, Z., Hidalgo, G., Simon, T., Wei, S. E. & Sheikh, Y. OpenPose: Realtime MultiPerson 2D Pose Estimation using Part Affinity Fields. IEEE Trans. Pattern Anal. Mach. Intell. 43, 172-186 (2018).18. Moons, K. G. M., Altman, D. G., Reitsma, J. B., loannidis, J. P. A., Macaskill, P., Steyerberg, E. W., Vickers, A. J., Ransohoff, D. F. & Collins, G. S. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. Ann. Intern. Med. 162, W1-W73 (2015).19. Collins, G. S., Reitsma, J. B., Altman, D. G. & Moons, K. G. M. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The TRIPOD Statement. https: / / doi.org / 10.7326 / M14-0697 162, 55-63 (2015).20. Battig, L., Dunner, C., Cserpan, D., Riiegger, A., Hagmann, C., Schmitt, B., Pisani, F. & Ramantani, G. Levetiracetam versus Phenobarbital for Neonatal Seizures: A Retrospective Cohort Study. Pediatr. Neurol. 138, 62-70 (2023).21. Sciortino, G., Farinella, G. M., Battiato, S., Leo, M. & Distante, C. On the Estimation of Children’s Poses. Leet. Notes Comput. Sci. (including Subser. Leet. Notes Artif. Intell. Leet. Notes Bioinformatics) 10485 LNCS, 410-421 (2017).DB2 / 650910081.2 46Attorney Docket No.: 104593-5055-WO22. Suryanto, M. E., Saputra, F., Kurnia, K. A., Vasquez, R. D., Roldan, M. J. M., Chen, K. H. C., Huang, J. C. & Hsiao, C. Der. Using DeepLabCut as a Real-Time and Markerless Tool for Cardiac Physiology Assessment in Zebrafish. Biology (Basel). 11, (2022).23. Insafutdinov, E., Pishchulin, L., Andres, B., Andriluka, M. & Schiele, B. DeeperCut: A Deeper, Stronger, and Faster Multi -Person Pose Estimation Model. Leet. Notes Comput. Sci. (including Subser. Leet. Notes Artif. Intell. Leet. Notes Bioinformatics) 9910 LNCS, 34-50 (2016).24. Lu, J., Jayakumari, A., Richter, F., Li, Y. & Yip, M. C. SuPer Deep: A Surgical Perception Framework for Robotic Tissue Manipulation using Deep Learning for Feature Extraction. Proc. - IEEE Int. Conf. Robot. Autom. 2021 -May, 4783-4789 (2020).25. He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learning for Image Recognition. Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. 2016-December, 770-778 (2015).26. Tomasev, N., Harris, N., Baur, S., Mottram, A., Glorot, X., Rae, J. W., Zielinski, M., Askham, H., Saraiva, A., Magliulo, V., Meyer, C., Ravuri, S., Protsyuk, I., Connell, A., Hughes, C. O., Karthikesalingam, A., Cornebise, J., Montgomery, H., Rees, G., Laing, C., Baker, C. R., Osborne, T. F., Reeves, R., Hassabis, D., King, D., Suleyman, M., Back, T., Nielson, C., Seneviratne, M. G., Ledsam, J. R. & Mohamed, S. Use of deep learning to develop continuous-risk models for adverse event prediction from electronic health records. Nat. Protoc. 2021 166 16, 2765-2787 (2021).27. Whalen, S., Schreiber, J., Noble, W. S. & Pollard, K. S. Navigating the pitfalls of applying machine learning in genomics. Nat. Rev. Genet. 2021 233 23, 169-181 (2021).28. Chambers, C., Seethapathi, N., Saluja, R., Loeb, H., Pierce, S. R., Bogen, D. K., Prosser, L., Johnson, M. J. & Kording, K. P. Computer Vision to Automatically Assess Infant Neuromotor Risk. IEEE Trans. Neural Syst. Rehabil. Eng. 28, 2431-2442 (2020).29. Cabrera-Quiros, L., Kommers, D., Wolvers, M. K., Oosterwijk, L., Arents, N., Van Der Sluijs-Bens, J., Cottaar, E. J. E., Andriessen, P. & Van Pul, C. Prediction of Late-Onset Sepsis in Preterm Infants Using Monitoring Signals and Machine Learning. Crit. care Explor. 3, E0302 (2021).30. Efron, B. & Tibshirani, R. J. An Introduction to the Bootstrap. (Chapman and Hall / CRC, 1994). doi:10.1201 / 9780429246593.DB2 / 650910081.2 47Attorney Docket No.: 104593-5055-WO31. Ho, J., Tumkaya, T., Aryal, S., Choi, H. & Claridge-Chang, A. Moving beyond P values: data analysis with estimation graphics. Nat. Methods 16, 565-566 (2019).32. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passes, A., Cournapeau, D., Brucher, M., Perrot, M. & Duchesnay, E. Scikit-leam: Machine Learning in Python. J. Mach. Learn. Res. 12, 2825-2830 (2011).33. Chen, T. & Guestrin, C. XGBoost: A Scalable Tree Boosting System. Proc. 22ndACM SIGKDD Int. Conf. Knowl. Discov. Data Min. doi: 10.1145 / 2939672.34. Sharpe, C., Sharpe, C., Reiner, G. E., Davis, S. L., Nespeca, M., Gold, J. J., Rasmussen, M., Kuperman, R., Jo Harbert, M., Michelson, D., Joe, P., Wang, S., Rismanchi, N., Le, N. M., Mower, A., Kim, J., Battin, M. R., Lane, B., Honold, J., Knodel, E., Arnell, K., Bridge, R., Lee, L., Ernstrom, K., Raman, R. & Haas, R. H. Levetiracetam Versus Phenobarbital for Neonatal Seizures: A Randomized Controlled Trial. Pediatrics 145, (2020).35. Peng, Z., Kommers, D., Liang, R. H., Long, X., Cottaar, W., Niemarkt, H., Andriessen, P. & van Pul, C. Continuous sensing and quantification of body motion in infants: A systematic review. Heliyon 9, (2023).36. Giordano, V., Deindl, P., Kuttner, S., Waldhor, T., Berger, A. & Olischar, M. The Neonatal Pain, Agitation and Sedation Scale reliably detected oversedation but failed to differentiate between other sedation levels. Acta Paediatr. 103, e515-e521 (2014).37. McPherson, C., Ortinau, C. M. & Vesoulis, Z. Practical approaches to sedation and analgesia in the newborn. J. Perinatol. 2020 413 41, 383-395 (2020).38. Carroll, A. E. & Christakis, D. A. Call for Papers on Artificial Intelligence Applied to Pediatric Care. JAMA Pediatr. 177, 884-885 (2023).39. Antink, C. H., Ferreira, J. C. M., Paul, M., Lyra, S., Heimann, K., Karthik, S., Joseph, J., Jayaraman, K., Orlikowsky, T., Sivaprakasam, M. & Leonhardt, S. Fast body part segmentation and tracking of neonatal video data using deep learning. Med. Biol. Eng. Comput. 58, 3049-3061 (2020).40. Bemtson, G. G., Thomas Bigger, J., Eckberg, D. L., Grossman, P., Kaufmann, P. G., Malik, M., Nagaraja, H. N., Porges, S. W., Saul, J. P., Stone, P. H. & Van Der Molen, M. W. Heart rate variability: origins, methods, and interpretive caveats. Psychophysiology 34, 623- 648 (1997).DB2 / 650910081.2 48Attorney Docket No.: 104593-5055-WO41. Boghossian, N. S., Geraci, M., Phibbs, C. S., Lorch, S. A., Edwards, E. M. & Horbar, J. D. Trends in Resources for Neonatal Intensive Care at Delivery Hospitals for Infants Born Younger Than 30 Weeks’ Gestation, 2009-2020. JAMA Netw. Open 6, e2312107-e2312107 (2023).42. Ho, J., Tumkaya, T., Aryal, S., Choi, H. & Claridge-Chang, A. Moving beyond P values: data analysis with estimation graphics. Nat. Methods 16, 565-566 (2019).43. Efron, B. & Tibshirani, R. J. An Introduction to the Bootstrap. (Chapman and Hall / CRC, 1994). doi:10.1201 / 9780429246593.44. Chen, T. & Guestrin, C. XGBoost: A Scalable Tree Boosting System. Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discov. Data Min. doi: 10.1145 / 2939672.45. Collins, G. S., Reitsma, J. B., Altman, D. G. & Moons, K. G. M. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The TRIPOD Statement. https: / / doi.org / 10.7326 / M14-0697 162, 55-63 (2015).46. Moons, K. G. M., Altman, D. G., Reitsma, J. B., loannidis, J. P. A., Macaskill, P., Steyerberg, E. W., Vickers, A. J., Ransohoff, D. F. & Collins, G. S. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. Ann. Intern. Med. 162, W1-W73 (2015).

[0169] CONCLUSION

[0170] The terminology used in the description of the invention herein is for the purpose of describing particular implementations only and is not intended to be limiting of the invention. As used in the description of the invention and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0171] All references cited herein are incorporated herein by reference in their entirety and for all purposes to the same extent as if each individual publication or patent or patentDB2 / 650910081.2 49Attorney Docket No.: 104593-5055-WO application was specifically and individually indicated to be incorporated by reference in its entirety for all purposes.

[0172] The foregoing description, for purpose of explanation, has been described with reference to specific implementations. However, the illustrative discussions above are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings. The implementations described herein were chosen and described in order to best explain the principles of the invention and its practical applications, to thereby enable others skilled in the art to best utilize the invention and various implementations with various modifications as are suited to the particular use contemplated.DB2 / 650910081.2 50

Claims

Attorney Docket No.: 104593-5055-WOWHAT IS CLAIMED IS:

1. A method for predicting a neurologic condition in a subject using a movement index derived from pose estimation, wherein the subject is a human subject less than one year old, the method comprising:(a) capturing video data of the subject over a duration of time using one or more cameras positioned to observe the body of the subject;(b) processing the video data using a pose estimation artificial intelligence model to detect and track a plurality of anatomical landmarks of the subject in successive video frames in the video data;(c) computing a movement index for a defined time interval based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarks across the interval within the video data, wherein the movement index is normalized to account for body size and camera distance; and(d) inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, wherein the neurologic condition represents a clinically relevant neurological state.

2. The method of claim 1, wherein the neurologic condition comprises sedation or cerebral dysfunction.

3. The method of claim 1 or 2, wherein the clinically relevant neurological state comprises sedation management in the settings of therapeutic hypothermia, pulmonary hypertension, or unrepaired Tetralogy of Fallot.

4. The method of any one of claims 1-3, wherein the clinically relevant neurological state comprises encephalopathy, hypoglycemia, sepsis, hyperammonenia, neonatal stroke, seizure, or adverse effect of medication.

5. The method of claim 4, wherein the encephalopathy comprises hypoxic ischemic encephalopathy.

6. The method of any one of claims 1-5, wherein the subject has a postmenstrual age of 64 weeks or less.DB2 / 650910081.2 51Attorney Docket No.: 104593-5055-WO7. The method of any one of claims 1-6, wherein the subject has a postmenstrual age of 48 weeks or less.

8. The method of any one of claims 1-7, wherein the subject has been administered with a sedative mediation or an anti-seizure medication.

9. The method of claim 8, wherein the anti-seizure medication comprises levetiracetam, phenobarbital, or pentobarbital.

10. The method of any one of claims 1-9, wherein the one or more cameras are positioned to capture at least two anatomical landmarks, in the plurality of anatomical landmarks, on the body of the subject.

11. The method of any one of claims 1-10, wherein the duration of time of video data comprises at least 1 minute, at least 10 minutes, or at least 1 hour of video data of the subject.

12. The method of any one of claims 1-11, wherein the plurality of anatomical landmarks comprise a subset selected from the group consisting of: nose, neck, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hand, right hand, left hip, right hip, left knee, right knee, left ankle, right ankle, left foot, and right foot.

13. The method of any one of claims 1-11, wherein the plurality of anatomical landmarks comprise one or more anatomical landmarks selected from the group consisting of: nose, neck, shoulders, elbows, hands, hips, knees, and feet.

14. The method of any one of claims 1-13, wherein the plurality of anatomical landmarks comprises at least shoulders and feet.

15. The method of any one of claims 1-14, wherein, for one or more video frames of the successive video frames, one or more anatomical landmarks in the plurality of anatomical landmarks are occluded.DB2 / 650910081.2 52Attorney Docket No.: 104593-5055-WO16. The method of any one of claims 1-15, wherein the plurality of anatomical landmarks comprises from 3 to 8 anatomical landmarks.

17. The method of any one of claims 1-16, wherein the pose estimation artificial intelligence model comprises a convolutional neural network or a deep convolutional neural network.

18. The method of any one of claims 1-17, wherein the pose estimation artificial intelligence model comprises DeepLabCut.

19. The method of any one of claims 1-18, wherein the pose estimation artificial intelligence model generates, for each respective video frame in the successive video frames, a corresponding vectorized pose for the subject, and wherein: the corresponding vectorized pose comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, corresponding X and Y coordinates for the respective anatomical landmark in the respective video frame.

20. The method of claim 19, wherein the corresponding vectorized pose further comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a corresponding likelihood for the respective anatomical landmark in the respective video frame.

21. The method of any one of claims 1-20, the method further comprising determining the movement index using, for each anatomical landmark in the plurality of anatomical landmarks, X and Y coordinates of the respective anatomical landmark across a subset of successive video frames during the defined time interval.

22. The method of claim 21, wherein the movement index comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a variance of movement derived from the X and Y coordinates of the respective anatomical landmark across the subset of successive video frames during the defined time interval.

23. The method of any one of claims 1-22, wherein the defined time interval comprises at least 30 seconds, at least 1 minute, or at least 2 minutes.DB2 / 650910081.2 53Attorney Docket No.: 104593-5055-WO24. The method of any one of claims 1-22, wherein the defined time interval is from 30 seconds to 2 minutes.

25. The method of any one of claims 1-24, wherein computing the movement index further comprises: computing, for each respective defined time interval in a plurality of defined time intervals, a corresponding movement index in the position of the tracked plurality of anatomical landmarks across the defined time interval within the video data, wherein the movement index is normalized to account for body size and camera distance; and determining a measure of central tendency for the movement index across the plurality of defined intervals.

26. The method of claim 25, wherein each respective defined time interval in the plurality of defined time intervals (i) comprises a respective portion of the duration of time for capturing video data and (ii) corresponds to a respective subset of successive video frames in the video data captured during the respective portion of the duration of time.

27. The method of claim 25 or 26, wherein each respective defined time interval in the plurality of defined time intervals comprises an equal amount of time.

28. The method of any one of claims 25-27, wherein the measure of central tendency for the movement index comprises a median of a plurality of movement variances, wherein each respective movement variant in the plurality of movement variances is determined within a 1- minute time interval for each respective anatomical landmark in the plurality of anatomical landmarks.

29. The method of any one of claims 1-28, wherein the normalization of the movement index accounts for subject body size and camera distance by using the distance between a nose landmark and a neck landmark of the subject.

30. The method of any one of claims 1-29, wherein the trained machine learning model comprises a logistic regression model, a support vector machine, or XGBoost.DB2 / 650910081.2 54Attorney Docket No.: 104593-5055-WO31. The method of any one of claims 1-30, wherein the trained machine learning model comprises a plurality of parameters that reflects a plurality of training data for a plurality of subjects, the plurality of training data comprising: for each respective subject in the plurality of subjects, (i) for each respective defined time interval in a plurality of defined time intervals, a corresponding variance in the position of a tracked plurality of anatomical landmarks across the defined time interval within the video data, wherein the movement index is normalized to account for body size and camera distance, and (ii) a respective indication of a neurologic condition in the respective subject.

32. The method of claim 31, wherein the respective indication of the neurologic condition is determined using electroencephalography.

33. The method of claim 31 or 32, wherein the plurality of parameters comprises at least 10, at least 1000, at least 10,000, or at least 1 x 106parameters.

34. The method of any one of claims 1-33, further comprising: performing continuous neurological monitoring of the subject by repeating the (a) capturing, (b) processing, (c) computing, and (d) inputting over a period of time, thereby obtaining an updated prediction of the neurologic condition in the subject.

35. The method of claim 34, wherein the period of time comprises at least 1 hour, at least 6 hours, at least 12 hours, or at least 24 hours.

36. The method of any one of claims 1-35, further comprising providing a report comprising at least the prediction of the neurologic condition in the subject and the clinically relevant neurological state.

37. The method of any one of claims 1-36, further comprising diagnosing the subject with the clinically relevant neurological state.

38. The method of any one of claims 1-37, further comprising adjusting a dosage or schedule of a medication based on the prediction of the neurologic condition.DB2 / 650910081.2 55Attorney Docket No.: 104593-5055-WO39. The method of any one of claims 1-38, further comprising determining a state of oversedation based on the prediction of the neurologic condition, and, responsive to determining the state of oversedation, reducing an amount of a medication administered to the subject.

40. The method of any one of claims 1-38, further comprising determining a state of undersedation based on the prediction of the neurologic condition, and, responsive to determining the state of undersedation, increasing an amount of a medication administered to the subject.

41. The method of any one of claims 38-40, wherein the medication is a sedative medication or an anti-seizure medication.

42. The method of any one of claims 38-40, wherein the medication comprises levetiracetam, phenobarbital, or pentobarbital.

43. The method of any one of claims 1-42, further comprising validating the neurologic condition using electroencephalography for the subject, where the neurologic condition comprises cerebral dysfunction.

44. The method of any one of claims 1-43, further comprising displaying a real-time graphical representation of the movement index over time for use by clinical personnel.

45. A system for predicting a neurologic condition in a subject, the system comprising: a camera configured to capture video data of a body of a subject over a duration of time, wherein the subject is a human subject less than one year old; and a processor operatively coupled to the camera and configured to perform a method comprising: processing the video data using a pose estimation artificial intelligence model to detect and track a plurality of anatomical landmarks of the subject in successive video frames in the video data; computing a movement index for a defined time interval based on a statistical measure of variance in the position of the tracked plurality of anatomical landmarksDB2 / 650910081.2 56Attorney Docket No.: 104593-5055-WO across the interval within the video data, wherein the movement index is normalized to account for body size and camera distance; and inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, wherein the neurologic condition represents a clinically relevant neurological state.

46. The system of claim 45, wherein the camera is positioned to capture at least two anatomical landmarks, in the plurality of anatomical landmarks, on the body of the subject, wherein the plurality of anatomical landmarks are selected from the group consisting of: nose, neck, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hand, right hand, left hip, right hip, left knee, right knee, left ankle, right ankle, left foot, and right foot.

47. The system of claim 45 or 46, wherein the camera is positioned to capture at least one or both shoulders and one or both feet of the subject.

48. The system of any one of claims 45-47, wherein the neurologic condition comprises sedation or cerebral dysfunction.

49. The system of any one of claims 45-48, wherein the clinically relevant neurological state comprises sedation management in the settings of therapeutic hypothermia, pulmonary hypertension, or unrepaired Tetralogy of Fallot.

50. The system of any one of claims 45-49, wherein the clinically relevant neurological state comprises encephalopathy, hypoglycemia, sepsis, hyperammonenia, neonatal stroke, seizure, or adverse effect of medication.

51. The system of claim 50, wherein the encephalopathy comprises hypoxic ischemic encephalopathy.

52. The system of any one of claims 45-51, wherein the subject has a postmenstrual age of 64 weeks or less.DB2 / 650910081.2 57Attorney Docket No.: 104593-5055-WO53. The system of any one of claims 45-52, wherein the subject has a postmenstrual age of 48 weeks or less.

54. The system of any one of claims 45-53, wherein the subject has been administered with a sedative mediation or an anti-seizure medication.

55. The system of claim 54, wherein the anti-seizure medication comprises levetiracetam, phenobarbital, or pentobarbital.

56. The system of any one of claims 45-55, wherein the duration of time of video data comprises at least 1 minute, at least 10 minutes, or at least 1 hour of video data of the subject.

57. The system of any one of claims 45-56, wherein the plurality of anatomical landmarks comprise a subset selected from the group consisting of: nose, neck, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hand, right hand, left hip, right hip, left knee, right knee, left ankle, right ankle, left foot, and right foot.

58. The system of any one of claims 45-56, wherein the plurality of anatomical landmarks comprise one or more anatomical landmarks selected from the group consisting of: nose, neck, shoulders, elbows, hands, hips, knees, and feet.

59. The system of any one of claims 45-58, wherein the plurality of anatomical landmarks comprises at least shoulders and feet.

60. The system of any one of claims 45-59, wherein, for one or more video frames of the successive video frames, one or more anatomical landmarks in the plurality of anatomical landmarks are occluded.

61. The system of any one of claims 45-60, wherein the plurality of anatomical landmarks comprises from 3 to 8 anatomical landmarks.

62. The system of any one of claims 45-61, wherein the pose estimation artificial intelligence model comprises a convolutional neural network or a deep convolutional neural network.DB2 / 650910081.2 58Attorney Docket No.: 104593-5055-WO63. The system of any one of claims 45-62, wherein the pose estimation artificial intelligence model comprises DeepLabCut.

64. The system of any one of claims 45-63, wherein the pose estimation artificial intelligence model generates, for each respective video frame in the successive video frames, a corresponding vectorized pose for the subject, and wherein: the corresponding vectorized pose comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, corresponding X and Y coordinates for the respective anatomical landmark in the respective video frame.

65. The system of claim 64, wherein the corresponding vectorized pose further comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a corresponding likelihood for the respective anatomical landmark in the respective video frame.

66. The system of any one of claims 45-65, the method further comprising determining the movement index using, for each anatomical landmark in the plurality of anatomical landmarks, X and Y coordinates of the respective anatomical landmark across a subset of successive video frames during the defined time interval.

67. The system of claim 66, wherein the movement index comprises, for each respective anatomical landmark in the plurality of anatomical landmarks, a variance of movement derived from the X and Y coordinates of the respective anatomical landmark across the subset of successive video frames during the defined time interval.

68. The system of any one of claims 45-67, wherein the defined time interval comprises at least 30 seconds, at least 1 minute, or at least 2 minutes.

69. The system of any one of claims 45-67, wherein the defined time interval is from 30 seconds to 2 minutes.

70. The system of any one of claims 45-69, wherein computing the movement index further comprises:DB2 / 650910081.2 59Attorney Docket No.: 104593-5055-WO computing, for each respective defined time interval in a plurality of defined time intervals, a corresponding movement index in the position of the tracked plurality of anatomical landmarks across the defined time interval within the video data, wherein the movement index is normalized to account for body size and camera distance; and determining a measure of central tendency for the movement index across the plurality of defined intervals.

71. The system of claim 70, wherein each respective defined time interval in the plurality of defined time intervals (i) comprises a respective portion of the duration of time for capturing video data and (ii) corresponds to a respective subset of successive video frames in the video data captured during the respective portion of the duration of time.

72. The system of claim 70 or 71, wherein each respective defined time interval in the plurality of defined time intervals comprises an equal amount of time.

73. The system of any one of claims 70-72, wherein the measure of central tendency for the movement index comprises a median of a plurality of movement variances, wherein each respective movement variant in the plurality of movement variances is determined within a 1- minute time interval for each respective anatomical landmark in the plurality of anatomical landmarks.

74. The system of any one of claims 45-73, wherein the normalization of the movement index accounts for subject body size and camera distance by using the distance between a nose landmark and a neck landmark of the subject.

75. The system of any one of claims 45-74, wherein the trained machine learning model comprises a logistic regression model, a support vector machine, or XGBoost.

76. The system of any one of claims 45-75, wherein the trained machine learning model comprises a plurality of parameters that reflects a plurality of training data for a plurality of subjects, the plurality of training data comprising: for each respective subject in the plurality of subjects, (i) for each respective defined time interval in a plurality of defined time intervals, a corresponding variance in the position of a tracked plurality of anatomical landmarks across the defined time interval within theDB2 / 650910081.2 60Attorney Docket No.: 104593-5055-WO video data, wherein the movement index is normalized to account for body size and camera distance, and (ii) a respective indication of a neurologic condition in the respective subject.

77. The system of claim 76, wherein the respective indication of the neurologic condition is determined using electroencephalography.

78. The system of claim 76 or 77, wherein the plurality of parameters comprises at least 10, at least 1000, at least 10,000, or at least 1 x 106parameters.

79. The system of any one of claims 45-78, further comprising: performing continuous neurological monitoring of the subject by repeating the (a) capturing, (b) processing, (c) computing, and (d) inputting over a period of time, thereby obtaining an updated prediction of the neurologic condition in the subject.

80. The system of claim 79, wherein the period of time comprises at least 1 hour, at least 6 hours, at least 12 hours, or at least 24 hours.

81. The system of any one of claims 45-80, further comprising providing a report comprising at least the prediction of the neurologic condition in the subject and the clinically relevant neurological state.

82. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the system to perform a method for predicting a neurologic condition, the method comprising: receiving video data of a subject captured by a camera, wherein the subject is a human subject less than one year old; applying a pose estimation artificial intelligence model to the video data to detect and track a plurality of anatomical landmarks of the subject across a sequence of video frames; computing a movement index for a defined time interval based on a statistical variance in the positions of the anatomical landmarks, wherein the movement index is normalized for patient body size and camera distance; and inputting the movement index to a trained machine learning model, thereby obtaining, as output from the model, a prediction of the neurologic condition in the subject, wherein the neurologic condition represents a clinically relevant neurological state.DB2 / 650910081.2 61

Images