Abstract: Techniques disclosed herein relate to identifying individuals in digital images. In some embodiments, a digital image may be acquired (802) that captures an environment containing at least a first subject. A first portion of the digital image depicting the first subject may be segmented (806) into a plurality of superpixels. For each superpixel of the plurality of superpixels: a semantic label may be assigned (810) to the superpixel; features of the superpixel may be extracted (812); and a measure of similarity between the features extracted from the superpixel and features extracted from a reference superpixel identified in a reference digital image may be determined (814), wherein the reference superpixel has a reference semantic label that matches the semantic label assigned to the superpixel. Based on the measures of similarity associated with the plurality of superpixels, it may be determined (818) that the first subject is depicted in the reference image.

Abstract: A method of explaining a machine learning model, including: receiving a plurality of disease states for a patient over time from a database, wherein the disease states have a plurality of features; generating a plurality of locally faithful explanation models for the patient for the disease states based upon the machine learning model; calculating an explanation with respect to one feature of the plurality of features over time using the locally faithful explanation models; and calculating the importance of the one feature of the plurality of features over time based upon the plurality of locally faithful explanation models.

Abstract: A method (100) for identifying time series data using a time series retrieval system (800), comprising: receiving (120) a plurality of time series, each time series comprising a plurality of datapoints, wherein a least some of the plurality of times series comprise datapoints obtained at irregular time intervals within the time period; storing (130) the received plurality of time series in a database; generate (140) a context vector for each of the plurality of time series; receiving (150) a request for identification of one or more of the plurality of time series based on similarity to a time series query; identifying (160), based on similarity to the query time series context vector, one or more of the stored generated context vectors; retrieving (170) each time series associated with the identified one or more stored generated context vectors; and providing (180) the retrieved time series.

Abstract: A device, system and method for generating a prediction model for a test patient. To generate the prediction model, a multi-dimensional clinical time series for each of a plurality of training patients is collected to generate a training population. A machine learning algorithm is then trained using the training population. Measurement data corresponding to the test patient is also received, the measurement data includes a multi-dimensional clinical time series for the test patient. The test patient is not included in the plurality of training patients. The prediction model is generated for the test patient based on i) the measurement data corresponding to the test patient and ii) training the machine learning algorithm using the training population.

Abstract: A method (100) for identifying time series data using a time series retrieval system (800), comprising: receiving (120) a plurality of time series, each time series comprising a plurality of datapoints, wherein a least some of the plurality of times series comprise datapoints obtained at irregular time intervals within the time period; storing (130) the received plurality of time series in a database; generate (140) a context vector for each of the plurality of time series; receiving (150) a request for identification of one or more of the plurality of time series based on similarity to a time series query; identifying (160), based on similarity to the query time series context vector, one or more of the stored generated context vectors; retrieving (170) each time series associated with the identified one or more stored generated context vectors; and providing (180) the retrieved time series.

Abstract: A method for training a probabilistic encoder-decoder having a latent space, the method including: extracting different types of medical data for a group of individuals; creating a data matrix X including the extracted medical data, wherein each row of the data matrix X includes data for one of the group of individuals; creating condition matrix C including features to define a clinical condition, wherein each row of the condition matrix C includes the condition data for one of the group of individuals; and training the encoder and the decoder to learn the latent space by minimizing the reconstruction loss and using a regularization effect to force clinically similar inputs to be close together in the latent space.

Abstract: The present disclosure is directed to systems and methods for developing an individual-specific patient baseline for a target patient. An exemplary method involves: determining one or more acuity scores for the target patient; identifying patient health data corresponding to one or more low acuity time periods; storing retrospective clinical data from a group of patients in a second database; comparing the patient health data corresponding to the one or more low acuity time periods with retrospective clinical data from a group of patients by identifying one or more patient subgroups; determining the individual-specific patient baseline using an adaptive baseline selection algorithm, wherein the adaptive baseline selection algorithm is used to determine whether to determine the individual-specific patient baseline using patient health data or using retrospective clinical data from one or more patient subgroups; and displaying, using a user interface, the individual-specific patient baseline.

Abstract: Techniques disclosed herein relate to learning and applying contextual patient similarities. Multiple template similarity functions (118) may be provided (602). Each template similarity function may compare a respective subset of features of a query entity feature vector with a corresponding subset of features of a candidate entity feature vector. A composite similarity function (120) may be provided (604) as a weighted combination of respective outputs of the template similarity functions. A plurality of labeled entity vectors may be provided (606) as context training data. An approximation function may be applied (608) to approximate a first context label for each respective labeled entity vector. A first context specific composite similarity function may be trained (610) based on the composite similarity function by learning first context weights for the template similarity functions using a first loss function based on output of application of the approximation function to the first context training data.

Abstract: A method is provided for developing adaptable predictive analytics for subjects in a medical facility.

Abstract: A method for reporting a user's risk of infection, comprising: (i) receiving, from a sensor of a wearable device worn by the user, user heart data; (ii) determining, from the received user heart data by a pulse rate variability phase acceleration algorithm, the user's pulse rate variability phase acceleration values for the first time period; (iii) determining, from the received user heart data by a pulse rate variability phase deceleration algorithm, the user's pulse rate variability phase deceleration values for the first time period; (iv) determining, by a trained infection risk prediction algorithm using the user's pulse rate variability phase acceleration values and the user's pulse rate variability phase deceleration values for the first time period, the user's risk of a current infection; and (v) providing, via a user interface, the determined user's risk of a current infection.

Abstract: A non-transitory computer-readable medium stores instructions readable and executable by at least one electronic processor (20) to perform an atrial fibrillation (AF) detection method (100). The method includes: generating a time-frequency representation of an electrocardiogram (ECG) signal acquired over a time interval; processing the time-frequency representation using a neural network (NN) (32) to output probabilities for rhythms of a set of rhythms including at least atrial fibrillation; assigning a rhythm for the ECG signal based on the probabilities for the rhythms of the set of rhythms output by the neural network; and controlling a display device (24) to display the rhythm assigned to the ECG signal.

Abstract: A method of explaining a machine learning model, including: receiving a plurality of disease states for a patient over time from a database, wherein the disease states have a plurality of features; generating a plurality of locally faithful explanation models for the patient for the disease states based upon the machine learning model; calculating an explanation with respect to one feature of the plurality of features over time using the locally faithful explanation models; and calculating the importance of the one feature of the plurality of features over time based upon the plurality of locally faithful explanation models .

Abstract: Techniques disclosed herein relate to learning and applying contextual patient similarities. In various embodiments, a first value for a query entity may be displayed (702) on an interface. The first value may be related to a first context. A first trained similarity function may be selected (704) from a plurality of trained similarity functions. The first trained similarity function may be associated with the first context. The first selected trained similarity function may be applied (706) to a set of features associated with the query entity and respective sets of features associated with a plurality of candidate entities. A set of one or more similar candidate entities may be selected (708) from the plurality of candidate entities based on application of the first trained similarity function. Information associated with the first set of one or more similar candidate entities may be displayed (710) on the interface.

Abstract: A method of explaining a machine learning model, including: receiving a plurality of disease states for a patient over time from a database, wherein the disease states have a plurality of features; generating a plurality of locally faithful explanation models for the patient for each disease state based upon the machine learning model; calculating an explanation with respect to one feature of the plurality of features over time using the locally faithful explanation models; and calculating the importance of the one feature of the plurality of features over time based upon the plurality of locally faithful explanation models.

Abstract: A method of implementing a task complexity learning system, including: learning a model for predicting the value of a continuous task variable y based upon an input variable x; learning an encoder that encodes a continuous task variable y into an encoded task value; calculating a loss function based upon the predicted value of y output by the model and the encoded task value output by the encoder; calculating a distortion function based upon the input continuous task variable y and the encoded task value, wherein learning the model and learning the encoder includes minimizing an objective function based upon the loss function and the distortion function for a set of input training data including x, y pairs.

Abstract: A non-transitory computer-readable medium stores instructions readable and executable by at least one electronic processor (20) to perform an atrial fibrillation (AF) detection method (100). The method includes: generating a time-frequency representation of an electrocardiogram (ECG) signal acquired over a time interval; processing the time-frequency representation using a neural network (NN) (32) to output probabilities for rhythms of a set of rhythms including at least atrial fibrillation; assigning a rhythm for the ECG signal based on the probabilities for the rhythms of the set of rhythms output by the neural network; and controlling a display device (24) to display the rhythm assigned to the ECG signal.

Abstract: An electronic clinical decision support (CDS) device (10) employs a trained CDS algorithm (30) that operates on values of a set of covariates to output a prediction of a medical condition. The CDS algorithm was trained on a training data set (22). The CDS device includes a computer (12) that is programmed to provide a user interface (62) for completing clinical survey questions using the display and the one or more user input devices. Marginal probability distributions (42) for the covariates of the set of covariates are generated from the completed clinical survey questions. The trained CDS algorithm is adjusted for covariate shift using the marginal probability distributions. A prediction of the medical condition is generated for a medical subject using the trained CDS algorithm adjusted for covariate shift (50) operating on values for the medical subject of the covariates of the set of covariates.

Abstract: A sensor device including a plurality of sensors producing sensor data, a storage device, a memory, a processor connected to the storage device and the memory, wherein the processor is configured to receive sensor data from the plurality of sensors, process the received sensor data from the plurality of sensors, receive external processing requests, perform external processing requests, and send external processing results.

Abstract: A method, system and computer-program product for identifying neural network inputs for a neural network that may have been incorrectly processed by the neural network. A set of activation values (of a subset of neurons of a single layer) associated with a neural network input is obtained. A neural network output associated with the neural network input is also obtained. A determination is made as to whether a first and second neural network input share similar sets of activation values, but dissimilar neural network outputs or vice versa. In this way a prediction can be made as to whether one of the first and second neural network inputs has been incorrectly processed by the neural network.

Abstract: A system (400) for monitoring an individual's sleep includes: (i) a patient monitor (410) configured to obtain a patient waveform, the patient waveform comprising information representative of a vital statistic of the patient; a processor (420) in communication with the patient monitor and configured to: (i) process the patient waveform to generate a segmented waveform; (ii) extract at least one feature from a segment of the waveform in a time domain and/or at least one feature from the segment of the waveform in the frequency domain; (iii) classify, using the at least one extracted feature, a sleep stage of the patient for the segment of the waveform; and (iv) generate, from classified sleep stages for a plurality of segments of the waveform, a sleep quality measurement; and a user interface (480) configured to report the generated sleep quality measurement.