Systems and methods for diagnosing a health condition based on patient time series data
Inventors
Wagner, Tyler • Aravamudan, Murali • BABU, Melwin • Barve, Rakesh • Soundararajan, Venkataramanan • Prasad, Ashim • CARPENTER, Corinne • Carlson, Katherine
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12718948-B2
Publication Date
2026-08-25
Expiration Date
Abstract
Disclosed systems, methods, and computer readable media can diagnose a health condition based on patient time series data. For example, a method for diagnosing a health condition based on patient time series data includes identifying a training set of health records comprising a first set of patient time series data, training a neural network using the training set of health records, and executing the trained neural network model to diagnose a health condition based on a second set of patient time series data. In further examples, the first set of patient time series data and the second set of patient time series data can each comprise electrocardiogram data and the health condition can comprise pulmonary hypertension.
Core Innovation
The disclosed method and system receive patient electrocardiogram time-series data measured prior to a diagnosis date and apply trained neural network models to predict whether the patient is at risk of developing a health condition within five years. The models process the time-series data through convolutional branches, including at least one branch configured to generate a one-dimensional waveform representation, and concatenate branch outputs. Training utilizes health records comprising electrocardiogram data measured within diagnosis windows rather than pre-emptively, correlated to positive diagnoses.
The system executes prediction via a non-transitory memory and hardware processors configured to run the models. Dual-branch neural network architectures process electrocardiogram waveforms with one-dimensional convolutions and spectrograms with two-dimensional convolutions, combining extracted features with demographic indicators including age and gender. Prediction evaluations encompass pulmonary hypertension across pre-diagnosis time windows alongside preliminary models for AL amyloidosis.
Claims Coverage
The claims recite two independent claims comprising a method and a system, detailing four inventive features directed to pre-diagnosis electrocardiogram risk prediction.
Pre-diagnosis electrocardiogram time-series data and five-year risk prediction
The method and system receive patient time-series data comprising patient electrocardiogram data measured from the patient prior to a diagnosis date and predict whether the patient is at risk of developing a health condition within the next five years.
Trained neural network model architecture
Trained neural network models comprise one or more convolutional branches, with at least one branch configured to process patient time-series data into a one-dimensional waveform representation, and a concatenating layer configured to concatenate outputs of the branches to generate a model output.
Training using diagnosis-window electrocardiogram data
The neural network models are trained with a set of health records comprising electrocardiogram data correlated to positive diagnoses of the health condition, with the data measured within diagnosis windows rather than pre-emptively to diagnoses.
Processor-based system architecture
Claim 11 recites a non-transitory memory and one or more hardware processors configured to read instructions from the non-transitory memory to receive the patient time-series data, execute the trained neural network models, and predict risk.
The independent claims cover a method and a processor-based system using pre-diagnosis electrocardiogram time-series data, convolutional branches, and concatenated model outputs to predict health condition risk within five years.
Stated Advantages
The patent reports strong prediction performance across cohorts, generally demonstrating AUCs around 0.85–0.91, including high AUCs for pre-diagnosis electrocardiograms in pulmonary-hypertension analyses.
Documented Applications
Predicting whether a patient is at risk of developing a health condition within the next five years based on patient electrocardiogram time-series data measured before diagnosis.
Predicting whether a patient is at risk of developing pulmonary hypertension using pre-diagnosis electrocardiogram data evaluated across pre-diagnosis time windows, with mean pulmonary artery pressure and tricuspid regurgitation velocity identified as diagnostic measures.
Predicting AL amyloidosis from electrocardiograms using a preliminary model.
Electrocardiogram-based binary classification, pre-diagnosis performance evaluation, and clinical-note classification.
Interested in licensing this patent?