Invention Title:

METHOD AND DEVICE TO PREDICT EXERCISE PEAK VO2, CARDIOVASCULAR OUTCOMES AND FUTURE DEATH USING ECG DEEP LEARNING MODELS

Publication number:

US20250261909

Publication date:

2025-08-21

Section:

Human necessities

Class:

A61B5/7267

Inventors:

Puneet Batra 🇺🇸 Cambridge, MA, United States

Steven Lubitz 🇺🇸 Boston, MA, United States

Patrick Ellinor 🇺🇸 Boston, MA, United States

Samuel Friedman 🇺🇸 Cambridge, MA, United States

Nathaniel Diamant 🇺🇸 Cambridge, MA, United States

Shaan Khurshid 🇺🇸 Boston, MA, United States

James Sawalla Guseh 🇺🇸 Boston, MA, United States

Assignees:

The General Hospital Corporation 🇺🇸 Boston, MA, United States

THE BROAD INSTITUTE, INC. 🇺🇸 Cambridge, MA, United States

Applicants:

The General Hospital Corporation 🇺🇸 Boston, MA, United States

The Broad Institute, Inc. 🇺🇸 Cambridge, MA, United States

Smart overview of the Invention

A method and device are introduced to estimate peak oxygen consumption ({dot over (V)}O2 PEAK) using electrocardiogram (ECG) data. This approach leverages a computer-implemented system where ECG waveform data is recorded and transmitted between computing devices. A convolutional neural network processes this data to predict {dot over (V)}O2 PEAK, which is then communicated back to the user or associated device.

Technical Field

The technology focuses on predicting oxygen consumption and cardiovascular outcomes using ECG data. It addresses the need for a system that can automatically and widely deliver {dot over (V)}O2 PEAK metrics, which are crucial for assessing cardiorespiratory fitness (CRF) and predicting health risks like cardiovascular diseases and mortality.

Background

{dot over (V)}O2 PEAK is a critical indicator of CRF, influencing prognostics in diverse health conditions. Traditional methods of measuring this metric are resource-intensive and not widely accessible. The invention seeks to make {dot over (V)}O2 PEAK assessment more accessible through deep learning models applied to resting 12-lead ECGs, providing a more efficient alternative to cardiopulmonary exercise testing.

Methodology

The system employs a convolutional neural network to analyze ECG data, generating a high-dimensional embedding of the ECG signals. This network can be pre-trained using datasets where {dot over (V)}O2 PEAK was measured via traditional methods. The process includes model training with techniques like Elastic Net penalization to ensure accuracy in predictions.

Applications and Devices

Beyond predicting oxygen consumption, the system estimates cardiorespiratory fitness and assesses cardiovascular risk using models like the Cox proportional hazards model. The technology can be integrated into wearable devices, providing real-time feedback on health metrics such as atrial fibrillation risk or potential heart failure, enhancing preventive healthcare measures.