ECG-PCG Foundation Model for Non-Invasive Cardiac Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model training protocols for cardiovascular conditions require large amounts of labeled data, which is time-consuming and often biased due to unrepresentative data composition, and/or invasive procedures, and/or lack of detection via standard techniques.
Innovation Solution
A foundation model trained with synchronized ECG and PCG data, utilizing a digital stethoscope to detect cardiovascular conditions, using a digital stethoscope to collect synchronized ECG and PCG data, and a foundation model to learn associations between features of ECG and PCG data in an unsupervised manner, with a smaller set of labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are trained with large amounts of labeled training data, then model accuracy may be improved, but the time-consuming labeling process and difficulty in finding experts worsen
Solution Approach 1:
The foundation model performs preliminary unsupervised pre-training on large volumes of unlabeled ECG and PCG data before fine-tuning with labeled data. This preliminary action extracts general features and patterns from unlabeled data, reducing the amount of labeled data needed and the time required for labeling while maintaining model accuracy.
2Measurement precision
If invasive procedures are used to confirm ground truth conditions for training data, then data accuracy may be improved, but patient risk and procedural complexity worsen
Solution Approach 1:
The foundation model acts as an intermediary that processes and analyzes non-invasive ECG and PCG signals to infer cardiovascular conditions. Instead of directly relying on invasive procedures to obtain ground truth, the model learns from patterns in non-invasive data, reducing patient risk while maintaining diagnostic accuracy.
3Measurement precision
If ground truth conditions are rare in the overall patient population, then model training data may be biased due to unrepresentative composition, but collecting sufficient training data becomes more challenging
Solution Approach 1:
The foundation model performs preliminary unsupervised pre-training on large volumes of unlabeled ECG and PCG data before fine-tuning with labeled data. This preliminary action allows the model to learn general cardiovascular patterns from abundant unlabeled data, then adapt to rare conditions with a smaller set of labeled examples, improving reliability without requiring large volumes of rare disease data.
Data Source
AI summary
The present description relates generally to methods and systems for detecting cardiovascular conditions using a foundation model. In one example, a method includes obtaining a synchronized ECG signal and a PCG signal from a patient, converting the PCG signal to a PCG mel-spectrogram, entering the ECG signal and the PCG mel-spectrogram as input to a trained specialized model configured to output a classification output based on the ECG signal and the PCG mel-spectrogram, the trained specialized model trained with labeled ECG and PCG signal pairs using a foundation model trained with unlabeled ECG and PCG signal pairs, and storing the classification output in memory and/or displaying the classification output on a display device.