Cardiac Foundation Model Training With Unlabeled ECG and PCG Data
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 the difficulty in obtaining expert labels and the need for invasive procedures, especially for rare conditions.
Innovation Solution
A method involving a foundation model trained with synchronized electrocardiogram (ECG) and phonocardiogram (PCG) signals, using unlabeled data to learn associations between features, and fine-tuning with a smaller set of labeled data to generate specialized models for conditions like atrial fibrillation and low ejection fraction.
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 is improved, but time consumption and complexity increase due to the need for expert labeling and invasive procedures
Solution Approach 1:
The patent applies preliminary action by pre-training a foundation model on large volumes of unlabeled cardiac data before fine-tuning with labeled data. This pre-training phase extracts general features and patterns from unlabeled ECG and PCG signals, reducing the subsequent need for extensive labeled data and expert annotation time while maintaining model accuracy.
2Measurement precision
If traditional machine learning models are trained with large amounts of labeled training data, then model accuracy is improved, but resource requirements and complexity increase due to the need for expert labels and invasive procedures
Solution Approach 1:
The patent applies preliminary action by pre-training a foundation model on large volumes of unlabeled cardiac data before fine-tuning with labeled data. This pre-training phase extracts general features and patterns from unlabeled ECG and PCG signals, reducing the subsequent need for extensive labeled data and expert annotation time while maintaining model accuracy.
Solution Approach 2:
The patent uses a foundation model as an intermediary between raw unlabeled cardiac data and the final specialized diagnostic model. This foundation model serves as a mediator that processes unlabeled data to create useful representations, which are then refined with a smaller amount of labeled data, simplifying the overall training pipeline and reducing dependency on expert-labeled datasets.
3Measurement precision
If labeled training data is obtained through invasive procedures to confirm ground truth conditions, then data accuracy is improved, but patient risk and procedural complexity increase
Solution Approach 1:
The patent uses a foundation model as an intermediary between raw unlabeled cardiac data and the final specialized diagnostic model. This foundation model serves as a mediator that processes unlabeled data to create useful representations, which are then refined with a smaller amount of labeled data, simplifying the overall training pipeline and reducing dependency on expert-labeled datasets.
Solution Approach 2:
The patent applies partial action by using a smaller subset of labeled data for fine-tuning after pre-training on unlabeled data. Instead of requiring complete labeled datasets for all training phases, the method uses labeled data only for the critical fine-tuning stage, reducing the number of invasive procedures needed while maintaining diagnostic accuracy.
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 and a confidence 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 and the confidence output in memory and/or displaying the classification output and the confidence output on a display device.