Systems and methods for a foundation model for cardiac data
Inventors
Mathew, George • Prince, John • Barbosa, Daniel • Venkatraman, Subramaniam
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12484858-B1
Publication Date
2025-12-02
Expiration Date
Abstract
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.
Core Innovation
The invention relates to obtaining an ECG signal and a PCG signal from a patient, wherein the ECG signal and PCG signal are synchronized. The PCG signal is converted to a PCG mel-spectrogram, and the ECG signal is partitioned into segments while the PCG mel-spectrogram is partitioned into patches. An input sequence is formed from the segments and the patches to represent synchronized ECG and PCG information for subsequent modeling.
The invention further provides entering the input sequence as input to a trained specialized model configured to output a classification output based on the input sequence. The trained specialized model is trained with labeled ECG and PCG signal pairs using a foundation model trained with unlabeled ECG and PCG signal pairs. The classification output is stored in memory and/or displayed on a display device, enabling patient-condition detection from synchronized ECG and PCG.
In addition, the foundation model uses a transformer-based foundation model trained using unlabeled training data input sets that include an ECG signal and a PCG mel-spectrogram. Unlabeled pre-training includes encoding segments and patches into encoded tokens, using mask tokens for masked segments and patches, and decoding to reconstruct the ECG signal and the PCG mel-spectrogram.
The invention also includes a deployed modeling approach in which a trained specialized model is used for classification of a patient condition based on a patient ECG signal and a patient PCG signal, using converting, partitioning, forming an input sequence, and entering that input sequence into the trained specialized model. In further refinements, classification can include a quality-level indication and a conditional two-stage approach using a second trained specialized model based on a quality threshold.
Claims Coverage
The independent claims are clm-00001, clm-00009, and clm-00016. Across these claims, the coverage centers on synchronized ECG and PCG representation as an input sequence of ECG segments and PCG mel-spectrogram patches, a foundation-model to specialized-model training approach using unlabeled and labeled pairs, and deployment to output a classification result for patient-condition detection.
Synchronized ECG and PCG signal conversion to mel-spectrogram input sequence
obtaining an ECG signal and a PCG signal from a patient, wherein the ECG signal and PCG signal are synchronized; converting the PCG signal to a PCG mel-spectrogram; partitioning the ECG signal into segments and partitioning the PCG mel-spectrogram into patches; and forming an input sequence from the segments and the patches
Trained specialized model output classification using foundation model training
entering the input sequence as input to a trained specialized model configured to output a classification output based on the input sequence, the trained specialized model trained with labeled ECG and PCG signal pairs using a foundation model trained with unlabeled ECG and PCG signal pairs
Storing and/or displaying classification output for patient condition detection
storing the classification output in memory and/or displaying the classification output on a display device
Data processing system with trained specialized model and foundation model pre-training
memory storing instructions and a trained specialized model, the trained specialized model trained with labeled, synchronized ECG and PCG signal pairs using a foundation model trained with unlabeled, synchronized ECG and PCG signal pairs, the foundation model including a pre-trained encoder that is pre-trained with the unlabeled, synchronized ECG and PCG signal pairs and a decoder, and the trained specialized model includes an encoder that comprises the pre-trained encoder that is further trained with the labeled, synchronized ECG and PCG signals pairs; and one or more processors configured to execute the instructions to obtain a patient ECG signal and a patient PCG signal from a patient, convert the patient PCG signal to a PCG mel-spectrogram, enter the patient ECG signal and the PCG mel-spectrogram as input to the trained specialized model configured to output a classification output, and store the classification output in memory and/or display the classification output on a display device
Generating a specialized model from a transformer-based foundation model for condition detection
training a transformer-based foundation model using a plurality of unlabeled training data input sets, each unlabeled training data input set comprising a first ECG signal and a first PCG mel-spectrogram; training the specialized model using a plurality of labeled training data input sets, each labeled training data input comprising a second ECG signal, a second PCG mel-spectrogram, and a label; and deploying the trained specialized model to detect a patient condition based on a patient ECG signal and a patient PCG signal, including converting the patient PCG signal to a PCG mel-spectrogram, partitioning the patient ECG signal into a plurality of segments and partitioning the PCG mel-spectrogram into a plurality of patches, forming an input sequence that includes the plurality of segments and the plurality of patches, and entering the input sequence as input to the trained specialized model
Across the independent claims, the inventive coverage is directed to converting synchronized ECG/PCG into a mel-spectrogram based representation, partitioning ECG and PCG mel-spectrogram into segments and patches to form an input sequence, and using a trained specialized model that produces a classification output for patient-condition detection, where the trained specialized model leverages a foundation model trained with unlabeled ECG and PCG signal pairs. Independent claim coverage also includes a system implementation that stores instructions and performs classification and a model-generation approach that trains a transformer-based foundation model on unlabeled data and trains/deploys a specialized model on labeled data.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?