Machine learning algorithm for the detection of cardiac amyloidosis from 12 lead ECG data
Inventors
Bose, Sluso Anne • Antony, Catherine Vinnarasi • T.A, Komala • Ali Shaikh, Rozina Moazzam
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12620488-B2
Publication Date
2026-05-05
Expiration Date
Abstract
The present disclosure provides systems and methods for detection of cardiac amyloidosis from electrocardiogram (ECG) signals. In particular, the present disclosure identified critical novel features that can be incorporated in systems and methods for the detection of cardiac amyloidosis from one or more ECG signals.
Core Innovation
A computer-implemented method annotates a plurality of electrocardiogram (ECG) parameters in a database as associated with a cardiac amyloidosis designation or a negative cardiac amyloidosis control designation. The cardiac amyloidosis designation encompasses cases labeled with light chain (AL) amyloidosis, transthyretin related (ATTR) amyloidosis, organ limited amyloidosis, lichen amyloidosis, heredofamilial amyloidosis, unspecified amyloidosis, neuropathic heredofamilial amyloidosis, and secondary systemic amyloidosis.
The method instructs a machine learning model to distinguish patterns of selected ECG parameters, including P wave duration/amplitude, R wave duration/amplitude, S wave amplitude, T wave duration/amplitude, PR Interval (PRI) value, and QT value. The distinguishing is performed by training on the annotated electrocardiograms to identify patterns distinguishing each of the eight amyloidosis subtypes from negative controls, thereby linking ECG parameter patterns to amyloidosis subtype-specific characteristics.
The method applies SHAP analysis to the distinguished ECG parameters by computing a numeric value for each parameter representing its contribution to the distinction between cardiac amyloidosis designations and negative controls. This provides numeric contribution values for informative ECG parameters, identifying one or more ECG parameters that are informative of cardiac amyloidosis.
A system receives an electrocardiogram (ECG) signal and extracts one or more of an R wave amplitude in lead V5, an S wave amplitude in lead V3, an R wave amplitude in lead aVL, an R wave amplitude in lead V6, an R wave amplitude in lead V4, an R wave amplitude in lead I, an S wave amplitude in lead V1, and an S wave amplitude in lead V2. A trained machine learning model trained on a plurality of cardiac amyloidosis cases with eight amyloidosis subtype designations applies logic to identify a cardiac amyloidosis pattern and subtype-specific characteristics in the specified amplitudes and outputs a binary classification indicative of a cardiac amyloidosis pattern or a non-cardiac amyloidosis pattern.
Claims Coverage
The document provides two independent claims. One claim focuses on annotating ECG parameters, training a machine learning model to distinguish amyloidosis-subtype patterns from negative controls, and applying SHAP to identify informative parameters. The other claim focuses on extracting specified R- and S-wave amplitudes from selected leads and outputting a binary diagnostic classification indicating a cardiac amyloidosis pattern versus a non-cardiac amyloidosis pattern.
Annotation of ECG parameters with amyloidosis and negative controls for subtype coverage
Annotating a plurality of electrocardiogram (ECG) parameters in a database as associated with a cardiac amyloidosis designation or a negative cardiac amyloidosis control designation, including labeling electrocardiograms with cardiac amyloidosis designations encompassing eight amyloidosis subtype categories.
Machine learning distinction of informative ECG parameter patterns distinguishing eight subtypes from negative controls
Instructing a machine learning model to distinguish one or more parameters selected from P wave duration/amplitude, R wave duration/amplitude, S wave amplitude, T wave duration/amplitude, PR Interval (PRI) value, and QT value based on a pattern present in the one or more cardiac amyloidosis designations that is not present in the negative cardiac amyloidosis control, by training on the annotated electrocardiograms to identify patterns distinguishing each of the eight amyloidosis subtypes from negative controls.
SHAP-based numeric contribution values identifying informative ECG parameters
Applying a SHAP analysis to the distinguished electrocardiogram (ECG) parameters by computing a numeric value for each parameter representing the contribution of that parameter to the distinction between cardiac amyloidosis designations and negative controls, thereby providing numeric contribution values and identifying one or more informative electrocardiogram (ECG) parameters.
Extraction of specified ECG lead amplitudes from an ECG signal
Receiving an electrocardiogram (ECG) signal and extracting one or more of an R wave amplitude in lead V5, an S wave amplitude in lead V3, an R wave amplitude in lead aVL, an R wave amplitude in lead V6, an R wave amplitude in lead V4, an R wave amplitude in lead I, an S wave amplitude in lead V1, and an S wave amplitude in lead V2.
Trained machine learning logic for cardiac amyloidosis pattern and subtype-specific characteristics
Using a trained machine learning model trained on a plurality of cardiac amyloidosis cases with eight amyloidosis subtype designations, to apply logic to identify a cardiac amyloidosis pattern and subtype-specific characteristics in the specified R and S wave amplitudes and to identify the amyloidosis subtype-specific characteristics based on these amplitudes.
Binary output classification indicative of cardiac amyloidosis pattern vs non-cardiac amyloidosis pattern
Outputting a binary classification indicative of a cardiac amyloidosis pattern in the ECG signal or a non-cardiac amyloidosis pattern in the ECG signal.
Across the independent claims, the document covers annotating ECG parameters, training a machine learning model to distinguish amyloidosis-subtype patterns from negative controls, applying SHAP to compute parameter contribution values, extracting specified lead amplitudes, identifying cardiac amyloidosis pattern and subtype-specific characteristics, and outputting a binary diagnostic classification.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?