Search

US-12620488-B2 - Machine learning algorithm for the detection of cardiac amyloidosis from 12 lead ECG data

US12620488B2US 12620488 B2US12620488 B2US 12620488B2US-12620488-B2

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.

Inventors

  • Sluso Anne Bose
  • Catherine Vinnarasi Antony
  • Komala T.A
  • Rozina Moazzam Ali Shaikh

Assignees

  • AccurKardia, Inc.

Dates

Publication Date
20260505
Application Date
20241213

Claims (20)

  1. 1 . A computer-implemented method for identifying one or more electrocardiogram (ECG) parameters(s) that are informative of cardiac amyloidosis and generating a diagnostic classification, comprising: annotating a plurality of electrocardiogram (ECG) parameters in a database as associated with a cardiac amyloidosis designation or a negative cardiac amyloidosis control designation, wherein annotating comprises labeling a plurality of electrocardiograms with one or more cardiac amyloidosis designation, wherein the cardiac amyloidosis designation encompasses cases labeled with one or more of: light chain (AL) amyloidosis, transthyretin related (ATTR) amyloidosis, organ limited amyloidosis, lichen amyloidosis, heredofamilial amyloidosis, unspecified amyloidosis, neuropathic heredofamilial amyloidosis, and secondary systemic amyloidosis; instructing a machine learning model to distinguish the plurality of electrocardiogram (ECG) parameters, wherein the machine learning model is instructed to distinguish one or more parameters selected from a group comprising a P wave duration, a P wave amplitude, an R wave duration, an R wave amplitude, an S wave amplitude, a T wave duration, a T wave amplitude, a PR Interval (PRI) value, and a QT value based on a pattern present in the one or more cardiac amyloidosis designation that is not present in the negative cardiac amyloidosis control by training the machine learning model on the annotated electrocardiograms to identify patterns distinguishing each of the eight amyloidosis subtypes from negative controls; applying a SHAP analysis to the plurality of distinguished electrocardiogram (ECG) parameters, wherein applying the SHAP analysis comprises computing a numeric value for each parameter that represents the contribution of that parameter to the distinction between cardiac amyloidosis designations and negative controls, thereby providing a numeric value that represents the contribution of each parameter to the one or more cardiac amyloidosis designation and identifies one or more electrocardiogram (ECG) parameters(s) that are informative of cardiac amyloidosis.
  2. 2 . The method of claim 1 , wherein the P wave duration corresponds to the P wave duration in lead I, the P wave duration in lead II, or the P wave duration in lead III.
  3. 3 . The method of claim 1 , wherein the P wave duration corresponds to the P wave duration in lead V 1 , the P wave duration in lead V 2 , the P wave duration in lead V 3 , the P wave duration in lead V 4 , the P wave duration in lead V 5 , or the P wave duration in lead V 6 .
  4. 4 . The method of claim 1 , wherein the P wave duration corresponds to the P wave duration in lead aVF, the P wave duration in lead aVR, or the P wave duration in lead aVL.
  5. 5 . The method of claim 1 , wherein the P wave amplitude corresponds to the P wave amplitude in lead I, the P wave amplitude in lead II, or the P wave amplitude in lead III.
  6. 6 . The method of claim 1 , wherein the P wave amplitude corresponds to the P wave amplitude in lead V 1 , the P wave amplitude in lead V 2 , the P wave amplitude in lead V 3 , the P wave amplitude in lead V 4 , the P wave amplitude in lead V 5 , or the P wave amplitude in lead V 6 .
  7. 7 . The method of claim 1 , wherein the P wave amplitude corresponds to the P wave amplitude in lead aVF, the P wave amplitude in lead aVR, or the P wave amplitude in lead aVL.
  8. 8 . The method of claim 1 , wherein the R wave duration corresponds to the R wave duration in lead I, the R wave duration in lead II, or the R wave duration in lead III.
  9. 9 . The method of claim 1 , wherein the R wave duration corresponds to the R wave duration in lead V 1 , the R wave duration in lead V 2 , the R wave duration in lead V 3 , the R wave duration in lead V 4 , the R wave duration in lead V 5 , or the R wave duration in lead V 6 .
  10. 10 . The method of claim 1 , wherein the R wave duration corresponds to the R wave duration in lead aVF, the R wave duration in lead aVR, or the R wave duration in lead aVL.
  11. 11 . The method of claim 1 , wherein the R wave amplitude corresponds to the R wave amplitude in lead I, the R wave amplitude in lead II, or the R wave amplitude in lead III.
  12. 12 . The method of claim 1 , wherein the R wave amplitude corresponds to the R wave amplitude in lead V 1 , the R wave amplitude in lead V 2 , the R wave amplitude in lead V 3 , the R wave amplitude in lead V 4 , the R wave amplitude in lead V 5 , or the R wave amplitude in lead V 6 .
  13. 13 . The method of claim 1 , wherein the R wave amplitude corresponds to the R wave amplitude in lead aVF, the R wave amplitude in lead aVR, or the R wave amplitude in lead aVL.
  14. 14 . The method of claim 1 , wherein the S wave amplitude corresponds to the S wave amplitude in lead I, the S wave amplitude in lead II, or the S wave amplitude in lead III.
  15. 15 . The method of claim 1 , wherein the S wave amplitude corresponds to the S wave amplitude in lead V 1 , the S wave amplitude in lead V 2 , the S wave amplitude in lead V 3 , the S wave amplitude in lead V 4 , the S wave amplitude in lead V 5 , or the S wave amplitude in lead V 6 .
  16. 16 . The method of claim 1 , wherein the S wave amplitude corresponds to the S wave amplitude in lead aVF, the S wave amplitude in lead aVR, or the S wave amplitude in lead aVL.
  17. 17 . The method of claim 1 , wherein the T wave amplitude corresponds to the T wave amplitude in lead I, the T wave amplitude in lead II, or the T wave amplitude in lead III.
  18. 18 . The method of claim 1 , wherein the T wave amplitude corresponds to the T wave amplitude in lead V 1 , the T wave amplitude in lead V 2 , the T wave amplitude in lead V 3 , the T wave amplitude in lead V 4 , the T wave amplitude in lead V 5 , or the T wave amplitude in lead V 6 .
  19. 19 . The method of claim 1 , wherein the T wave amplitude corresponds to the T wave amplitude in lead aVF, the T wave amplitude in lead aVR, or the T wave amplitude in lead aVL.
  20. 20 . The method of claim 1 , wherein the PR interval is a value in lead I, wherein the PR interval is a value in lead II, or wherein the PR interval is a value in lead III.

Description

FIELD The present disclosure relates to systems and processes for detection of cardiac amyloidosis. Cardiac Amyloidosis is a progressive restrictive cardiomyopathy that leads to heart failure and poor patient prognosis. Amyloidosis arises from the mis-folding of precursor proteins that become insoluble and deposit in the tissues, including heart muscle. The vast majority of cardiac amyloid cases result from light chain fibrils deposition (AL Amyloid) or Transthyretin (ATTR Amyloid) deposition. SUMMARY This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Other features, details, utilities, and advantages of the claimed subject matter will be apparent from the following written Detailed Description including those aspects illustrated in the accompanying drawings and defined in the appended claims. Cardiac amyloidosis is a protein misfolding disorder involving deposition of amyloid fibril proteins in the heart. The associated fibrosis of the conduction tissue results in conduction abnormalities and arrhythmias. “Classical” electrocardiogram (ECG) findings in cardiac amyloidosis include that of low voltage complexes with increased left ventricular wall thickness on echocardiography. However, this “classical” finding is neither sensitive nor specific. Nor is it consistent: conventional wisdom in recent literature suggests that ECG patterns are inconsistent in cardiac amyloidosis patients. Ng, P. L., Lim, Y. C., Evangelista, L. K., Wong, R. C., Chai, P., Sia, C. H., Loi, H. Y., Yeo, T. C., Lin, W. (2022). See, e.g., utility and pitfalls of the electrocardiogram in the evaluation of cardiac amyloidosis. Annals of Noninvasive Electrocardiology, 27, e12967. See also, Cappelli F, Vignini E, Martone R, Perlini S, Mussinelli R, Sabena A, Morini S, Gabriele M, Taborchi G, Bartolini S, Lossi A, Nardi G, Marchionni N, Di Mario C, Olivotto I, Perfetto F. Baseline ECG Features and Arrhythmic Profile in Transthyretin Versus Light Chain Cardiac Amyloidosis. Circ Heart Fail. 2020. In some aspects, the invention describes systems and processes for detection of cardiac amyloidosis from one or more meticulously selected electrocardiogram (ECG) parameters (i.e., added as features into a machine learning model). The disclosure describes the process for annotation of a large database (approximately 1 million patients) with a label associated with one or more cardiac amyloidosis diagnosis or a negative control non-amyloidosis diagnosis. The designation for the cardiac amyloidosis can be selected from, e.g., the group consisting of: light chain (AL) amyloidosis, transthyretin related (ATTR) amyloidosis, organ limited amyloidosis, lichen amyloidosis, heredofamilial amyloidosis, unspecified amyloidosis, neuropathic heredofamilial amyloidosis, and secondary systemic amyloidosis. Notably, the selection of features including a P wave duration, a P wave amplitude, an R wave duration, an R wave amplitude, an S wave amplitude, a T wave duration, a T wave amplitude, a PR Interval (PRI) value, or a QT value is demonstrated by the disclosure to provide a suitable basis for the development of a system where a feature (or a subset of features) strongly associated with a cardiac amyloidosis designation correctly identifies an ECG from a subject afflicted with cardiac amyloidosis with high sensitivity and specificity. In some aspects, the disclosure provides, a method for identifying one or more electrocardiogram (ECG) parameters(s) that are informative of cardiac amyloidosis, comprising: annotating a plurality of electrocardiogram (ECG) parameters in a database as associated with one or more cardiac amyloidosis designations or a negative cardiac amyloidosis control designation, whereby the designation for the cardiac amyloidosis is selected from the group consisting of: light chain (AL) amyloidosis, transthyretin related (ATTR) amyloidosis, organ limited amyloidosis, lichen amyloidosis, heredofamilial amyloidosis, unspecified amyloidosis, neuropathic heredofamilial amyloidosis, and secondary systemic amyloidosis; instructing a machine learning model to distinguish the plurality of electrocardiogram (ECG) parameters, wherein the machine learning model is instructed to distinguish one or more parameters selected from a group comprising a P wave duration, a P wave amplitude, an R wave duration, an R wave amplitude, an S wave amplitude, a T wave duration, a T wave amplitude, a PR Interval (PRI) value, and a QT value based on a pattern present in the one or more cardiac amyloidosis designation(s) that is not present in the negative cardiac amyloidosis control; applying a machine learning (e.g., SHAP) analysis to the plurality of distinguished electrocardiogram (ECG) parameters thereby pro