A Topology Informed Random Forest Classifier for ECG Classification

Abstract

This paper accompanies Team Cordi-Ak's entry to the classification of 12-lead ECGs for the PhysioNet/Computing in Cardiology Challenge 2020. Our approach leverages mathematically computable topological signatures of 12-lead ECGs as proxy for features informed by medical expertise to train a two-level random forest model in a multi-class classification task. We view ECGs as multivariate time series data and convert different segments and groupings of leads to point cloud embeddings. This stores both local and global structures of ECGs, and encodes periodic information as attractor cycles in high-dimensional space. We then employ topological data analysis on these embeddings to extract topological features based on different summaries available in the literature. We supplement these features with demographic data and statistical moments of RR intervals based on the Pan-Tompkins algorithm for each lead to train the classifier. Our classifier achieved a challenge validation score of 0.304, and a final score of -0.113 on the full hidden test data, placing us 37th out of 41 officially ranked teams that participated in this year's Challenge.