Machine Learning Tools for Intracardiac Electrogram Mapping

I used signal processing tools: Band pass filters to denoise the signals, and Multiscale Entropy, Multiscale Frequency methods to extract the key features from the irregular heart rhythms

The heat map shows how electrical activity works in different heart regions.

This is the pipeline I developed for performing non-supervised learning analysis on special data types exported from a clinical setting system.

For this coding supplement, I processed arrhythmia recordings exported from the CARTO 3 clinical mapping system as time-series data. Using Python and mathematical libraries, I extracted spatially-aware feature maps from each signal sequence. I then reconstructed the corresponding 3D chamber geometry and overlaid heatmaps of signal entropy and dominant frequency to reveal electrophysiological patterns in context. Finally, I developed and evaluated machine-learning classifiers—implementing SVM and Random Forest models in Python—to distinguish noisy artifacts from true sinus signals and to correlate signal dynamics with key anatomical landmarks.

The classification result shown how the features in mulitdimension could be used to built a classification tool for different rhythms. To further elaborate this idea, we use machine learning tools ro develop the boundaries between clusters.

This is a collaboration with the Mayo Clinic. Here is the link to the paper for further technical details: https://pubmed.ncbi.nlm.nih.gov/40039295/