Intellectual Property India Publishes Patent Application for 'Feature-Level Interpretability In Machine Learning Models For Cardiac Arrhythmia Classification' Filed by Dr. Ruhina Quazi; Dr. Rahil Khan; Dr. Vignesh V; Mr. Gunasekar M; Dr. Benjamin Arul S; Ms. L. Steffina Morin; Ms. C. Jeyalakshmi; Ms. Nithya S; Ms. A. Jothi Soruba Thaya; and Dr. Jose Anand A

MUMBAI, India, Jan. 23 -- Intellectual Property India has published a patent application (202521125495 A) filed by Dr. Ruhina Quazi; Dr. Rahil Khan; Dr. Vignesh V; Mr. Gunasekar M; Dr. Benjamin Arul S; Ms. L. Steffina Morin; Ms. C. Jeyalakshmi; Ms. Nithya S; Ms. A. Jothi Soruba Thaya; and Dr. Jose Anand A, Nagpur, Maharashtra, on Dec. 11, 2025, for 'feature-level interpretability in machine learning models for cardiac arrhythmia classification.'

Inventor(s) include Dr. Ruhina Quazi; Dr. Rahil Khan; Dr. Vignesh V; Mr. Gunasekar M; Dr. Benjamin Arul S; Ms. L. Steffina Morin; Ms. C. Jeyalakshmi; Ms. Nithya S; Ms. A. Jothi Soruba Thaya; and Dr. Jose Anand A.

The application for the patent was published on Jan. 23, under issue no. 04/2026.

According to the abstract released by the Intellectual Property India: "This study investigates the interpretability of machine learning models for ECG-based arrhythmia classification using SHAP explainability techniques. Using 12,186 single-lead ECG recordings from the PhysioNet dataset, we extracted 222 clinically meaningful features capturing rhythm variability, waveform morphology, noise characteristics, and time-frequency patterns. An AutoML framework (AutoGluon) was used to identify the optimal classifier, with XGBoost achieving the highest accuracy and F1-score. Six SHAP explainers were systematically compared to determine which method yields the most reliable interpretability, evaluated through explanation error, computation time, and Keep/Remove metrics. Results show that the Permutation Partition and Permutation explainers provide the most accurate explanations, while Tree Approximation, despite its speed, suffers from substantial reconstruction error. The selected SHAP method enabled multi-level interpretation-global, class-specific, and local-revealing that RR interval variability dominates the global importance landscape, while noise and waveform features differentiate rhythm classes. The findings demonstrate that SHAP provides transparent, clinically aligned insights into model behavior and supports trustworthy deployment of arrhythmia classification systems."

Disclaimer: Curated by HT Syndication.