Commentary On CEFEs, A CNN Explainable Framework for ECG Signals

Author(s): Professor. Prabhakaran Balakrishnan, Sagnik Dakshit

In healthcare, trust, confidence, and functional understanding are critical for automated decision support systems, therefore, presenting challenges in the prevalent use of black-box deep learning (DL) models. This work attempts to summarize the global interpretable performance explanation methods developed for the task of Arrhythmia classification using deep learning. CEFEs provides a task independent modular framework for investigation of learned features for 1D time-series signals in terms of shape, frequency, and clinical features. The framework allows for functional understanding of the model, quantification of model capacity, and comparison of deep learning models through interpretable explanations. The results also demonstrate understanding of model performance improvement using synthetic signals. While the modules are independent of the task, the tests for each module are task specific and demonstrated for Arrhythmia classification using ECG signals.

View PDF View Fulltext