Class-Oriented and Subject-Oriented Approaches for Arrhythmia Classification Using Two-Channels ECG with MobileNet-Based CNN

Sudden Cardiac Death (SCD) is a significant global public health concern, accounting for approximately 15-20% of all fatalities. It often arises due to severe complications accompanied by arrhythmias. The use of a portable Electrocardiogram (ECG), known as a Holter Monitor, is common among doctors for conducting diagnosis related to arrhythmias. Extensive research has been conducted in the development of ECG-based arrhythmia detection systems to reduce the time cost of observation process. Research that specifically detects premature contractions (PC) and bundle branch blocks (BBB) and adopts a subject-oriented approach has been undertaken, but all of that research used the MLII channel. This paper presents experiments classifying arrhythmias using a two-channels ECG with a MobileNet-based CNN. Both class-oriented and subject-oriented approaches were employed to determine the performance of the model. The class-oriented approach achieves an impressive 97.9% accuracy. However, subject-oriented models, while excelling with Normal and LBBB at sensitivity with values of 92% and 90% respectively, it struggle in certain classes such as RBBB, PAC, and PVC with values of 47%, 7%, and 70%, respectively. Nevertheless, this research offers valuable insights, with specificity for the normal class remaining high at 79%. This proves advantageous, as misclassifying arrhythmia as normal is more critical.