Real-Time CNN Based ST Depression Episode Detection Using Single ECG-Lead

LAKSHMAN TAMIL1, Erhan Tiryaki1, Akshay Sonawane2
1University of Texas at Dallas, 2The University of Texas at Dallas


As an ST segment derivation, ST segment depression is one of the indications for Ischemic Heath Disease. In this study, we proposed a CNN-based ST depression detection algorithm for different length of depression episodes. The certain parts of consecutive ECG beats are used for forming matrixes which act as input images for CNN. The use of different number of consecutive ECG beats enables the method to detect different duration of ST segment depression episodes. The algorithm is evaluated using European ST-T Database. Binary classifications are performed to distinguish normal and ST depression episodes for subject independent and subject dependent analysis. The best results in subject independent analysis are shown as 0.95, 0.98, and 0.91 for accuracy, sensitivity, and specificity consecutively. The best results for two subjects are seen better than subject independent results. Finally, we developed a model using proposed method and deployed it to a mobile application for real time usage.