Risk Level Classification of Drug-Induced Arrhythmias and Long-QT Syndrome Diagnosis by SCLMISHCOLU-CNN and SBT-FUZZY

Abstract

Abnormal heart rhythms that are caused by some medications are called Drug-induced Arrhythmias (DA), which also affect the functioning of the heart. In the prevailing works, abnormal heart rhythms are examined grounded on an Electrocardiogram (ECG) signal only, but the consequences of DA are not focused. Therefore, grounded on ECG and arrhythmia’s consequences data, a risk-level classification and a long-QT syndrome diagnosis model are proposed. Primarily, the ECG signal is preprocessed, followed by time series segmentation. Subsequently, to get the accurate P, Q, R, S, and T waves, the segmented signal is transformed into a waveform, and the time intervals between the wavelet components are estimated. Likewise, the segmented signal is also converted to image format, and the image is masked with the full heartbeat. Then, from the masked image, the features are extracted. In addition, the historical DA consequence data are preprocessed. Further, from the DA consequence’s features, masked image features, and wave interval features, the optimal features are chosen. The SCLMish Collapsing Linear Unit-Convolutional Neural Network (SCLMishCoLU-CNN) classifier detects the normal and DA types grounded on the optimal features. Lastly, by using the Singleton Beta Trapezoidal Fuzzy (SBT-Fuzzy) algorithm, the risk level is identified from the detected output; also, grounded on QT intervals, the long-QT syndrome is diagnosed here. The robustness of the framework is proved by the analysis outcomes, thus achieving classification accuracy and precision of 99.79% and 99.49%, correspondingly.