Visual Explanations of Deep Learning Architectures in Predicting Cyclic Alternating Patterns Using Wavelet Transforms

Cyclic Alternating Pattern (CAP) is a sleep instability marker defined based on the amplitude and frequency of electroencephalogram signal. Because time intensive process labeling data, different machine learning automatic approaches are proposed. However, due to low accuracy traditional approach black box approach, proposed systems remain untrusted by physician. This study contributes accurately estimating CAP in Frequency-Time domain A-phase its subtypes prediction transforming monopolar deviated signals into corresponding scalograms. Subsequently, various computer vision classifiers were tested for using scalogram images. It was found that MobileNetV2 outperformed all other achieving average accuracy, sensitivity, specificity values 0.80, 0.75, 0.81, respectively. The trained model further fine-tuned prediction. To verify visual ability models, Gradcam++ employed identify targeted regions network. verified areas identified match focused experts predictions, thereby proving clinical viability robustness. motivates development novel deep methods patterns predictions.