Sleep Stage Aware ECG-EEG Combined Apnea Type Classification and Severity Grading Using STG-Saru And FNFRL

Abstract

Sleep apnea is a common sleep disorder that involves periodic interruption of airflow during sleep, resulting in numerous health consequences like cardiovascular diseases and cognitive decline. Current detection techniques are mostly based on polysomnography, which is expensive and impractical for large-scale applications. This paper suggests a new sleep stage-aware apnea type classification and severity grading system that incorporates Electroencephalogram (EEG) and Electrocardiogram (ECG) signals. The approach suggested uses Spatio-Temporal Gated Self-Attention Recurrent Unit (STG-SARU) for classification of sleep stages and detection of apnea and Frobenius Norm Fuzzy Regularized Logic (FNFRL) for grading severity. Minimum Probability Gaussian Mixture Model (MinPro-GMM), Multi-Scale Entropy (MSE) analysis, and Canonical Kernelized Cross-Correlation Approximation (CKCCA) are employed together to extract and fuse discriminatory features from EEG and ECG signals. In addition, noise and artifact removal is maximized through Zero-Crossing Discrete Boundary Smoothing Wavelet Transform (ZDBSWT) and Quasi-Random Independent Sequential Component Analysis (QRISCA). Experimental verification on the PhysioNet Sleep Apnea dataset proves enhanced classification performance, outperforming traditional approaches in accuracy, precision, and robustness. The new framework provides a valid, non-invasive solution for early and accurate sleep apnea diagnosis, facilitating personalized treatment planning.