Classification of Cardiac Arrhythmia Diseases from Obstructive Sleep Apnea Signals using Decision Tree Classifier

Abstract

Sleep is a judgmental to health and well-being. Deficient quality sleep is similar to a wide range of negative outcomes that vary from schizophrenia to cardiovascular disorders. Obstructive sleep apnea (OSA) is one of the sleep disorders. OSA is a respiratory episode; it is observed that there is a relationship within the peripheral system such as the cardiovascular system. Both elongated QRS duration and sleep apnea are connected with hypertension, unexpected cardiac death, and heart failure. The objective of the project is to provide a computer-based solution for identifying various cardiac deceases like Bradycardia, Tachycardia from OSA signals using electrocardiogram features. MIT-BIH Polysomnographic and UCD Sleep Apnea Database collected as input signals from the PhysioNet website is used in this study and the implementation of the proposed method is evaluated. In the preprocessing stage, various filters like Wavelet, Median, IIR Notch, and FIR Filter are applied and it is found that Wavelet (sym7) has obtained better results based on evaluation parameters like MSE, SNR, PSNR, etc. The output of the preprocessed signal is smoothened by using the Savitzky- Golay filter. Later RR intervals were detected by using the Pan Tompkins method which is modified in this work. The advantages of using the Pan-Tomkins algorithm compared to other available techniques for feature extraction are the sensitivity and efficiency of the Pan-Tompkins algorithm are more than 99%. Totally 11 features were extracted from the sleep signals and classification is done. By comparing with various classifiers out of them, Decision Tree classifiers have shown with better accuracy of 99.82%, the sensitivity of 94% and specificity of 79.48% in detecting and classifying the Cardiac Arrhythmia.