Non Invasive Anomaly Detection of ECG Signals with Emphasis on Diabetes and Deep Learning Techniques

Abstract

Biomedical signals contain useful information about the activity of different parts of the body. Biomedical signals are basically non-linear and non-stationary in nature. Hence it is very difficult to get useful information from these signals, directly in the time domain, just by observing them. Hence, signal processing techniques are employed to extract important features from these signals for the diagnosis of different diseases. These non-invasive automated techniques which exhibit very high accuracy of detection of anomaly, can serve as an assisted tool to clinicians to diagnose and manage several diseases. Biosignal analysis was mainly done earlier employing machine learning based methods. Now, shift to deep learning methods is happening. ECG (Electrocardiogram) signal indicates the working of autonomic nervous system (ANS) which regulates the normal rhythm of heart. Like any other biosignal, ECG signal are also non-linear and non-stationary in nature. The analysis of ECG gives a clue to different diseases. One of the important anomalies which can be detected from ECG derived signal is diabetes. Diabetes is incurable. If diabetes is not controlled properly with medications /insulin support, it is likely to worsen to severe complicated conditions like stroke, kidney failure and heart attack. Timely detection is of great importance in managing diabetes. The main objective of this thesis is to find out methods to diagnose diabetes using heart rate variability (HRV) signals employing deep learning so as to get very high accuracy of detection. We devise HOS based machine learning method, then deep learning based method and then an improvement of our deep learning work to achieve increased accuracy values to achieve automated diabetes detection using HRV signals. As auxiliary works, diagnosis of arrhythmia and further diagnosis of atrial fibrillation were done from ECG signals. A review book chapter which tells details about non-invasive automated diabetes diagnosis methods with HRV signals as input and another..