Hybird RNN based feature extraction for early prediction of CVDs using ECG Signals for type 2 diabetic patients

Abstract

Diabetes mellitus patients are at an increased risk of cardiovascular illness, and cardiovascular complications are the primary cause of morbidity. Diabetes is linked to both morbidity and mortality. Type-2 Diabetes causes a prothrombotic state that leads to acute coronary syndromes by causing endothelial damage and lowering antiaggregant factors like nitric oxide and prostacyclin, as well as increasing thrombotic substances like fibrinogen and factor VII, and suppressing fibrinolysis with factors like plasminogen activator inhibitors. The accurate identification and diagnosis of CVD (Cardio Vascular Disease) is dependent on the correct detection of the ECG signal from the heart. The ECG signal is extremely important in the early detection of cardiac problems. The ECG signal of diabetic individuals offers vital information about the heart and is one of the most important diagnostic tools used by doctors to identify cardiovascular disorders. The time gap between two consecutive QRS complexes appearing contiguous in an ECG is known as heart rate.  The most appealing feature is that HRV (Heart Rate Variability) measurement is non-invasive and repeatable. A number of machine learning techniques have been proposed for the non-invasive automated identification of diabetes. This paper discusses innovative methods for analyzing electrocardiogram (ECG) signals in order to extract important diagnostic information. The ECG signal is first treated using a dual tree complex wavelet transform (DTCWT-SG) with threshold method. Subsequently, the features are extracted from detailed coefficients of DTCWT-SG filter, Eigen vectors by minimum normalization method and Rajan Transform. Main key features are extracted using these three methods. These features are classified and analyzed by different machine learning classifiers. The proposed approach was tested on DICARDIA, MIT-BIH and Physionet database and the performance analysis shows that the hybrid Recurrent Neural Network (RNN) (LSTM+GRU Gated Recurrent Units) achieves better prediction of 98.8% compared to state of art techniques.