Signal Processing methods for ECG Signal Analysis

ECG signals give important diagnostic information in cardiology. To take advantage of it, the ECGs are to be properly processed for effective analysis and interpretation. New methods for processing and analysis of ECG signals are proposed in this work. In general, the ECG signals are quasi-periodic of low amplitude (several mV). They are often affected by noise signals which include powerline interference, baseline wander, EMG artifacts, etc,. Hence, the data may be corrupted with these noise signals and it is essential to be filtered for the real-time heart monitoring systems. A new method called Gaussian Mean Variant (GMV) filtering is proposed for ECG filtering. Statistical measures like root mean square error (RMSE root mean square deviation (RMSD) and root mean square variation (RMSV) established the efficacy of the proposed method compared to adaptive filters and other conventional methods. The successive step of ECG signal analysis is formation of a feature space. To this end, a method based on Integrated Peak Analyzer (IPA) is proposed for extraction of features, which are useful descriptors that fully capture the essence of the ECG. Based on QRS and PQRS estimations, a set of 33 features were extracted using the proposed method. The extracted features should provide non-redundant information pertaining to ECG and should exhibit high discriminatory power for use in ECG classification. Hence, a novel feature optimization method named Intensity Weighted Fire-Fly Optimization (IWFFO) method is proposed for selecting the best features. The proposed IWFFO method optimized the feature set to 18 from a set of 33 features identified by IPA method representing ECG. The challenging objective of ECG analysis is to identify heart disfunctions and their causes. Classification is the process of recognizing selected features of ECG to determine the abnormalities. An efficient method named Multimodal Decision Learning (MDL) algorithm is proposed for classification of ECGs. All the proposed methods performed well and their efficacy is established by comparing with few existing methods concerned. Test signals were taken from MIT-BIH arrhythmia database and tested on MATLAB? R2012a. In respect of final classification of ECGs, the analysis is carried out by the proposed multimodal decision learning algorithm for detecting the heart beats consisting of 1 normal and 12 abnormalities namely normal sinus rhythm (NSR), paced rhythm (PR), ventricular tachycardia (VT), first degree AV block (FDB), supraventricular tachycardia (SVT), idioventricular tachycardia (IVT), atrial fibrillation (AF), ventricular flutter (VF), ventricular trigeminy (VTR), ventricular bigeminy (VBG), pre-excitation (PE), second degree AV block (SDB) and sinus bradycardia (SBC). To prove the efficacy of the proposed method, the results are compared with Multilayer Neural Networks (MNN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Support Vector Machine (SVM) based algorithms. For the proposed MDL method, the test results revealed an average accuracy of 95.57%, where it is 92.7%, 81.25% and 76.67% for MNN, ANFIS and SVM methods respectively.