Detection Of Abnormal Heart Conditions Based On ECG Signal Characteristics

Heart diseases are one of the most important death causes across the globe.Therefore,early detection of heart diseases is crucial to reduce the rising death rate.Electrocardiogram(ECG)is widely used to diagnose many types of heart diseases such as abnormal heartbeat rhythm(arrhythmia),Myocardial infarction(MI),etc.However,the nonlinearity and the complexity of the abnormal ECG signals make it very difficult to detect its characteristics.Besides,it may be time-consuming to check these ECG signals manually.In addition,early detection of some abnormal heart conditions such as MI(also known as a cardiac attack)is critical for the reduction of the rising of the death rate.To overcome these limitations,we have proposed fast and accurate classifier that simulates the diagnosis of the cardiologist to classify the ECG signals into normal and abnormal from single lead ECG signal and better than other well-known classifiers.In addition,we have proposed an effective computer-aided diagnosis(CAD)system to detect MI signals using the two-dimensional convolution neural network(CNN).The first part of this thesis introduces the proposed classifier that simulates the diagnosis of a cardiologist to classify ECG signal data into normal and abnormal classes from single lead ECG signals.In this part,firstly,an accurate algorithm is used for correcting the ECG signals from noise and extracting the major features of each ECG signal.After that,we simulated the characteristics of the ECG signals and created the proposed classifier from these characteristics.Two Neural Network(NN)classifiers,four Support Vector Machine(SVM)classifiers and K-Nearest Neighbor(KNN)classifier are employed to classify the ECG signals and compared with the proposed classifier.The total 13 features extracted from each ECG signal used in the proposed algorithm and set as input to the other classifiers.Finally,the performance of the proposed classifier is presented at the end of this part.In the second part of this thesis,we propose an effective method to detect MI signals using two-dimensional CNN.Firstly,we have employed two ways of the transfer learning technique to retrain the pre-trained VGG-Net and obtained two new networks VGG-MI1 and VGG-MI2.After that,data augmentation techniques are employed to increase the classification performance.Our methods are tested on several publicly available databases for ECG.Experimental results show that the proposed methods are efficient,robust and reliable than existing algorithms.According to the advantages of the proposed method,it can be implemented in clinical settings and can help the experts to detect the MI signals more precisely.