الفهرس | Only 14 pages are availabe for public view |
Abstract A reliable automatic system for classifying multichannel electrocardiographic (ECG) heartbeats is presented. The dataset was collected from the American Heart Association (AHA) database. Five different beats were selected from the AHA database: normal beats (N), paced beats (P), ventricular beats (V), atrial flutter beats (F), and a sinus bradycardia beats (R), a number of beats of each type was manually separated from each channel of the two channels of the AHA database. These separated heartbeats were then processed using signal processing techniques. Features are extracted from the isolated beats using five different techniques: time domain features, statistical features, morphological features, frequency domain features, and wavelet domain features. Each of the five feature vectors was inputted to an artificial neural network. Two topologies were utilized to design the neural classifier and their performance was compared. It is concluded that a neural network classifier of five output neurons and a Meyer mother function wavelet features provided the best correct classification (CC) rate; up to 96.2%, with true positive of 96.12%, false positive of 3.79%, true negative of 4.76%, and false negative of 3.73%. In an attempt to improve the classification results, a data fusion approach at the classifier level technique was used for further improvement in the correct classification CC rate. As a result of using different data fusion techniques with a set of score normalization methods, it has been found that classification correct rate reaches 99.12%. The obtained results are satisfactory and higher than those reported by other researchers |