Electrocardiogram signal processing and classification using neural network computing

Abstract

Neural network computing is applied into medical electrocardiogram signal processing, classification and diagnosis. The electrocardiogram (ECG) is the measurement of the electrical activity of the heart, and it is called the language of the heart, since from which the heart function and abnormality can be assessed, and this analysis process can be greatly assisted by current computer technology.
In this thesis, artificial neural networks are introduced, and applied into the analysis and diagnosis of ECGs. The heart rate diagram is produced using the Backpropagation Neural Network (BPNN). A diagnosis conclusion of certain disease according to the heart rate is given here. Then the beat classification result for any types of record is achieved by the Adaptive Resonance Theory (ART) Network, by which thousands of QRS complexes (the main part of the ECU) are classified into less than sixty-six categories, these classification results can be treated as a simplification of the original record.
All the test records are taken from the MIT-BIH Database stored in a particular CD. In this thesis, the basic knowledge of ECU diagnosis and Neural Network computing is introduced in the early chapters. The principles of the BPNN and ART network are demonstrated in the middle chapters, the program results are provided at the end of the thesis.