The Detection and Recognition of Electrocardiogram’s Waveform Based on Sparse Decomposition and Neural Network

 Electrocardiogram (ECG) is an important method of diagnosis. Using computer to detect and recognize all kinds of wave of ECG can release doctor’s tension coming from more and more ECG data, and decrease error because of tiredness, negligence, and subjective warp. Based on Gabor dictionary modified and the particle swarm optimization, ECG data are decomposed by the theory of sparse decomposition. Comparing of original ECG, its solution vector of sparse decomposition is very sparse (only a few components are nonzero). Every nonzero value corresponds to an atom selected form dictionary, the atom represents the characters of ECG. Based on magnitude of nonzero value in the solution vector and waveform of atom corresponding, some information, such as amplitude, wave width, waveform, position, can be found out. Then using prior knowledge of ECG, it can be made sure which kinds of waveform represented by atom (P wave, QRS complex or T wave). And then, it is constructed that training sample of neural network (NN). Passing through training, NN will complete automatic detection and recognition of ECG. Experiments have proved that this algorithm could complete P wave, QRS complex and T waves’ automatic detection and recognition of ECG, at the same time.