Classification of ballistocardiography using wavelet transform and neural networks

Ballistocardiography (BCG) has the interesting feature that no electrodes are needed to be attached to the body during measurements. This provides a potential application to assess a patient´s heart condition in the home. Artificial neural networks (ANNs) have several properties that make them promising for automatic signal classification problems. In the time domain of the BCG classification, the whole cardiac cycle of BCG waveform needs a large-sized neural network which makes the classification a computationally intensive task. By classifying the data in a compressed format, savings in computer time may be realised. Here, the authors used wavelet multiresolution property analysis that allows one to obtain a significant information content from the BCG signal. Small subsets of the wavelet coefficients were used to classify the normal, hypertension and heart attack risk patients by neural networks. It is shown that the proposed system achieved an overall 95.39% correct classification rate for the test data set. The classification of the features obtained by principal component analysis can only achieve a 92.8% overall performance. The advantage of the proposed classification system is that it can be easily implemented into a portable device