Applying novel Supervised Fuzzy Adaptive Resonance Theory (SFART) neural network and Biorthogonal wavelets for ballistocardiogram diagnosis

In this study, we applied biorthogonal wavelets to extract essential features of the ballistocardiogram (BCG) signal and to classify them using a novel neural network so-called supervised fuzzy adaptive resonance theory (SF-ART). SF-ART has two stages. At first stage, pre-classification level, the input data is clustered roughly to arbitrary (M) classes using self-organized fuzzy ART tuned for fast learning. At the second stage, post-classification level, the SF-ART performs supervised clustering using a special array called affine look-up table (ALT) with M elements, which are used to store the labels of corresponding input samples. In testing mode, first the self-organized fuzzy ART classifies the input data roughly. In the next step, the content of an ALT cell with address equal to the index of the first stage´s winning output line will be read. The read value declares the class that input data belongs to. Initial tests with BCG from six subjects (both healthy and unhealthy people) indicate that the method can classify the subjects into three classes with a high accuracy, high learning speed (elapsed time for learning around half second), and very low computational load compared with the well-known neural networks such as multilayer perceptrons (elapsed time for learning above five minutes). The method is insensitive to latency and non-linear disturbance. Moreover, the applied wavelet transform requires no prior knowledge of the statistical distribution of data samples