Bowel sound based digestion state recognition using artificial neural network

A new method of recognizing the digestion states from the bowel sound signals is proposed by using an artificial neural network. The bowel sound signals are filtered using a two-step adaptive filter first for noise cancelling. The network uses the extracted features from the processed signals as the input. The adaptive filter can effectively cancel both the environment noise and the body noise/disturbance with the Least Mean Square algorithm (LMS) and an improved two-stage structure. Unlike the traditional studies, features in both the time and frequency domains are extracted to form a feature vector with 420 elements, to train the back-propagation neural network with 60 hidden neurons. The threshold detection block compares the function outputs that reflect the continuous variation of intestinal motility to well-chosen threshold values and tell the digestion states. The respective experiments on three volunteers show that this method has a personal recognition rate of over 70 % with the proposed network.