Classification of pre- and post-defecation states using intestinal peristalsis sounds and machine learning

Excretory care is one of the most demanding and burdensome tasks for caregivers. If defecation timing can be predicted with accuracy, it may greatly enhance the efficiency of nursing care. In this study, intestinal peristalsis sounds were recorded before and after defecation and acoustical parameters were compared between these periods. In addition, the feasibility of discrimination of pre- and post-defecation based on peristalsis sounds was evaluated using supervised machine learning. Intestinal peristalsis sounds were recorded from 10 healthy male subjects seated in an anechoic chamber, using four stethoscope-microphones attached to the subjects’ abdomens. Subjects had controlled diets 2 days prior to measurement. Acoustic features were extracted from data for large peristaltic sound peaks and surrounding periods. After selecting significant components using principal component analysis, Random Forest was used to classify before and after defecation. The results showed that the intensity of intestinal peristalsis sound was significantly higher before defecation than after. Also, the classification accuracy of pre- and post-defecation using machine learning was approximately 70%, and the most important features identified were kurtosis, rms, and low-frequency energy. These findings suggest the potential for predicting defecation based on intestinal peristalsis sounds, although further improvement in classification performance is needed.