Deep residual attention network for human defecation prediction using bowel sounds

Background and objective Fecal incontinence may lead to incontinence-associated dermatitis (IAD), affecting the physical health of the patient. Since human defecation is related to intestinal activity, and bowel sounds can reflect bowel motility, a prediction method for human defecation based on deep residual attention network (DRAN) using bowel sounds was proposed to prevent IAD. Methods We collected 1140 bowel sounds of 20 seconds from 15 volunteers. These bowel sounds were transformed into time-frequency maps by wavelet packet transform (WPT). Then the time-frequency maps are taken as input to the DRAN. DRAN classified bowel sounds to predict whether the patient would defecate. Results After training, the defecation prediction accuracy, precision and recall of DRAN could reach 91.18%, 91.67% and 90.59% respectively, Conclusion The result indicates that the proposed method could provide a high-precision defecation prediction result for patients with fecal incontinence, so as to prepare for defecation in advance and prevent the occurrence of IAD.