Long‐term, automated stool monitoring using a novel smart toilet: A feasibility study

Jin Zhou,Yuying Luo,Julia W. Darcy,Kyle J. Lafata,Jose R. Ruiz,Sonia Grego

DOI: https://doi.org/10.1111/nmo.14954

2024-11-03

Neurogastroenterology & Motility

Abstract:Burden and inaccuracy of patients self‐reporting of stool consistency is a challenge. We developed an approach for automatic collection of stool images and computer vision classification providing long‐term monitoring and objective detection of abnormal stool consistency. Background Patients' report of bowel movement consistency is unreliable. We demonstrate the feasibility of long‐term automated stool image data collection using a novel Smart Toilet and evaluate a deterministic computer‐vision analytic approach to assess stool form according to the Bristol Stool Form Scale (BSFS). Methods Our smart toilet integrates a conventional toilet bowl with an engineered portal to image feces in a predetermined region of the plumbing post‐flush. The smart toilet was installed in a workplace bathroom and used by six healthy volunteers. Images were annotated by three experts. A computer vision method based on deep learning segmentation and mathematically defined hand‐crafted features was developed to quantify morphological attributes of stool from images. Key Results 474 bowel movements images were recorded in total from six subjects over a mean period of 10 months. 3% of images were rated abnormal with stool consistency BSFS 2 and 4% were BSFS 6. Our image analysis algorithm leverages interpretable morphological features and achieves classification of abnormal stool form with 94% accuracy, 81% sensitivity and 95% specificity. Conclusions Our study supports the feasibility and accuracy of long‐term, non‐invasive automated stool form monitoring with the novel smart toilet system which can eliminate the patient burden of tracking bowel forms.

gastroenterology & hepatology,neurosciences,clinical neurology