Cholera is a bacterial disease which induces extremely liquid diarrhea. It affects millions of people, resulting in up to about 150,000 deaths per year. In this study, a sensor is developed which can non-invasively determine if an outbreak may occur in an area, acting as an early detection method so that resources can be employed to stop the rapid spread of the disease. The sensor uses a microphone to collect audio samples of various excretion events. The collected acoustic data are pre-processed to produce mel spectrograms which capture the distinct temporal frequency characteristics of each excretion event. These mel spectrograms are input into a pre-trained convolutional neural network to classify the event as either diarrheal or non-diarrheal with up to 98.1% accuracy. The algorithm is also capable of classifying other excretion events such as urination, flatulence, and defecation. The sensor developed here could be applied to identify other use cases such as tracking bowl movements for hospice patients or for those with inflammatory bowel diseases like Crohn’s disease.
Meeting abstract. No PDF available.
The feces thesis: Using machine learning to detect diarrhea
Maia Gatlin, David S. Ancalle, Anthony Popa, Ashima Taneja, Cade Tyler, David Meyer, David L. Hu, Alexis Noel; The feces thesis: Using machine learning to detect diarrhea. J. Acoust. Soc. Am. 1 October 2022; 152 (4_Supplement): A50. https://doi.org/10.1121/10.0015504
Download citation file: