The categorization of non–toilet-trained children’s stool consistency remains difficult. The purpose of this study was to see if it was possible to use machine learning to automate the categorization of stool consistencies from diaper pictures (ML). Following independent ethical research clearance, 2687 usable smartphone photographs of diapers with faeces from 96 children younger than 24 months were collected. Stool consistency was rated separately by study participants and two researchers from each photo using the original seven kinds of the Brussels Infant and Toddler Stool Scale. Using transfer learning to re-train the classification layer of a pretrained deep convolutional neural network model, a proof-of-concept ML model was created on this gathered photo set. The agreement rate between study participants and both researchers was 58.0 percent and 48.5 percent, respectively, and 77.5 percent between researchers. The model correctly categorised 60.3 percent of the test pictures, matching the final score. The agreement between model-based and researcher categorization was 77.0 percent for the 4-class grouping of the 7 Brussels Infant and Toddler Stool Scale categories.
The ML model’s automated and objective assessment of stool consistency from diaper pictures demonstrates strong agreement with human raters and overcomes the drawbacks of existing approaches that rely on caregiver reporting. This novel framework for photo database generation and ML classification, which is integrated with a smartphone application, has various potential applications in clinical research and home evaluation.
