Artificial Intelligence App Outperforms Patient-Reported Stool Assessments

Researchers have determined that an artificial intelligence (AI) algorithm can determine stool characteristics with high accuracy using smartphone images, outperforming patient self-reporting, according to a study published in the American Journal of Gastroenterology this month.

Stool form assessment is a key patient-reported outcome (PRO) used to evaluate gastrointestinal (GI) disorder treatment response, the researchers noted. However, PROs are subjective and can vary significantly between different individuals and within a single individual over time. The subjectivity of other PROs outside of stool assessment, such as abdominal pain, bloating, and nausea, can also impact treatment and placebo response rates.

For clinicians treating irritable bowel syndrome (IBS), the US Food and Drug Administration (FDA) outlines guidance related to reporting and outcomes for stool consistency before and after an intervention. Under this guidance, many gastroenterologists utilize the Bristol Stool Scale (BSS), which ranks stools from one to seven. The FDA also recommends that IBS patients report symptoms and bowel form daily, but the researchers stated that subjects’ interpretations of tools like the BSS could lead to reporting inaccuracies.

To combat this, researchers evaluated an AI-based smartphone application to assess digital images of individual bowel movements. The app is designed to enhance the capture, quality assurance, annotation, and analysis of human stool images. It requires patients to photograph each bowel movement throughout the day rather than attempt to use the BSS or another tool themselves at the end of each day, which is the standard reporting method.

The AI component of the app was trained to predict five clinically relevant stool characteristics based on 30,000 manually annotated stool images: consistency, fragmentation, edge fuzziness, volume, and BSS. To validate the model, 39 IBS with diarrhea (IBS-D) patients participating in a randomized controlled trial for a new drug therapy were asked to report their daily average stool form and use the app to capture images of each of their bowel movements for two weeks.

Two-thirds of the images were recorded on the app and evaluated by the AI component based on the five visual stool characteristics. The remaining one-third were collated and blindly assessed by two gastroenterologists, who also used the same five characteristics for evaluation.

When comparing the AI to each gastroenterologist, the researchers found good agreement between them on BSS, consistency, fragmentation, and edge fuzziness scores, and moderate-to-good agreement on volume. The AI was also found to outperform patient-reported BSS categorization, achieving a 95 percent accuracy rate compared to 89 percent for patients.

At the end of the study, participants were asked two optional questions about their experiences using the app: “Overall, how was your experience using the application in this trial?” and “How do you like tracking your stools in general?”

Twenty subjects responded, with 50 percent describing their experience as “very pleasant and easy” and the remaining 50 percent saying it was “somewhat easy and pleasant.” For the second question, 40 percent of the subjects responded with “doesn't matter to me,” 30 percent with “like it,” 15 percent with “hate it,” 10 percent with “love it,” and the remaining 5 percent with “don't like it.”

These findings show that current PROs for assessing stool are inadequate and that AI assessments could provide more objective outcome measures for stool characterization in gastroenterology, the researchers posited.

Further, since this study was conducted with subjects with IBS, research into the use of such an app in broader GI disease is needed, they added.