In a recent study published in the American Journal of Gastroenterology, researchers at Cedars-Sinai Medical Center in the United States evaluated an artificial intelligence (AI)-based smartphone application (app) trained to assess a patient's stool characteristics.
Study: A Smartphone Application Using Artificial Intelligence Is Superior To Subject Self-Reporting When Assessing Stool Form. Image Credit: Josep Suria / Shutterstock
Functional gastrointestinal (GI) disorders, especially luminal ones, require that a patient self-report stool form and frequency. However, since the symptoms of diarrhea common in irritable bowel syndrome with diarrhea (IBS-D) patients are subjective, the inability to accurately report or assess stool form and frequency makes it challenging to determine the effectiveness of therapeutic interventions in these conditions.
The Bristol Stool Scale (BSS) is the United States Food and Drug Administration (US-FDA) approved 7-point scale that ranks stool consistency from 1 (hard lumps) to 7 (liquid). However, inconsistent and inaccurate self-reporting of stool forms causes issues, especially among cases of IBS-D. In such cases, AI algorithms could help systematically assess digital images of a person's bowel movements.
In the present study, researchers enrolled subjects participating in a randomized clinical trial for IBS-D to validate AI determinations for stool images based on five distinct visual stool characteristics, viz., edge fuzziness, consistency, BSS, volume, and fragmentation. In another set of individuals from the same trial, they assessed how the app findings aligned with the self-reported BSS scores. Lastly, the team compared subject-determined BSS and AI-determined stool characteristics scores with standardized diarrhea severity scores.
The participating subjects captured all stool images during the two-week screening phase of the trial. The app processed the results and determined five visual stool characteristics and bowel frequency. Two experts validated AI images from the first one-third of the subjects. Later, the team also graded stool images annotated by AI, self-reported by the study participants and two experts, into categories, BSS <3 (constipation), BSS ≥ 3, but BSS ≤5 (normal), and BSS >5 (diarrhea). Lastly, the team computed sensitivity, specificity, accuracy, and diagnostic odds ratios of self-reported and AI-graded BSS scores by comparing them with experts' evaluations, which they considered the gold standard.
There were a total of 39 study participants, of which 14 provided 219 stool images for the validation phase. The team used data from the other 25 subjects for the implementation phase. Both AI and expert gastroenterologists presented BSS scores from one to seven and their evaluations were in good agreement for all five stool characteristics and so were AI and expert evaluations.
The average specificity and sensitivity rates of AI-graded BSS score categorization were 11% and 16% higher, respectively. The mean diagnostic odds ratio and accuracy rate was higher for AI at 30.64 vs. 3.67 and 95% vs. 89% compared to subject-reported scores. The agreement between subject-reported and AI-graded BSS scores was 0.31 during the validation phase but attained a value of 0.61 during the implementation phase. On average, the visual stool characteristics determined by AI between the two phases remained similar.
Further, the authors observed a good correlation between the AI-graded daily average BSS scores and diarrhea severity scores in IBS-D subjects. The other four visual stool characteristics reported by the app also correlated rather well with diarrhea severity scores. Notably, all the subjects found the app easy to use, and 50% of those who responded to queries regarding user experience described their experience as easy and very pleasant.
Earlier, IBS drug testing often depended on weekly GI symptom assessments. Later, the US FDA developed new guidelines for IBS, which mandated trial sponsors to ask all the participants to report and characterize symptoms daily to improve accuracy. Yet, BSS remains critical in evaluating self-reported stool hardness and rating individual stool types during clinical trials.
An inaccurate self-reporting of stool form could stem from inadequate subject understanding and recall bias. Although intuitive, the patient needs to be familiarized with BSS to avoid misperception. It becomes challenging when subjects with diarrhea report a daily average BSS while they have several varied bowel movements in a day. The current study findings support that self-reported daily scores differed from BSS scores given by the two experts.
AI catalogs characterized stool form in an objective 'true' sense, given a subject photo-documented each bowel movement. The digital stool images allowed full assessment of a drug effect and objectively quantifying the side effects of therapies for bowel disorders. Also, these images assessed the stool characteristics beyond the BSS. The evaluation of four novel characteristics facilitated considering each bowel movement separately and avoided the need to collate daily averages. Overall, the observed pattern of test characteristics suggested that AI results were superior. Further, they reduced trial costs as sponsors could now design trials with a large number of subjects to dampen the effect of inconsistency and inaccuracy of self-reporting. Accordingly, in the future, objective measurement of stool form would need fewer subjects to test drugs.
To conclude, the AI-based app used in the study accurately characterized stool compared with self-reporting and correlated well with diarrhea severity. It has the potential to become a valuable tool for use in trials of luminal GI diseases, including IBS-D, as it was both accurate and objective in defining stool features beyond the BSS.
