Home > Gastroenterology > ECCO 2022 > Miscellaneous Topics > AI model distinguishes between histologic activity and remission in ulcerative colitis

AI model distinguishes between histologic activity and remission in ulcerative colitis

Presented by
Dr Tommaso Parigi, Humanitas University, Italy
ECCO 2022
A novel artificial intelligence (AI) model was able to distinguish between histological remission and disease activity in biopsies of participants with ulcerative colitis (UC). The model can improve the histological evaluation of patients in clinical practice; an evaluation that is key in distinguishing mild activity from remission.

Although there are over 30 histological score systems, the use of these tools in clinical practice is limited, according to Dr Tommaso Parigi (Humanitas University, Italy). The largest obstacles for daily practice are the high interobserver variability and the complexity of the available score systems. The current study aimed to develop a simple and reliable histological score that is implementable in an AI system. The next objective was to develop an AI model to distinguish histological activity from remission. In total, 614 biopsies from the PICaSSO study were used to develop the score system [1,2].

The novel PICaSSO Histologic Remission Index (PHRI) measures the presence of neutrophils in epithelium and lamina propria. PHRI displayed a correlation with endoscopic assessment ranging between 0.69–0.78, depending on the endoscopic score system that was used for comparison. Moreover, it shows excellent inter-reader agreement (ICC 0.84). Subsequently, the authors created an AI model based on the PHRI to detect neutrophils and discriminate between disease activity and remission. The data of the AI model was compared with the gold standard of an annotation by a human pathologist.

Neutrophil detection via AI had a sensitivity of 0.72, a specificity of 0.84, a positive predictive value of 0.75, a negative predictive value of 0.83, and an accuracy of 0.80. Furthermore, the AI model showed a high specificity (0.94) and positive predictive value (0.90) for the detection of disease activity. The sensitivity, negative predictive value, and accuracy for detecting disease activity were 0.62, 0.73, and 0.79, respectively.

Dr Parigi added that this AI model might eventually replace the pathologist for histologic assessments. However, the sensitivity needs to be improved, and the AI should be further trained to predict other histologic scores.

  1. Villanacci V, et al. A new simplified histology artificial intelligence system for accurate assessment of remission in Ulcerative Colitis. OP15, ECCO 2022, 16–19 February.
  2. Iacucci M, et al. Gastroenterology. 2021;160(5):1558–1569.

Copyright ©2022 Medicom Medical Publishers

Posted on