Standardized, repeatable ulcerative colitis histology scoring and endpoint assessment using an automated, foundation...

Area of Science:

• Gastroenterology
• Digital Pathology
• Artificial Intelligence in Medicine

Background:

• Pathological assessment of ulcerative colitis (UC) severity relies on histological indices like Geboes Score, Robarts Histopathology Index (RHI), and Nancy Histologic Index (NHI).
• Current histological scoring methods exhibit significant inter- and intra-reader variability, hindering consistent interpretation and defining histologic remission criteria.
• There is a need for standardized, reproducible methods to assess UC histology in clinical trials.

Purpose of the Study:

• To develop and validate an artificial intelligence-based measurement (AIM) tool for scoring UC histology in mucosal biopsies (AIM-HI UC).
• To evaluate the performance of AIM-HI UC against expert pathologists in predicting histological parameters and remission states.
• To assess the reproducibility and correlation of AIM-HI UC scores with key histological features of inflammation.

Main Methods:

• An AI model (AIM-HI UC) was developed using additive multiple instance learning leveraging the PLUTO pathology foundation model, trained on 10,230 UC biopsies.
• The model predicts Geboes subgrades, enabling calculation of Geboes grade-level score, RHI, and NHI.
• Performance was evaluated on a separate set of clinical trial specimens, comparing AIM-HI UC agreement with pathologist consensus.

Main Results:

• AIM-HI UC demonstrated non-inferiority to pathologists for all seven Geboes subgrades, grade-level Geboes, RHI, NHI, and histologic improvement/remission.
• The AI tool was superior to pathologists in predicting several Geboes subgrades, grade-level Geboes, RHI, and positive percent agreement for 2A histologic remission.
• AIM-HI UC achieved >99% repeatability and strongly correlated with inflammation markers like epithelial inflammation, crypt neutrophil infiltration, and ulceration.

Conclusions:

• AIM-HI UC shows potential to enhance the consistency and standardization of UC histology interpretation in clinical trials.
• The AI tool offers a reproducible and accurate method for assessing UC disease severity and remission.
• This technology could facilitate more reliable comparisons across studies and improve patient care through standardized diagnostics.