The treatment goals for Ulcerative Colitis (UC) are becoming increasingly stringent. Although not yet an official target, multiple studies have demonstrated that histological remission plays a crucial role in maintaining clinical remission, and experts emphasize the importance of comprehensive disease control incorporating histological outcomes1,2. At this year's ECCO Congress in Berlin, I presented a study on the application of digital pathology and artificial intelligence (AI) from a gastroenterology perspective. During the preparation of this research, I delved into relevant background knowledge, which provided insights into the ongoing discussions and current research trends in this field.
When considering histological remission as a treatment target in gastroenterology, two primary concerns arise. Drawing from my prior research on endoscopic disease activity3 and prognostic predictions, I aim to outline the background that led to this study.
The first concern is the objectivity of histological assessment tools and the reliance on pathologists' interpretations. Well-established histological scoring indices include the Geboes score4, the Robarts Histopathology Index5 and the Nancy Histological Inde6. However, each index employs different methodologies for evaluating disease activity, which inherently leads to inter-observer variability. The second concern pertains to the role of histological assessment in treatment decision-making. Even if histological remission is established as a stringent treatment target, how should clinicians adjust therapy when histological interpretations are not quantified into a numerical score? Should treatment modifications be based solely on endoscopic findings and clinical symptoms in such cases? AI-driven central reading systems could serve as a crucial solution to these challenges. These systems can mitigate the time-consuming nature of pathology evaluations while reducing inter-observer variability, thereby enabling more standardised and objective treatment decisions.
AI-based research in digital pathology has been expanding7. One notable study utilised the PICaSSO Histological Remission Index model, which applies a neutrophil-only approach to analyse whole slide images (WSI)8. This method compresses feature data from selected image patches, converges them into a linear model and applies weighted values to predict patient outcomes. Another study proposed a different approach: initial evaluation of inflammation activity on WSI by pathologists is followed by cropping and analysis of relevant areas, and the results are then interpreted using class activation mapping to visualise neutrophil location and density9. These AI-based models have demonstrated comparable performance, highlighting the need to shift discussions toward their clinical implementation. However, several unresolved issues remain. For instance, the presence of neutrophils in the lamina propria is a clear indicator of active inflammation, but the relationship between chronic inflammatory cell infiltration and UC disease activity is not yet well defined. Notably, the role of eosinophils remains controversial. Some studies report that eosinophil counts decrease with treatment, while others suggest that higher eosinophil counts correlate with better responses to anti-TNF therapy. The fact that eosinophil assessment is included as a subscore in the Geboes score raises the question of whether evaluating disease activity based solely on neutrophils is sufficient10. Although annotating individual cells remains a labour-intensive process in research settings, as histological assessment criteria become more rigorous and prediction accuracy gains in importance, such methods could ultimately become valuable clinical tools.
The multidisciplinary approach fostered by the ECCO Community provides an ideal research platform for IBD management. The ability to translate clinical concerns into research initiatives and share findings within this collaborative framework is invaluable.
I would like to express my sincere gratitude to Aart Mookhoek, Chair of H-ECCO, for the opportunity to contribute to this ongoing discourse.
