A group of investigators at PathAI have been able to leverage a pair of artificial intelligence (AI) models to better assess disease activity in patients with ulcerative colitis.

The work, presented by Archit Khosla, compared 2 distinct quantitative approaches to predict disease activity scores and histological remission using AI-powered digital pathology.

The data was presented during the Crohn’s and Colitis Congress 2023 in Denver.

Assessing Disease Activity

Microscopic inflammation is a crucial indicator of disease activity in patients with ulcerative colitis. However, manual histologic scoring is only semi-quantitative and often subject to interobserver variation. In addition, artificial intelligence based solutions often lack interpretability.

The 2 solutions compared in the study were a random forest classifier (RFC) and a graph neural network (GNN). Both these models were expected to identify histological features informing model predictions to provide explainability and biological insight.

In the study, the investigators developed convolutional neural network using > 162k annotations in 820 WSI of H&E-stained colorectal biopsies for pixel-level identification of tissue regions and cell types.

Each WSI was scored by 5 board-certified pathologists using the Nancy Histological Index (NHI) to establish consensus ground truth, while a rich, quantitative set of human interpretable features captured and extracted CNN predictions of the tissue region and cell type across each WSI. They also predicted slide-level NHII scores by training an RFC.

The team tested the hypothesis that tissue region spatial relationships and cellular composition can inform AI-based predictions of disease activity by training a separate GNN using nodes defined by spatially-resolved CNN model-generated outputs to predict NHI scores.

Finally, the investigators calculated feature importance for all combinations of RFC and applied the GNNExplained to locate crucial interactions between regions in the tissue and identify features significantly contributing to GNN predictions.

Results

The results show both models predicted histologic remission with high accuracy (weighted kappa 0.87 and 0.85, respectively) and identified histologic features relevant to disease activity predictions. However, some features, including infiltrated epithelium or neutrophil cell were well-established to distinguish cases with histologic remission.

The models also identified features beyond what was assessed by the NHI, including area proportion of basal plasmacytosis, which was associated with predictions of NHI 2 and 3.

There were some new features not previously implicated in ulcerative colitis disease activity, including intraepithelial lymphocytes, which differentiated cases with NHI 3.

“We report quantitative and interpretable AI-powered approaches for UC histological assessment,” the authors wrote. “CNN identification of UC histology was used as input to two distinct disease activity classifiers that showed strong concordance with consensus pathologist scoring. Both approaches provide interpretable features that explain model predictions and that may be used to inform biomarker selection and clinical development efforts.”

The study, “QUANTITATIVE AND EXPLAINABLE ARTIFICIAL INTELLIGENCE (AI)-POWERED APPROACHES TO PREDICT ULCERATIVE COLITIS DISEASE ACTIVITY FROM HEMATOXYLIN AND EOSIN (H&E)-STAINED WHOLE SLIDE IMAGES (WSI),” was published online by the Crohn’s and Colitis Congress.