An AI tool achieved a high correlation with histopathologists for assessing mucosal inflammation, reinforcing the potential of AI for clinical use in IBD.
Mucosal healing is an emerging treatment goal in the management of ulcerative colitis. One of the most widely used scores to evaluate this endpoint is the Nancy score, which allows assessment of acute and chronic inflammatory disease activity in the mucosa. However, scoring histological images is not only time-consuming but also requires pathologist training, which might not be available, especially in nonacademic institutions or smaller hospitals. Moreover, there is the unsolved problem of inter- and intra-observer variability. Therefore, Professor Laurent Peyrin-Biroulet of Nancy University Hospital in France and his team assessed whether an artificial intelligence (AI) tool using image processing and machine learning algorithms that assigns a Nancy index value to histopathology slides might be helpful in assessing histological disease activity. Eight global sites submitted 600 UC histological images, which were added to the 200 images used in a smaller preliminary study. Almost all (90%) probes were used for training the algorithm and 10% for testing.
Cell and tissue regions of each training image were manually assessed by three histopathologists and assigned a Nancy index. These results were used to further train the AI, allowing the AI tool to fully characterize histological images, identify tissue types, cell types, cell numbers and locations, and measure the Nancy Index for each image.
The average intra-class correlation was 92.1% among the histopathologists and 91.1% between the histopathologists and the AI tool in all stages of disease progression. An even higher consensus was achieved at the extremes of the Nancy Index.
This study shows that the robustness of the AI tool was substantially improved by adding a larger number of tissue samples by maintaining accuracy.
Copyright ©2023 Medicom Medical Publishers