Quantitative method for intestinal eosinophils in inflammatory bowel disease using artificial intelligence-based software

Inflammatory bowel disease (IBD) is a condition of unknown etiology. Eosinophilia is sometimes prominent in pathological specimens of the small and large intestines of patients with IBD, but no consensus exists on the clinical significance of eosinophils in IBD. Eosinophil quantification methods include manual counting by pathologists and software analysis. In this study, we aimed to establish a quantitative technique for tissue eosinophils in small and large intestinal pathological specimens of IBD using artificial intelligence-based software and confirm its pathological validity. Small and large intestine specimens were collected from 40 patients. Hematoxylin and eosin-stained slides were prepared and scanned using WSI. In the eosinophil-rich field of view for the 39 cases, the eosinophil counts of the two pathologists and the software were performed. The software was trained in advance using the remnant case. Significant positive correlations were found between the eosinophil counts of the two pathologists' average and the software (Spearman's rank correlation coefficient = 0.919, P = 1.62e-16) and of the two pathologists (Spearman's rank correlation coefficient = 0.767, P = 0.0000000122). Moreover, the average eosinophil counts of the two pathologists were predicted using the eosinophil count software. These results will enable future investigation of the clinicopathological significance (differential diagnosis, treatment response, and prognosis) of the total tissue eosinophil count in IBD, as well as reproducible assessment of disease activity in terms of eosinophils.