Deep Learning–Driven Analysis and Quantification of Histopathologic Features in a Dextran Sulfate Sodium–Induced Colitis Mouse Model

Dextran-based colitis models have been used extensively to study the pathophysiology of inflammatory bowel disease, including ulcerative colitis and Crohn disease. Histologic studies of animal colon sections require an unbiased scoring to yield meaningful outcomes. These examinations tend to be tedious, time-consuming, and influenced by inherent human bias, which can be reduced with digital pathology tools. In this study, a deep learning-based classifier was used to identify and quantify key histologic features in colon sections from a dextran sodium sulfate-induced colitis mouse model. The artificial intelligence (AI)-based classifiers were trained on HALO, an interactive human-in-the-loop image analysis platform, to sequentially identify tissues, mucosa, submucosa, and lymphoid tissues followed by quantification of different levels of tissue damage and cell infiltration within subregions, as well as goblet cell numbers and several cell types based on nuclear phenotyping. AI-based assessments could accurately identify key pathologic features and detect alleviation of the damage. This quantitative AI-based assessment of tissue damage exhibited good correlation and better sensitivity to detect histopathologic changes compared with the pathologists' assessment. This proof-of-concept study successfully established the efficiency of machine learning-based classifiers in characterizing disease pathology and their utility during the initial phases of drug discovery.