Artificial intelligence in the diagnosis of inflammatory bowel diseases: focus on clinical and laboratory markers

AIM : to work out artificial neural networks (ANNs) for the screening and differential diagnostics of inflammatory bowel diseases (IBD) based on the analysis of clinical and laboratory data. PATIENTS AND METHODS : the study retrospectively evaluated the clinical manifestations, medical history and laboratory data of patients with irritable bowel syndrome (IBS), ulcerative colitis (UC), and Crohn’s disease (CD) with colonic involvement during exacerbation. The most typical characteristics for each patient group (IBS-IBD, UC-CD) were estimated and were used to construct the models. In order to create ANNs that could make a decision on the presence of IBD and provide a differential diagnosis of UC and CD, we used simple neural networks multilayer perceptrons (MLP) and radial basis functions. RESULTS : the MLP 13:13-5-1:1 identified IBD in the test sample with a sensitivity of 89.3% and a specificity of 100%. Across the entire dataset, the model demonstrated a sensitivity of 92.7% and a specificity of 99.0%. The highest accuracy for the differential diagnosis of UC and Crohn’s disease CD was observed with the MLP 9:9-8-1:1, which identified 76.81% of CD cases and 86.67% of UC cases in the test sample. Across the entire dataset, the model detected 70.16% of CD cases and 86.40% of UC cases. CONCLUSION : the ANNs demonstrated high efficacy in identifying IBD and moderate performance — in the differential diagnosis of UC and CD. Following validation, the model may serve as a convenient tool for screening inflammatory bowel diseases in clinical practice.