Computed tomography enterography radiomics and machine learning for identification of Crohn’s disease

Abstract

Abstract Background: Crohn’s disease (CD) is a severe chronic and relapsing inflammatory bowel disease. Contrast-enhanced computed tomography (CT) enterography (CTE) has been used frequently to evaluate CD. However, imaging findings of CD are not always specific and may overlap with other bowel diseases. Recent researches have shown that radiomics-based machine learning algorithms have been used to assist in the diagnosis of medical images. In this study, we aim to develop a non-invasive method for detecting bowel lesions from Crohn’s disease using computed tomography (CT) enterography (CTE) radiomics and machine learning algorithms. Methods: Patients (n=139) with pathologically confirmed Crohn’s disease were retrospectively enrolled into the study. Radiomics features were extracted from both the arterial- and venous-phase CTE images for both the bowel lesions with Crohn’s disease and segments of normal bowel. A machine learning classification system was built combining 6 selected radiomics features and 8 classification algorithms. The models were trained with leave-one-out cross-validation and were evaluated for accuracy. Results: The classification model showed robust performance and high accuracy, with the area under the curve (AUC) reaching 0.981 and 0.978 for the arterial- and venous-phase CTE images, respectively. The model achieved an accuracy of 0.9375, and 0.9615 for the arterial-phase and venous-phase image, respectively. Conclusions: Our study identified a CTE radiomics machine learning method that could be used to differentiate Crohn’s disease bowel lesions from normal bowel. Future studies with a larger sample size and external cohorts should be performed to validate our results.