Psoas muscle CT radiomics-based machine learning models to predict response to infliximab in patients with Crohn's disease.

Abstract

Background: Crohn's disease (CD) is a chronic inflammatory bowel disease, with infliximab (IFX) commonly used for treatment. However, no clinically applicable model currently exists to predict the response of patients with CD to IFX therapy. Given the strong association between sarcopenia and IFX treatment outcomes, this study developed computerized tomography radiomics-based machine learning (ML) models, utilizing psoas muscle volume as a proxy for skeletal muscle mass, to predict the response of patients with CD to IFX therapy.

Methods: In this retrospective study, patients with CD from two institutions were recruited between January 2010 and January 2023, following stringent inclusion and exclusion criteria. Regions of interest were delineated using 3D Slicer software, and radiomics features were extracted with the Pyradiomics package in Python. Z score standardization and independent sample t test were applied to identify optimal predictive features, which were then utilized in seven ML algorithms for training and validation. Model performance was assessed through receiver-operating characteristic curves, precision-recall curves, and calibration curve analyses, evaluating accuracy and clinical applicability. Binary logistic regression was employed to identify predictors of IFX treatment response.

Results: A total of 134 patients were included, divided into a training cohort (n = 84) and a validation cohort (n = 50). Twenty differential radiomics features were selected for integration into the ML models. All models demonstrated strong predictive performance in the validation cohort, with a mean area under the curve of 0.849. The eXtreme Gradient Boosting algorithm outperformed others, achieving an area under the curve of 0.910.

Conclusion: Psoas computerized tomography radiomics-based ML models effectively predict the response of patients with CD to IFX therapy, with the eXtreme Gradient Boosting model exhibiting superior performance.