Impact of abdominal adipose tissue segmentation on radiomics-based prognostic evaluation using CT enterography in Crohn's disease.

BACKGROUND

To evaluate the impact of different abdominal adipose tissue segmentation strategies on radiomics-based models for predicting 3-year disease progression in patients with Crohn's disease (CD).

METHODS

In this retrospective dual-center study, patients with CD who underwent baseline CT enterography (CTE) were included. Visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) were segmented using three approaches: (1) whole-abdomen segmentation (T-VAT/SAT); (2) single-slice segmentation of adipose tissue surrounding the most severely affected bowel segment (LL-VAT/SAT); and (3) single-slice segmentation at individual vertebral levels from L3 to S2 (V-VAT/SAT). Radiomic features were extracted after preprocessing, and redundant features were removed. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO), followed by construction of logistic regression models. A combined model incorporating radiomics and clinical features was also constructed. Model performance was evaluated using receiver operating characteristic (ROC), area under the precision-recall curve (AUPRC), and decision curve analysis (DCA).

RESULTS

In the external test set, the T-VAT model achieved an AUC of 0.877 (95% CI: 0.718-0.937) and an AUPRC of 0.657 (95% CI: 0.368-1.000), performing better than other VAT- and SAT-based models. The combined model, which incorporated clinical features with the T-VAT model, showed an AUC of 0.873 (95% CI: 0.711-1.000) and the highest AUPRC of 0.666 (95% CI: 0.376-1.000). DCA further demonstrated that the combined model provided greater net clinical benefit compared with both the T-VAT and clinical models.

CONCLUSIONS

Radiomics based on whole-abdominal VAT segmentation achieved the best prognostic performance for predicting disease progression in CD. Appropriate selection of adipose tissue segmentation strategy is critical for developing robust radiomics models for outcome prediction.

CLINICAL TRIAL NUMBER: Not applicable.