Deep learning radiomic nomogram using visceral fat on non-contrast CT predicts GIST risk grade

A deep-learning radiomic nomogram (DLRN) using visceral fat features on noncontrast CT achieved AUC 0.936 (internal) and 0.862 (external) for preoperative GIST risk stratification, outperforming clinical models. Retrospective, small cohort — further validation needed.

• Retrospective study of 211 patients from two institutions (158 derivation, 53 external test).
• DLRN showed higher net clinical benefit than all comparison models in decision curve analysis.
• External test AUC 0.862 but wide 95% CI (0.62–1.00) due to small sample; needs larger multicenter validation.

RadPigeon summaries are original and for information only. They are not clinical advice.