Fusion model combining habitat-radiomics and 2.5D deep learning accurately differentiates lipid-poor adrenal adenomas from metastases

A fusion model (2.5D deep learning + habitat-radiomics + clinical features) differentiated lipid-poor adrenal adenomas from metastases with an external test AUC of 0.886 (multicenter, 390 patients, retrospective).

• Multicenter retrospective study using 390 patients divided into training, internal validation, and external test sets; automatic segmentation performed with Medical SAM.
• Standalone 2.5D deep learning model achieved AUCs up to 0.980, while habitat-radiomics alone reached up to 0.967, but the fusion model performed best overall.
• Retrospective design with external validation limited to one center; no prospective clinical validation is reported.

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