xsCT-based radiomics for adrenal incidentaloma characterization.

PURPOSE: The aim of this retrospective monocentric study is to examine the role of radiomics in distinguishing benign and malignant adrenal incidentalomas (AIs) analyzing computed tomography (CT) images.

MATERIAL AND

METHODS

We retrospectively selected patients with clinical or histologic diagnosis of adrenal pathology studied in the period 2011-2025 with pre-operative or pre-diagnostic CT in our center. Two radiology residents and a radiologist with 10 years of experience delineated the volume of interest of the AI using 3D-slicer software. All 135 patients included in the analysis underwent segmentation of the non-contrast acquisition; among these, 115 were also segmented on the portal phase acquisition. Statistical analysis was performed by dividing patients into two groups: group 0, benign adrenal pathology (adenoma, myelolipoma, ganglioneuroma, pseudocyst); group 1, malignant adrenal pathology (pheochromocytoma, adrenal cortical carcinoma, metastasis, lymphoma).

RESULTS

135 patients were included. On non-contrast acquisition, the logistic regression model LASSO selected three first-order features and two second-order features able to distinguish patients belonging to Group 0 or Group 1. The features selected were used to construct a radiomic model that showed excellent AUC/discrimination (AUC: 0.88-95% CI 0.83-0.94). For portal phase, the logistic regression model LASSO selected three first-order features and two second-order features able to distinguish Group 0 and Group 1. The features selected were used to construct a radiomic model that showed excellent AUC/discrimination (AUC: 0.84- 95% CI 0.77-0.91).

CONCLUSION

The radiomic models developed from both non-contrast and portal CT phases were effective in differentiating benign from malignant AIs.