P32 - ECE_3469 - Noninvasive differentiation of adrenal adenomas, pheochromocytomas, and malignant tumors using CT radiomics
TL;DR
Preliminary data suggest that CT texture features, particularly when combined with conventional density metrics, may enhance non-invasive differentiation of adrenal adenomas, pheochromocytomas, and malignant lesions and support the potential integration of radiomic analysis into routine adrenal imaging workflows and guide clinical decision-making.
Abstract
Non-contrast computed tomography (CT) remains the standard imaging modality for characterizing adrenal lesions and guiding subsequent clinical management. Conventional assessment relies on tumor size, density, and lipid content; however, these parameters may lack sufficient specificity, occasionally resulting in unnecessary surgical interventions or delayed malignancy diagnosis. Radiomics offers advanced quantitative image analysis capable of capturing tumor heterogeneity beyond visual assessment. This study aimed to evaluate whether CT-derived texture features can improve differentiation of adrenal lesions. In this single-center retrospective analysis, adrenal incidentalomas from patients treated at the Clinic of Endocrinology, Oncological Endocrinology, and Nuclear Medicine, University Hospital Krakow (2017-2025) were reviewed. Lesions were assessed in a native CT scan and confirmed histopathologically. Eighty-two focal adrenal lesions from 80 patients (58 women, 22 men) were included: 22 adenomas, 35 pheochromocytomas, 16 carcinomas, 3 oncocytomas, 2 borderline oncocytic tumors, and 3 oncocytic carcinomas. Manual segmentation of lesions on native-phase imaging was performed, followed by extraction of first- (GLRLM_GreyLevelNonUniformity) and second-order texture parameters (IntensityBasedRobustMeanAbsoluteDeviation (IBRAMD) and IntensityHistogramEntropyLog10 (IHEL 10) using Local_Image_Feature_Extraction (LIFEx) software. Diagnostic performance was assessed via receiver operating characteristic (ROC) analysis for individual features and multivariable logistic regression for combined models. Statistical analyses were conducted using STATISTICA 13.3. When integrated with tumor density, selected texture parameters GLRLM, IBRMAD, and Intensity IHEL10 demonstrated significant discriminatory potential. These parameters differentiated adenomas from pheochromocytomas (P < .01), adenomas from carcinomas (P < .01), and pheochromocytomas from adrenal carcinomas (P < .01 for GLRLM; P = .97 for IBRMAD; P = .85 for IHEL10). For adenomas from pheochromocytomas the texture-based feature HU + GLRLM achieved the highest discriminative ability, with an AUC of 0.95 (SE = 0.03; 95% CI: 0.89-1.00; P < .001). For adenomas from carcinomas the texture-based feature HU + GLRLM achieved the highest discriminative ability, with an AUC of 1 (SE = 0.00; 95% CI: 1.00-1.00; P < .001). For pheochromocytomas from carcinomas the texture-based feature HU + GLRLM achieved the highest discriminative ability, with an AUC of 0,89 (SE = 0.04; 95% CI: 0,8-0,97; P < .001). Diagnostic accuracy remained high across different tumor subtypes, with minimal differences between individual parameters. Preliminary data suggest that CT texture features, particularly when combined with conventional density metrics, may enhance non-invasive differentiation of adrenal adenomas, pheochoromocytomas, and malignant lesions. These findings support the potential integration of radiomic analysis into routine adrenal imaging workflows and guide clinical decision-making.