Non-invasive prognostication of adrenocortical cancer using CT radiomics

Abstract Background Adrenocortical carcinoma (ACC) is an aggressive malignancy with limited non-invasive tools for early prognostication. Radiomics—the extraction of high-dimensional features from medical images—offers a promising way to bridge this gap. We explored whether radiomic features from contrast-enhanced CT scans correlate with tumour aggressiveness, including Ki-67 index, tumour size, TNM staging, and future metastasis, to support non-invasive risk stratification in ACC. Method We retrospectively analysed contrast-enhanced CT scans of 58 patients with histologically confirmed ACC. Fifty-three cases were from The Cancer Imaging Archive, and 18 from the Scottish Adrenal Cancer Registry. Thirteen TCIA scans were excluded due to a lack of portovenous contrast, yielding 58 eligible cases. Manual 3D segmentation of adrenal tumours was performed. A standardised PyRadiomics pipeline was used to extract 107 features per case. Pearson and Spearman correlations were used for continuous outcomes, Kruskal-Wallis for ordinal outcomes, and Mann–Whitney U tests for binary outcomes. Results Shape features correlated strongly with tumour biology: elongation predicted Ki-67 index (r = 0.50, AUC = 0.72), with NGTDM Strength, Maximum, and Kurtosis improving model performance (AUC = 0.76). MajorAxisLength showed near-perfect correlation with tumour size (r = 0.92, P < 10−16). Entropy and intensity-based features distinguished TNM stages and metastatic status (P < 0.05). Conclusion Radiomic features, particularly shape and texture, correlate with ACC aggressiveness and could help develop non-invasive tools for diagnosis, prognostication, and personalised follow-up. This supports our original aim of identifying imaging biomarkers that reflect tumour biology, demonstrating that radiomics may offer a valuable adjunct to traditional pathology in guiding early, non-invasive risk stratification in ACC.