A NEW study revealed that artificial intelligence (AI) can significantly improve the diagnosis of renal masses using pre-operative computed tomography (CT) scans. The research, which analysed 13,261 CT scans from 4,557 patients, found that AI-driven deep learning models outperform experienced radiologists in predicting the malignancy and aggressiveness of kidney tumours.
Currently, treatment decisions for incidental renal masses are often made with pathologic uncertainty, leading to overtreatment of benign tumours or inadequate management of aggressive cancers. In this study, researchers developed two convolutional neural network (CNN) models to address these diagnostic challenges. The first model, designed to distinguish between benign and malignant renal masses, achieved an area under the curve (AUC) of 0.871 in the prospective test set, surpassing the diagnostic accuracy of seven expert radiologists. The second model, aimed at differentiating aggressive tumours from indolent ones, achieved an AUC of 0.783. Both AI models outperformed conventional radiomics models and the nephrometry score nomogram, which are commonly used in current clinical practice.
The findings suggest that deep learning can serve as a powerful tool for non-invasive renal cancer assessment, enabling more accurate preoperative risk stratification. By improving diagnostic precision, AI has the potential to reduce unnecessary surgeries for benign tumours while ensuring timely and appropriate treatment for aggressive cancers.
The study’s authors emphasise that integrating AI into clinical workflows could revolutionise kidney cancer management, leading to more personalised treatment strategies and improved patient outcomes. Future research will focus on refining these models further and integrating them into routine clinical decision-making.
Aleksandra Zurowska, EMJ
Reference
Xiong Y et al. Artificial intelligence links CT images to pathologic features and survival outcomes of renal masses. Nat Commun. 2025: DOI; 10.1038/s41467-025-56784-z.
