Future perspectives on equivocal renal masses: AI and imaging
The EAU Section of Urological Imaging session “Renal cancer: Imaging challenges in the management of renal masses” covered future perspectives on the characterisation of equivocal renal masses. Section Chair Prof. Francesco Sanguedolce (ES) and moderators Prof. Axel Bex (GB), Prof. Peter Mulders (NL), and Prof. Maxine Tran (GB) spearheaded the Plenary Session, which concluded with the inaugural EAU Imaging Vision Award 2025.
Radiological imaging
During the lecture “What’s new in radiological imaging?”, presenter Prof. Alexandre Ingels (FR) underscored the importance of multimodal approaches for equivocal masses. The new technologies mentioned included:
- Photon-counting CT scan: The latest and most promising technology, capturing spectral information (i.e., more accurate tissue attenuation properties).
- Dual-energy CT (DECT): Eliminates pseudoenhancement (i.e., corrects beam-hardening artifacts), provides colour-coded iodine overlay images, and reduces the need for additional imaging tests (similar to MRI).
- Contrast-enhanced ultrasound imaging: Uses a contrast agent consisting of gas microbubbles stabilised by a phospholipid shell.
Prof. Ingels also emphasised the importance of standardising classifications such as the Bosniak Classification of Cystic Renal Masses (version 2019) which reduces interobserver variability, formally incorporates MRI features, and categorises a greater proportion of cases as lower-risk. The new classification, Clear-Cell Likelihood Score (ccLS), is a Likert scale (ranging from 1 point = very unlikely to 5 points = very likely) designed to improve the prediction of histopathological features of small renal masses in multiparametric MRI studies. He stated, “These new technologies will offer new opportunities for artificial intelligence (AI) applications.”
Molecular imaging
In his presentation “What’s new in molecular imaging?”, Dr. Giuseppe Basile (IT) concluded that while FDG-PET (Fluorodeoxyglucose Positron Emission Tomography) and PSMA-PET have poor diagnostic accuracy for characterising primary tumours, they still perform better in detecting metastatic disease (i.e., PSMA-PET has higher sensitivity).
In addition, although Sestamibi SPECT/CT (Single-Photon Emission Computed Tomography combined with Computed Tomography) is a low-cost and highly available tool, it has poor diagnostic accuracy for oncocytic tumours.
Dr. Basile also stated that 89ZrZr-girentuximab PET/CT demonstrated optimal diagnostic accuracy in differentiating intermediate renal masses (ccRCC), offering a safe approach with a better tumour-to-background ratio and high specificity.
Potential AI applications
During his presentation, “What would be the role of AI?”, Dr. Enrico Checcucci (IT) discussed the potential application of AI for the imaging of renal masses in terms of kidney tumour segmentation, classification, staging and grading, as well as nephrometry score assessment.
Dr. Checcucci stated that the use of fully automatic segmentation in kidney tumour segmentation is the first step in Machine Learning-based cancer classification, staging, and treatment planning (e.g., UNet and nnUNet algorithms).
Regarding classification, AI has the potential to differentiate between benign and malignant lesions. Deep Learning (DL) algorithms can differentiate oncocytoma from RCC with an accuracy of 0.75. With Convolutional Neural Networks (CNN), it is possible to differentiate benign from malignant masses with an accuracy of 78% and an Area Under the Curve (AUC) of 0.82. Using radiomics, demographic, histological, and genomic data can be combined to further improve the prediction of classification.
On staging, the integration of clinical variables and radiological data with DL algorithms allows for the prediction of the RCC stage. Furthermore, the inclusion of radiomic features allows for achieving an AUC of 0.83.
Concerning grading, multiple radiomic studies have focused on evaluating tumour aggressiveness through the prediction of nuclear grade.
Regarding nephrometry score assessment, various studies have explored AI’s capability to assess different nephrometry scores, such as PADUA, RENAL, and C-Index.
Potential future applications include the use of radiogenomics to overcome the limitations of tissue-based biomarkers and predict the presence of somatic gene mutations, as well as 3D virtual models to better understand tumour complexity, define the surgical strategy, and optimise staging, grading, and histopathological characterisation.
Award premiere
Mr. Jasper Twilt (NL), winner of the first EAU Imaging Vision Award 2025, which recognises the most innovative imaging study published in urology in 2024, presented “Evaluating Biparametric versus Multiparametric Magnetic Resonance Imaging for Diagnosing Clinically Significant Prostate Cancer: An International, Paired, Non-Inferiority, Confirmatory Observer Study.” The study concluded that bpMRI demonstrated statistically non-inferior diagnostic performance compared to mpMRI on a large scale. Prospective studies are warranted to investigate the clinical outcomes of bpMRI reading in biopsy decision-making.
Webcasts, videos, posters, and full-text abstracts are accessible via the EAU25 Resource Centre.