News Brief
Recently, the urology team led by Prof. Junhua Zheng and Researcher Wei Zhai from the Department of Urology, Renji Hospital, Shanghai Jiao Tong University School ofkk Medicine, together with multiple collaborating centers, achieved a major breakthrough in precision prognosis of kidney cancer. Their multimodal prediction of recurrence score (MPRS) model was successfully published in npj Digital Medicine, a leading global digital medicine journal under the Nature portfolio. This model provides crucial support for recurrence-risk assessment and individualized treatment in patients with clear cell renal cell carcinoma (ccRCC).
Clinical Background and Unmet Need
Renal cell carcinoma is a common malignancy of the urinary system, with clear cell renal cell carcinoma accounting for about 70% of cases. Surgery remains the mainstay of treatment, yet approximately 20–30% of patients still experience postoperative recurrence and metastasis.
Current clinical tools such as the Leibovich score, UISS score, and the KEYNOTE-564 risk-stratification system have clear limitations. They rely mainly on clinical–pathological variables like tumor size and TNM stage, and cannot integrate multimodal prognostic information. Molecular-testing-based models are costly and hard to implement widely.
More importantly, these tools are prone to “risk misclassification”: some truly high-risk patients are underestimated and miss the window for adjuvant therapy, while some truly low-risk patients are overestimated and receive overtreatment, leading to both psychological and economic burden.
Development of the MPRS Multimodal AI Model
To address this clinical gap, the research team integrated clinical data, preoperative contrast-enhanced CT images, and postoperative whole-slide pathology images from a total of 1,648 patients drawn from six domestic centers plus the TCGA database. Based on these multimodal inputs, they innovatively constructed the MPRS multimodal AI model.
Compared with single-modality models and classical clinical tools (Leibovich score, UISS score, and KEYNOTE-564 trial risk classification), MPRS showed overwhelming superiority. In the internal validation cohort, the C-index reached 0.886, and 0.838 in the external validation cohort. The AUC values for 3-year and 5-year recurrence prediction were consistently above 0.829. Across different centers and imaging devices, the model maintained excellent performance, with calibration and robustness far exceeding existing tools.
Accurate Risk Re-Stratification and Model Interpretability
A particularly noteworthy advantage of MPRS is its ability to achieve precise risk re-stratification. Among patients misclassified as low risk by the KEYNOTE-564 system but who ultimately relapsed, MPRS successfully reclassified 83.3% into the high-risk group, thereby avoiding missed opportunities for adjuvant treatment. At the same time, 57.7% of non-relapsed patients who had been misclassified as intermediate/high risk were correctly reclassified as low risk, preventing unnecessary treatment and toxicity.
Using SHAP analysis and Grad-CAM visualization, the model can accurately capture key prognostic features such as irregular tumor borders and necrotic regions. The decision patterns identified by the AI closely match established clinical and pathological understanding, further supporting the reliability and clinical credibility of the model’s predictions.
Clinical Translation and Application Prospects
To promote clinical translation, the team deliberately chose data types that are already routine in standard care—CT scans and pathology slides—without requiring additional molecular testing or other high-cost examinations. The architecture is built on a lightweight ResNet design, substantially lowering the barrier to implementation in real-world hospitals.
The MPRS model can assist clinicians in designing personalized follow-up schedules and treatment strategies for ccRCC patients, and also offers a standardized risk-stratification tool for clinical research in kidney cancer. It is expected to reshape the diagnostic and therapeutic pathway of ccRCC and become an important pillar of precision oncology in this field.
Authors and Acknowledgements
This study lists Dr. Zang Xinyi (Renji Hospital, Department of Urology), Dr. Xia Yujia (Shanghai Jiao Tong University), Prof. Xiao Haibing (The First Affiliated Hospital of Anhui Medical University), and Prof. Luo Haolun (Kaohsiung Chang Gung Memorial Hospital, Taiwan) as co–first authors. Prof. Wang Keliang (Fourth Affiliated Hospital of Harbin Medical University), Prof. Liang Chaochao (The First Affiliated Hospital of Anhui Medical University), Prof. Yu Zhangsheng (Shanghai Jiao Tong University), Prof. Junhua Zheng (Renji Hospital, Department of Urology), and Researcher Wei Zhai(Renji Hospital) serve as co–corresponding authors.
The work was supported by the National Natural Science Foundation of China, the National Key Research and Development Program, and the Shanghai Municipal Education Commission’s special program on “Artificial Intelligence Empowering New Research Paradigms and Disciplinary Advancement,” among other grants, with strong institutional support from Shanghai Jiao Tong University and Renji Hospital.
