Recently, the Department of Urology at Jiangsu Provincial People’s Hospital, led by Professor Qiang Lü, published a multicenter study in EClinicalMedicine (CAS Zone 1, Impact Factor = 10.0). By integrating multicenter MRI data, the research team developed a multimodal fusion–based deep learning model designed to achieve noninvasive preoperative survival prediction in bladder cancer patients.The model not only significantly improves prognostic accuracy, but also identifies patient subgroups likely to benefit from perioperative therapy, providing valuable evidence to support personalized treatment decision-making and demonstrating strong potential for clinical application.
Key Findings of the Study
1. Outstanding Model Performance
The Multimodal Fusion Deep Learning Model (MF-DLM) achieved strong predictive performance, with C-index values of 0.902, 0.864, and 0.841 in the training, validation, and external test cohorts, respectively.
2. Effective Risk Stratification
The model successfully stratified patients into low-risk and high-risk groups, with significant survival differences between the two. This stratification ability remained robust across multiple clinical subgroups.
3. Potential to Guide Treatment Decisions
The model demonstrated meaningful clinical utility in guiding perioperative therapy:
- Low-risk pT3/4 patients showed significant survival benefit from adjuvant therapy
- High-risk pT3/4 patients exhibited no significant survival difference, regardless of adjuvant treatment
- In the neoadjuvant therapy setting, low-risk patients demonstrated survival benefit even without achieving pathological complete response
4. Strong Model Interpretability
The model showed significant correlations with tumor pathological characteristics, including tumor grade, invasion patterns, and lymphatic architecture, providing a biological rationale for its predictive capability.
Corresponding Author
Qiang Lü
First Author
Lingkai Cai
