Development and Validation of a Multimodal Fusion Artificial Intelligence Model for Predicting the Efficacy of Neoadjuvant Treatment of Bladder Cancer

Bladder cancer is one of the most common malignancies of the genitourinary system worldwide. For muscle - invasive bladder cancer amenable to radical resection, the standard treatment is neoadjuvant therapy combined with radical cystectomy, with neoadjuvant therapy playing a crucial role. Currently, numerous studies have shown that cisplatin - based neoadjuvant chemotherapy can downstage tumors, reduce the risk of mortality in bladder cancer patients, improve survival rates, and enhance prognosis. Other treatment approaches such as neoadjuvant immunotherapy, targeted therapy, and combination therapies are also under investigation. However, responses to neoadjuvant therapy vary among bladder cancer patients, with some not achieving the desired therapeutic goals. Therefore, accurately predicting participants' response to treatment can provide an important reference for personalized and precise treatment of bladder cancer.

In recent years, as advancements in computational power and data storage capacity, artificial intelligence (AI) has been widely applied in the field of digital diagnostics. AI technologies can extract and integrate a large number of features from multimodal data such as pathology, imaging, and clinical records, enabling precise disease diagnosis, prognosis assessment, and treatment prediction. In the field of tumor treatment prediction, multimodal AI technologies have achieved numerous breakthroughs, developing efficacy prediction models for tumors such as rectal and breast cancer based on imaging and pathological data, and validating the models' generalization capabilities through external validation.

Therefore, the investigators plan to construct and validate a "Bladder Cancer Neoadjuvant Treatment Efficacy Prediction Model" based on multimodal data (including MRI images, digital pathology images, and clinical records) of bladder cancer patients, and develop an AI-assisted prediction software for neoadjuvant treatment efficacy in bladder cancer. The study will adopt a combined retrospective and prospective data collection design to ensure sufficient sample size and model robustness. This study plans to enroll a total of 550 patients, including 500 retrospective cases and 50 prospective cases. Based on sample size calculation formulas and preliminary study results, the minimum sample size required for model development is estimated to be 220 cases. Considering imaging quality control, internal validation, and external validation needs, we plan to retrospectively enroll 500 cases from both the local center and external institutions. Due to the relative rarity of bladder cancer patients receiving neoadjuvant therapy, 50 prospective cases planned to verify the model's performance and clinical applicability.

Inclusion criteria

• Bladder occupying lesions, with histopathological confirmation of bladder cancer after resection.
• Planned neoadjuvant therapy and radical cystectomy.

Exclusion criteria

• Patients who have not undergone standard bladder imaging examinations or have missing imaging or pathological data.
• Patients who have received local treatments (such as interventional embolization) or systemic treatments (such as radiotherapy, chemotherapy, immunotherapy, or targeted therapy).
• Poor quality of imaging or pathological images.