Survival Prediction for Bladder Cancer Using Multimodal Data With Quantum Neural Networks and Transformer Architectures

Background: To address the challenges of cross-modal information fusion in high-dimensional multimodal medical data for cancer prognosis, this study presents a hybrid diagnostic accuracy model for cancer survival prediction, integrating quantum computing with classical deep learning in a retrospective analysis of bladder cancer patients. Methods: We propose QTMPN (Quantum-Transformer Multimodal Prognostic Network), a novel framework integrating quantum neural networks (QNNs), Transformers, and graph neural networks (GNNs). For high-dimensional whole-slide pathological images (WSIs), a quantum feature extractor (QFE) is designed using parallel quantum encoding and a hybrid quantum network to capture long-range dependencies. Multimodal data—including clinical and image features—are fused via a Transformer-GNN Collaborative Fusion (TCF) module employing attention-guided dynamic graphs. Results: Evaluated on the TCGA-BLCA dataset, QTMPN attained a survival prediction accuracy of 76.1% , outperforming baseline models such as PARADIGM and CMTA (up to 70.0%). This improvement suggests the model’s enhanced capability to capture cross-modal prognostic features. Further ablation experiment validated the effectiveness of the hybrid QNNs feature extract part (QFE) in QTMPN. Conclusions: QTMPN presents a promising quantum-classical framework for survival risk prediction in bladder cancer, effectively modeling complex multimodal interactions. The approach contributes to improving prognostic accuracy in oncology and supporting precision medicine.