Dual-stream cross-modal fusion alignment network for survival analysis.

Area of Science:

• Oncology
• Computational Biology
• Medical Imaging

Background:

• Survival prediction is crucial for personalized cancer treatment.
• Current methods struggle with integrating histopathological images and genomic data effectively.
• Limitations include over-reliance on global features and suboptimal cross-modal fusion.

Purpose of the Study:

• To develop a novel cross-modal fusion alignment framework, DSCASurv, for enhanced survival prediction.
• To address limitations in feature representation and data fusion in multimodal cancer research.
• To improve patient stratification and treatment optimization in precision oncology.

Main Methods:

• Leveraging convolutional layers for local feature extraction and scanning state space models for long-range dependencies.
• Utilizing dual parallel mixer architectures for generating cross-modal representations.
• Employing a cross-modal attention module for inter-modal information exchange and complementary information transfer.

Main Results:

• DSCASurv effectively extracts intra-modal and cross-modal representations.
• The framework enhances and recalibrates complementary information for improved survival prediction.
• Experiments on five benchmark cancer datasets show superior performance over existing methods.

Conclusions:

• DSCASurv offers a significant advancement in multimodal survival prediction for cancer.
• The proposed fusion alignment framework improves accuracy by integrating local and global features effectively.
• This approach holds promise for optimizing treatment strategies in precision oncology.