Multimodal Fusion Framework Based on Low-Rank Interaction for Tumor Prognostic Prediction.

Area of Science:

• Oncology
• Bioinformatics
• Medical Imaging

Background:

• Current cancer survival prediction models often rely on single-modal data, limiting accuracy.
• Existing multimodal fusion techniques can be overly complex and computationally intensive.

Purpose of the Study:

• To develop an innovative Multimodal Fusion Framework based on Low-rank Interaction (MF2LI) for enhanced cancer patient survival prediction.
• To overcome limitations of single-modal prediction and complex fusion methods.

Main Methods:

• The MF2LI model utilizes low-rank multimodal fusion (LMF) and optimal weight integration (OWI) for data fusion.
• A parallel decomposition strategy is employed to reduce model complexity and facilitate component-based fusion.
• Validation was performed on the GBMLGG and KIRC datasets from The Cancer Genome Atlas (TCGA).

Main Results:

• MF2LI achieved high C-index scores of 0.895 ± 0.007 (GBMLGG) and 0.728 ± 0.030 (KIRC), surpassing existing methods.
• Visualizations of risk ratios showed strong correlation with actual tumor grade classifications.
• The model demonstrated superior performance in integrating pathological images and genomic data.

Conclusions:

• The MF2LI framework significantly improves prognosis prediction accuracy for cancer patients.
• The model offers considerable clinical value by providing more reliable survival predictions.
• MF2LI represents an effective approach to multimodal data integration in cancer research.