Interpretable Multimodal Fusion Model for Bridged Histology and Genomics Survival Prediction in Pan-Cancer.

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Informatics

Background:

• Accurate cancer prognosis is vital for effective clinical diagnosis and treatment planning.
• Multimodal artificial intelligence (AI) fusion models enhance tumor heterogeneity analysis for improved prognosis prediction.
• Clinical utility of AI models is limited by incomplete multimodal patient data.

Purpose of the Study:

• To develop an interpretable bridged multimodal fusion model combining histopathology, genomics, and transcriptomics.
• To improve the precision of cancer patient prognosis, especially when molecular data is missing.
• To create a clinically applicable AI model for cancer prognosis.

Main Methods:

• Developed an interpretable bridged multimodal fusion model.
• Integrated histopathology, genomics, and transcriptomics data.
• Validated the model across 12 cancer types with complete and missing data modalities.

Main Results:

• The model achieved optimal predictive performance across 12 cancer types.
• Demonstrated high accuracy in both complete and missing data scenarios.
• Showcased the model's utility for patients with incomplete molecular features.

Conclusions:

• The developed AI model offers precise cancer prognosis predictions, even with missing data.
• Highlights the potential for clinically applicable multimodal fusion models in oncology.
• Aims to reduce healthcare burden and enhance clinical decision-making for cancer patients.