Agnostic-Specific Modality Learning for Cancer Survival Prediction From Multiple Data.

Area of Science:

• Oncology
• Computational Biology
• Bioinformatics

Background:

• Cancer is a leading global cause of mortality, necessitating improved survival prediction.
• Accurate prediction aids clinicians in developing effective treatment strategies and enhancing patient quality of life.
• Integrating diverse data, including pathological images and genomics, is key to advancing cancer survival prediction.

Purpose of the Study:

• To propose a novel agnostic-specific modality learning (ASML) framework for accurate cancer survival prediction.
• To address challenges of modality gap and semantic redundancy in multimodal cancer data.
• To enhance the comprehensive integration of diverse cancer-related data for improved predictive performance.

Main Methods:

• Developed an agnostic-specific learning strategy to identify commonalities and unique features across different data modalities.
• Employed a cross-modal fusion network to integrate multimodal information by modeling correlations.
• Utilized a divide-and-conquer approach to diminish semantic redundancy within integrated data.

Main Results:

• The ASML framework demonstrated superior performance compared to existing methods in cancer survival prediction.
• Experiments were conducted on three The Cancer Genome Atlas (TCGA) datasets, validating the framework's effectiveness.
• The proposed method successfully bridged the modality gap and reduced semantic redundancy in multimodal cancer data.

Conclusions:

• The ASML framework offers a significant advancement in multimodal cancer survival prediction.
• This approach effectively integrates diverse data sources, overcoming key challenges in the field.
• ASML provides a promising computational tool for improving cancer patient outcomes and treatment planning.