From Classical Machine Learning to Emerging Foundation Models: Review on Multimodal Data Integration for Cancer...

Area of Science:

• Computational oncology
• Artificial intelligence in medicine
• Multimodal data integration

Background:

• Cancer research increasingly relies on integrating diverse data modalities like genomics, proteomics, imaging, and clinical data.
• Extracting actionable insights from these complex, heterogeneous datasets presents a significant challenge.
• Foundation models (FMs) offer promising avenues for biomarker discovery, improved diagnostics, and personalized cancer treatment.

Purpose of the Study:

• To provide a comprehensive review of multimodal data integration strategies in oncology.
• To examine emerging trends in machine learning (ML) and deep learning (DL), focusing on the transition to FMs.
• To identify state-of-the-art FMs, repositories, and tools for advancing data-driven cancer research.

Main Methods:

• Systematic review of literature on multimodal data integration in oncology.
• Analysis of machine learning (ML) and deep learning (DL) methodologies, including foundation models (FMs).
• Examination of trends, frameworks, validation protocols, and open-source resources.

Main Results:

• The study maps the transition from conventional ML to advanced FMs for multimodal data integration in oncology.
• Identifies key advancements, challenges, and state-of-the-art FMs for integrating multi-omics and imaging data.
• Highlights publicly available multimodal repositories and tools crucial for data integration.

Conclusions:

• Current integrative methods using FMs lay the groundwork for next-generation large-scale AI models in oncology.
• This review is the first to systematically chart the evolution from ML to FMs in multimodal cancer data integration.
• These advancements are foundational for the future of AI-driven cancer research and personalized medicine.