Computational pathology in precision oncology: Evolution from task-specific models to foundation models

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Biomedical informatics

Background:

• Artificial intelligence (AI) is increasingly integrated into clinical pathology workflows, improving accuracy and efficiency.
• Traditional AI models require extensive labeled datasets for specific tasks, limiting scalability for rare diseases and open-set identification.
• The need for adaptable AI solutions necessitates a shift from task-specific models to more versatile foundation models (FMs).

Purpose of the Study:

• To review recent advancements in pathological foundation models (FMs).
• To categorize FMs based on their architecture and functionalities (image, image-text, image-gene).
• To explore the applications, challenges, and opportunities of FMs in precision oncology.

Main Methods:

• Literature review of recent research on pathological foundation models.
• Classification of FMs into three main categories: pathology image FMs, pathology image-text FMs, and pathology image-gene FMs.
• Analysis of the application scenarios and functionalities of each FM category, particularly in oncology.

Main Results:

• Pathological FMs have emerged as a powerful tool, categorized into image, image-text, and image-gene types.
• These FMs demonstrate diverse functionalities and application scenarios, enhancing diagnosis, treatment, prognosis, and biomarker discovery.
• Significant progress has been made in applying FMs to oncology, paving the way for precision medicine.

Conclusions:

• Foundation models represent a paradigm shift in computational pathology, addressing limitations of task-specific approaches.
• Pathological FMs offer scalable and adaptable solutions for a wide range of clinical tasks, including rare disease analysis.
• FMs hold immense potential for advancing precision oncology by enabling more sophisticated biomarker discovery and personalized treatment strategies.