PathOrchestra: a comprehensive foundation model for computational pathology with over 100 diverse clinical-grade...

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Digital pathology

Background:

• High-resolution pathological images pose challenges for computational pathology.
• AI foundation models require extensive data, storage, and computational resources.
• Clinical validation is crucial for AI models in pathology.

Purpose of the Study:

• To present PathOrchestra, a versatile pathology foundation model.
• To evaluate its performance on a wide range of computational pathology tasks.
• To assess its clinical readiness and potential for digital medicine.

Main Methods:

• Trained PathOrchestra on 287,424 slides across 21 tissue types from three centers.
• Evaluated the model on 112 tasks from 61 private and 51 public datasets.
• Assessed performance on tasks including preprocessing, classification, prediction, and report generation.

Main Results:

• Achieved >0.950 accuracy in 47 tasks, including pan-cancer classification and lymphoma subtyping.
• Demonstrated high performance on whole slide and region-of-interest images.
• First model to generate structured reports for colorectal cancer and lymphoma.

Conclusions:

• PathOrchestra shows clinical readiness for large-scale, self-supervised pathology foundation models.
• The model offers high accuracy and potential for integration into digital medicine.
• Highlights the advancement of AI in pathology for clinical applications.