Related Experiment Videos
Deep-learning triage of three-dimensional pathology datasets for comprehensive and efficient pathologist assessments
Gan Gao1, Renao Yan1,2, Andrew H Song3,4,5
1Department of Mechanical Engineering, University of Washington, Seattle, WA, USA.
Nature Biomedical Engineering
|August 12, 2026
Summary
This study introduces TRICARE, a deep learning framework for 3D pathology, to efficiently identify high-risk tissue sections for pathologist review. This AI-powered approach aims to improve disease detection and optimize workloads in clinical practice.
Area of Science:
- Digital Pathology
- Artificial Intelligence in Medicine
- Biomedical Imaging
Background:
- Standard 2D histopathology undersamples heterogeneous tissue, limiting diagnostic accuracy.
- Manual review of large 3D pathology datasets is time-prohibitive for clinical practice.
Purpose of the Study:
- To develop and validate TRICARE, a deep learning framework for triaging 3D pathology datasets.
- To enable efficient pathologist evaluation of 3D tissue volumes by identifying high-risk 2D cross-sections.
Main Methods:
- TRICARE, a deep learning triage framework, was developed to analyze 3D pathology datasets.
- The framework leverages contextual information from neighboring depth levels to assign risk scores to 2D cross-sections.
- Performance was evaluated using prostate cancer biopsies and Barrett's esophagus endoscopic biopsies.
Main Results:
- TRICARE outperforms models using isolated 2D levels by incorporating depth context.
- AI-triaged 3D pathology demonstrated improved detection of high-risk diseases compared to 2D histopathology.
- The framework shows potential for optimizing pathologist workloads.
Conclusions:
- TRICARE facilitates time-efficient pathologist evaluation of large 3D pathology datasets.
- AI-driven triage in 3D pathology offers a pathway for accelerated adoption in clinical settings.
- This approach enhances disease detection while maintaining pathologist involvement for final diagnoses.