Related Experiment Video
Updated: Aug 5, 2026
06:18
Pioneering Patient-Specific Approaches for Precision Surgery Using Imaging and Virtual Reality
Published on: April 5, 2024
Prototype-based AI triage for 3D pathology
Renao Yan1,2, Gan Gao1, Andrew H Song3,4,5
1Department of Mechanical Engineering, University of Washington, Seattle, WA, USA.
Biorxiv : the Preprint Server for Biology
|July 29, 2026
Summary
SCOPE, a novel AI framework, enhances 3D pathology by guiding risk assessment using segmentation and cross-slice learning. This interpretable approach improves 2D slice selection for pathologists, overcoming limitations of current methods.
Area of Science:
- Computational pathology
- Artificial intelligence in medicine
- Medical imaging analysis
Background:
- Non-destructive 3D pathology offers advanced visualization but poses challenges for manual review due to large datasets.
- Current AI-assisted 3D pathology triage methods lack interpretability and optimal performance, especially with limited labeled data.
Purpose of the Study:
- To introduce SCOPE, a Segmentation-guided Cross-slice Prototype learning framework for accurate risk assessment of 2D levels within 3D pathology datasets.
- To improve interpretability and performance in AI-assisted 3D pathology triage.
Main Methods:
- SCOPE utilizes clustering-based pretraining on unlabeled volumetric data to create morphology-aware prototypes.
- It incorporates segmentation-derived structural priors to guide prototype learning.
- Cross-slice (2.5D) prototype aggregation is employed for slice-level risk prediction.
Main Results:
- SCOPE demonstrates superior performance compared to attention-based and prototype-based multiple instance learning baselines in both binary and multiclass prediction tasks.
- The framework achieved consistent outperformance across prostate and esophageal cancer datasets.
- SCOPE enables depth-resolved risk profiling using interpretable morphological prototypes.
Conclusions:
- SCOPE provides a robust and interpretable solution for AI-assisted triage in 3D pathology.
- The framework effectively addresses the challenges of large volumetric datasets and limited labeled data.
- SCOPE facilitates more efficient and accurate pathologist review of 3D pathology specimens.
