Prototype-based AI triage for 3D pathology
Year: 2026
Type: preprint
Abstract: Non-destructive 3D pathology enables high-resolution slide-free imaging of intact clinical specimens, providing comprehensive visualization of tissue structures beyond what conventional slide-based 2D histopathology can provide. However, the scale and complexity of volumetric datasets make exhaustive manual review impractical, motivating AI-assisted triage methods to select a small number of high-risk 2D slices for pathologist review. While prior triage models have shown promise, interpretability is poor and performance can be suboptimal, especially in the nascent field of 3D pathology in which labeled data is limited. We present SCOPE, a S egmentation-guided C r O ss-slice P rototyp E learning framework for comprehensive risk assessment of 2D levels within 3D pathology datasets. SCOPE combines (i) clustering-based pretraining on large-scale unlabeled volumetric data to initialize morphology-aware prototypes, (ii) segmentation-derived structural priors from publicly available models to guide proto-type learning, and (iii) cross-slice (2.5D) prototype aggregation across neighboring slices to generate slice-level risk predictions. In prostate and esophageal data cohorts, SCOPE consistently outperforms attention-based and prototype-based multiple instance learning baselines for both binary and multiclass prediction tasks, enabling depth-resolved risk profiling for 3D triage based on morphological prototypes that are interpretable to pathologists.
Institutions: University of Washington, Stanford Medicine, Stanford University, The University of Texas MD Anderson Cancer Center, Harvard University, Mass General Brigham, Broad Institute, Harvard–MIT Division of Health Sciences and Technology, Dana-Farber Cancer Institute, Massachusetts Institute of Technology, University of Pennsylvania, Georgia Institute of Technology, Emory University, The Wallace H. Coulter Department of Biomedical Engineering
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Awards: RadxTools for assessing tumor treatment response on imaging, (PQ6) Radiogenomics of colorectal polyps to assess benign proliferative vs. premalignant states., Lung cancer in East Africa and the relationship to HIV-1 infection: epidemiology, molecular characterization and imaging, Genetic Determinants of Barrett's Esophagus and Esophageal Adenocarcinoma, Instrumentation platform for 3D pathology with open-top light-sheet microscopy, G3112464035, Prostate cancer risk stratification via computational 3D pathology, CHIRP Computerized Histologic Risk Predictor (CHiRP) for Early Stage Lung Cancers, Graduate Research Fellowship Program (GRFP), HistoTools: A suite of digital pathology tools for quality control, annotation and dataset identification, MR Fingerprinting and Computerized Decision Support for Prostate Cancer, Computerized histologic image predictor of cancer outcome, Quantitative Histomorphometric Risk Classifier (QuHbIC) in HPV + Oropharyngeal Carcinoma, Biomarkers for Reducing Mortality of Cancers of the Colon and Esophagus, Esophageal Cancer from Cells to Population: A Multiscale Approach, G7967259766, Multi-modal machine learning to guide adjuvant therapy in surgically resectable colorectal cancer, Computational 3D pathology for Barrett's esophagus risk stratification, The Pacific Northwest Prostate Cancer SPORE, Novel Radiomics for Predicting Response to Immunotherapy for Lung Cancer, Biomarkers for optimizing risk prediction and early detection of cancers of the colon and esophagus, Oral Cavity Quantitative Histomorphometric Risk Classifier (OHbIC) in Oral Cavity Squamous Cell Carcinoma (OC-SCC), Prognostic and Predictive Digital Tissue Image Assay for Prostate Cancer, An AI-enabled Digital Pathology Platform for Multi-Cancer Diagnosis, Prognosis and Prediction of Therapeutic Benefit
OpenAlex ID: W7168385491