Nature Communications· 2026Q1
Comparative analysis of end-to-end learning and foundation models for clinical-grade prostate pathology
- 13citations
- Q1SCImago
- 2026year
Short summary
Foundation models (FMs) for prostate pathology perform comparably to, and are sometimes surpassed by, task-specific (TS) models when sufficient labeled data is available, despite FMs consuming up to 35x more energy.
AI-generated from the title and abstract; the full text is not read.
Key points
- Foundation models (FMs) and task-specific (TS) models achieve similar performance in prostate pathology with sufficient labeled data (quadratic Cohen's kappa ~0.91).
- TS models outperform FMs in some cases with full training data on external cohorts (0.87 for TS vs. 0.86 and 0.85 for FMs).
- Task-specific training significantly reduces misgrading, misdiagnosis of challenging morphologies, and technical variability.
- FMs can consume up to 35 times more energy than TS models.
AI-generated from the title and abstract; the full text is not read.
Abstract
Abstract Foundation models (FMs) using self-supervised pre-training are widely advocated as a universal solution for pathology, yet their advantages over task-specific (TS) models remain unclear. Here we show, using over 100,000 prostate biopsies from 6268 patients across 15 sites, that while FMs demonstrate utility in data-scarce scenarios, their performance converges with—and is in some cases surpassed by—TS models with sufficient labeled training data. Concordance with pathologists’ grading reaches quadratic Cohen’s kappas of 0.91 (95% CI 0.90-0.92; all models) on internal and 0.87 (0.86-0.88; TS), 0.86 (0.84-0.87; UNI FM) and 0.85 (0.84-0.86; Virchow2 FM) on external cohorts at full training. Task-specific training markedly reduces misgrading, misdiagnosis of challenging morphologies, and technical variability. Notably, FMs consume up to 35 times more energy. We advocate for considering both FMs and end-to-end learning for robust, resource-efficient, and clinically applicable AI pathology.
The authors' abstract, as published at the source. Nature Communications, 2026 · DOI ↗
Continue with a free account
Ask the paper: 3 free questions a day about this paper; save it, get its citation, new summaries every day for your field. Takeaways are Premium.
Continue free on the webSign in with Google or Apple; no card needed. You come back to this paper.
On your phone:
Field: Artificial Intelligence
Artificial IntelligenceComputer Science