Widespread H&E preparation imperfections challenge AI-driven digital pathology transition
The study reports longitudinal external quality assessment data and a 2024 national survey from France examining H&E slide preparation quality. Investigators found that up to 25.5% of slides exhibited technical imperfections such as excessive thickness or tissue stretching, while 8.3% to 23.8% showed suboptimal staining with inadequate nuclear-cytoplasmic contrast. The authors emphasize that pre-analytical standardisation must be prioritised before scaling whole-slide imaging and AI algorithms, as robust computational models require consistent, high-quality inputs rather than relying on post-hoc correction. Optimising slide preparation at the source is presented as essential to reduce computational costs, improve diagnostic reliability, and ensure ethical transparency in digital pathology workflows.
The original study
Current quality challenges in H&E preparation: The critical foundation for increased use of digital pathology and AI emergence.
- Authors
- Tondon C, Erb G, Egele C, Michot JP, Fetique D, Chenard MP, et al.
- Journal
- Virchows Archiv : an international journal of pathology
- Type
- Journal Article
- PMID
- 42730840
Original abstract
Hematoxylin and Eosin (H&E) glass slide preparation is fundamental to histopathology. Pathologists' expertise routinely mitigates potential issues due to quality variability into or across laboratories, ensuring diagnostic accuracy. However, as artificial intelligence (AI) induced by whole slide imaging (WSI) emerges as a future advancement in pathology, a shift toward greater technical standardisation of the preparations or adaptive algorithms deserves consideration. This study examined longitudinal data (2019-2024) from the French national external quality assessment (EQA) programme integrated with a national survey conducted in 2024. We report that up to 25,5% of H&E slides, varying by tissue type and assessment year, exhibited technical preparation imperfection from multiple stages, including microtomic sectioning, excessive thickness, and tissue stretching. Separately, 8,3% to 23,8% of slides showed suboptimal staining, characterised mainly by insufficient nucleo-cytoplasmic contrast or intensity fluctuations. Advocating sustained quality improvement to prepare for future diagnostic pathology, we emphasise that optimising pre-analytical H&E preparation quality before AI deployment is critical: robust AI performance requires consistent, high-quality inputs rather than reliance on hypothetical post-hoc algorithmic compensation. Prioritising quality at the source reduces computational burden, environmental impact, and deployment costs while enhancing ethical transparency. Our findings provide technical areas for improvement to enhance H&E testing while anticipating AI integration at scale.