This presentation explores the transformative role of artificial intelligence (AI) and digital pathology in biomarker discovery, focusing on the analysis of haematoxylin and eosin (H&E) slides. AI-driven approaches are increasingly used to identify molecular profiles and predict biomarkers directly from H&E images. According to experts this could potentially reduce scientists’ reliance on more complex and costly assays.
Foundation models trained on vast datasets of H&E slides now support mutation detection and decision-making in pathology. Some models have achieved high precision in identifying specific mutations and have gained regulatory approvals, such as FDA breakthrough status for certain applications. The workflow described moves from preclinical studies using H&E-based models to infer key biomarkers like mitotic rate and p21 expression, to successful translation in clinical samples, including non-small cell lung cancer tissue.
Technical and regulatory challenges remain, including sample normalisation, inter-pathologist variability, and increasing regulatory requirements for clinical trials. For example, FDA validation for biopsies is now more stringent, impacting the adoption of H&E-based AI models. The presentation highlights the importance of robust, generalisable foundation models and advocates for community collaboration to overcome these hurdles.
Francesa Trapani, Scientific Director & Molecular Pathology Laboratory Head, Boehringer Ingelheim demonstrated that AI models can reliably infer mitotic incidents and biomarker status from H&E slides, with strong correlation to traditional immunohistochemistry results. These models have been validated in both preclinical and clinical settings, showing promise for routine implementation. The ultimate goal is to develop foundation models that work across indications, helping to stratify patients and predict responses in clinical trials and everyday practice.
To wrap up, AI quantification of H&E slides is not just a future possibility but a present reality, offering significant potential to streamline biomarker discovery and improve patient outcomes. Trapani highlighted that continued collaboration and innovation are essential for realising the full benefits of these technologies.
