Validating AI screening for gBRCA mutations directly from pathology slides

Recently published research at ESMO 2025 (European Society for Medical Oncology Congress, Berlin, Germany, October 2025), highlights robust results achieved across multiple independent cohorts from Centre Léon Bérard (France), Washington University School of Medicine in St. Louis (USA), and King’s College London (UK).

Breast cancer remains the leading cause of cancer-related deaths among women worldwide. Identifying patients with germline BRCA1/2 mutations is essential for guiding targeted treatment decisions, but testing is not always consistently offered to eligible individuals, and implementation varies between countries.

The model concept was initiated as part of Owkin’s ongoing partnership with Gustave Roussy and Centre Léon Bérard through PortrAIt, a French consortium dedicated to advancing precision medicine with AI-enabled digital pathology. Fostered through collaboration with AstraZeneca, as announced in October 2024, the development of BRCAura aims to accelerate and expand identification of BRCA mutations in patients with breast cancer.

Trained on an enriched cohort of germline BRCA mutation patients from Gustave Roussy, BRCAura RUO demonstrated strong and reproducible performance:

• area under the curve (AUC) of 80% in predicting mutation status
• ability to rule out approximately 40% of patients unlikely to carry gBRCA mutations with a high sensitivity of 93%
• robustness across different staining types and scanners, ensuring consistent performance across real-world conditions.

“We developed BRCAura with the vision of expanding access to gBRCA testing and identifying more patients at risk of harboring BRCA mutations to determine the best personalised treatment approach,” explains Meriem Sefta, Owkin’s Chief Diagnostic Officer. “This first step is being leveraged in research settings, to enhance workflows, but also to gain an understanding of how morphological factors could be systematically leveraged to expand and accelerate gBRCA testing.”

“These recent results are promising as they highlight a robustness to pre-analytical and clinical variability that is a necessary requirement for any screening solution to optimise clinical routine diagnosis of gBRCA mutations,” says Sheeba Irshad, Professor of Cancer Immunology at King’s College London and Breast Cancer Medical Oncologist. “While further evaluation will be essential, approaches like this may eventually help support earlier and better-informed decisions about patient management.”

These results support BRCAura RUO as a powerful, standardised histopathology-based solution with increased reliability when compared to current clinical scoring methods relying upon age, and family history. These methods, typically validated, remain highly diverse in approach across different hospital settings. This paves the way for a clinically approved version of the device to introduce a more standardised approach to help streamline patient selection for genetic testing and next-generation sequencing (NGS). Additionally, by ruling out patients unlikely to carry the gene, BRCAura could help expand testing access to those patients not routinely considered for genetic testing.

“To advance our bold oncology ambitions, we are building a collaborative ecosystem across leading AI companies and academia to develop AI solutions that can transform care,” said Greg Rossi, Senior Vice President Oncology, Europe and Canada, AstraZeneca. “These validation results underscore how histopathology-based AI can help bridge the gap between pathology and genomics – paving the way for faster and broader genetic testing.”

The PortrAIt consortium is a French consortium financed by the government within the framework of France 2030 and by the European Union - Next Generation EU within the framework of the France Relance Plan. It consists of the following members: Owkin; Gustave Roussy; Centre Léon Bérard; Unicancer; Cypath; Tribun Health.