Study shows AI support improves HER2 biomarker scoring in breast cancer

Showcasing the capabilities of Ibex’s AI technology, the manuscript, titled Fully Automated Artificial Intelligence Solution for HER2 Immunohistochemistry Scoring in Breast Cancer: A Multireader Study, demonstrates that this ‘zero-click’ decision-support tool for pathologists results in highly accurate, reproducible, and efficient delineation of HER2 expression into four standard scores: 0, 1+, 2+ and 3+, based on the 2023 ASCO/CAP guidelines, including the highly subjective and challenging HER2-low cases.

HER2, one of the proteins responsible for division and proliferation of breast cancer cells, is expressed in many breast tumours and its accurate assessment is critical for identifying patients who are likely to benefit from HER2-directed therapies. Principal investigator, Professor Savitri Krishnamurthy MD, Professor of Pathology and Laboratory Medicine at The University of Texas MD Anderson Cancer Center, and corresponding study author commented: “Our analysis shows the algorithm performs very well. The study provides evidence that AI can be a useful ancillary tool to aid pathologists in agreeing with expert breast pathologists, as well as creating more concordance amongst themselves for HER2 protein expression at the lower end of the spectrum, including HER2 0 and 1+ cases. This AI tool could be extremely useful for practising pathologists, particularly in today’s era, where there is a lot of pressure to create HER2-low results in a more reproducible and objective way.”

The study included 120 breast cancer patients’ biopsies from four different medical centres in the US, Europe and the Middle East, and evaluated multiple pathologists’ performance when assisted by Ibex Breast HER2 versus the standard of care (manual HER2 quantification and scoring). Their performance was compared with the reference HER2 scores established by a panel of five international breast pathology experts. The findings illustrate the promise of Ibex’s AI as a valuable addition to the HER2 IHC diagnostic workflow, particularly for distinguishing between HER2-low cases close to the HER2 0/1+ cut-off, a critical distinction contributing to metastatic breast cancer treatment decisions.

The study showed AI support was able to offer:

• Increased consistency: pathologists’ average inter-observer agreement was significantly higher when assisted by AI (83.7%) than with the standard of care (75%), in all slides, and specifically for HER2 0 and 1+ slides (87.4% with AI vs. 69.8% without AI)
• Improved accuracy (relative to experts): pathologists supported by AI demonstrated improved accuracy, when scoring HER2 0 and 1+ slides (88.8% with AI vs. 81.9% without AI)
• Validation and robustness: the AI demonstrated exceptional robustness, with high performance across multiple labs, HER2 antibodies, scanners and patient demographics.

Professor Stuart J Schnitt MD, Chief of Breast Oncologic Pathology at Dana-Farber Brigham Cancer Center, Associate Director at Dana-Farber Cancer Institute-Brigham and Women's Hospital Breast Oncology Program, Professor of Pathology, Harvard Medical School, one of the study’s expert breast pathologists, said: “With the availability of new drugs to treat patients with low levels of HER2 expression, there is a need for a computational image analysis and quantification solution. The AI identifies areas of invasive cancer and categorises the different classes of HER2 protein expression very clearly. A case’s final score includes visualisation of the AI findings and staining pattern percentages, leading to confidence in the assessment.”

• Krishnamurthy S, Schnitt SJ, Vincent-Salomon A, et al. Fully Automated Artificial Intelligence Solution for Human Epidermal Growth Factor Receptor 2 Immunohistochemistry Scoring in Breast Cancer: A Multireader Study. JCO Precis Oncol. 2024;8:e2400353. doi:10.1200/PO.24.00353