ASCO 2025: Can AI assist in identifying HER2-low and HER2-ultralow breast cancers?
A multinational study reveals how AI can help improve the accuracy in identifying and classifying HER2-low and HER2-ultralow breast cancers.
Research being presented at the ASCO Annual Meeting (30 May–3 June, IL, USA), has revealed that AI can assist in classifying breast cancers with low levels of HER2 protein expression, reducing the risk of misclassification and allowing more patients to choose HER2-targeted treatments.
The need to detect not only HER2-positive breast cancers, but also HER2-low and HER2-ultralow breast cancers, has become increasingly important in recent years due to the development of HER2-directing antibody–drug conjugates, a promising treatment option for these subtypes.
Roughly 55% of breast cancers are defined as HER2-low, and a further 10% are defined as HER2-ultralow. However, accurately detecting HER2 protein expression in HER2-low and HER2-ultralow breast cancers can be challenging and time-consuming, with pathologists disagreeing in the classification in roughly one-third of cases.
Traditionally, techniques such as immunohistochemistry (IHC) testing and in situ hybridization (ISH) are used to detect HER2 protein expression and nucleic acid sequences. With both techniques, an accurate diagnosis relies primarily on the human eye detecting abnormalities. This can sometimes lead to HER2-ultralow breast cancers mistakenly being labeled as HER2-null, resulting in eligible patients missing out on HER2-targeted therapy with antibody–drug conjugates – a treatment option that may prolong their survival.
ASCO 2025 guide: breaking news and key themes
To help you maximize your experience at ASCO, we compiled breaking news headlines and key conference themes in one handy article.
The study presented at ASCO involved 105 pathologists from 10 countries across Asia and South America who assessed 20 digital breast cancer cases using the AI-supported ComPath Academy platform. Pathologists performed 1,940 readings that were conducted during three separate exams, with AI assistance only provided during the third exam.
The results of the study were compared to ground-truth IHC scores from a central reference center – a consensus among expert pathologists who independently review and score HER2 IHC-stained tissue samples. The results revealed that with AI assistance, scoring sensitivity rose from 76% to 90% and pathologists’ agreement with the central reference scores improved by 13%, from 76.3% to 89.6%.
Additionally, pathologists’ accuracy in correctly identifying cases as HER2-positive, HER2-low, HER2-ultralow or HER2-null improved by nearly 22%, from 66.7% to 88.5%.
AI assistance also reduced the number of misclassifications of HER2-ultralow cases as HER2-null by more than 25%, with only 4% of readings misclassified, as opposed to 29.5% of readings without AI assistance.
“Our study provides the first multinational evidence that AI can help close a critical diagnostic gap and open the door to new therapies like antibody-drug conjugates for a majority of patients who, until recently, had not been offered these options,” commented lead author Marina De Brot (Camargo Cancer Center, São Paolo, Brazil).
Next, the researchers aim to conduct multicenter implementation studies, which will embed the AI-supported ComPath Academy platform into routine diagnostics, allowing downstream clinical effects to be measured.