One of the most important first steps after receiving a breast cancer diagnosis is understanding one’s specific type of cancer. Recently, exciting new tools are emerging that can make this process even better: artificial intelligence (AI).
In a study presented at the 2025 American Society of Oncology (ASCO) Annual Meeting by Dr. Marina De Brot, associate pathologist and postgraduate professor at A.C.Camargo Cancer Center in São Paulo, Brazil, AI-assisted training bettered pathologist concordance and reduced misclassification in HER2-low and HER2-ultralow breast cancer, highlighting the value of AI systems in biomarker interpretation training.
“From a patient perspective as well, what this does is this: with the increasing use and application of anti-HER2-directed therapies, especially in the low and ultralow populations, patients now have access to potentially life-changing medications with this reclassification,” said ASCO president, Dr. Robin Zon.
In the study, pathologists were given access to the ComPath Academy training platform, an AI training masterclass. Across 1940 readings from 105 pathologists in 10 countries, the rate of accuracy in pathologists not using AI was just under 90%, compared with just over 96% in those using AI assistance.
The accuracy in identifying HER2 clinical categories also improved from approximately 90% without AI to 95% with AI.
The study also looked at how much pathologists agreed with each other on HER2 scores, called “concordance.” With AI, oncologists’ agreement got much better, and in turn, the pathologists were likely to come to the same, correct conclusion.
Sometimes, it can be difficult for pathologists to tell if a cancer has very little or no HER2 (HER2-low or HER2-null). With AI, the ability to correctly identify these cases significantly increased, which is of importance because it means more patients who could benefit from HER2-targeting treatments could now be correctly identified and have access to those medicines. For example, the study showed that the accuracy in finding HER2-null cases jumped from about 54% to over 88% with pathologists’ use of AI.
De Brot noted how, while many HER2-low and -ultralow tumors are now targetable, they are often mislabeled as HER2-null, and patients are potentially missing out on access to effective therapies. She pointed to an observed 30% discordance rate among pathologists in scoring HER2-low and -ultralow tumors, underscoring the imperative for advanced training. Reducing misclassification of these cases “potentially enabl[es] more patients to access HER2-directing [antibody-drug conjugate (ADC)] therapies.”
In her presentation, De Brot outlined the next steps for this research, including rolling out breast HER2 masterclasses to additional countries and pathologists and building a global, unified, international database to map scoring gaps and guide AI training solutions. De Brot also identified a plan to conduct a multicenter implementation study where an AI tool would be embedded in routine diagnostics to measure downstream clinical effects, which would allow investigators to assess changes in treatment allocation and time to therapy in patients with HER2-low and HER2-ultralow breast cancers.
These findings show that AI systems can be valuable tools for doctors, including enhanced decision-making and improved access to treatment.
Outcomes4Me in partnership with CURE brings you access to exclusive, hand-selected coverage of ASCO abstracts, sessions and key data specific to Breast Cancer, curated with scientific excellence and a patient-first perspective.
Explore additional research takeaways from ASCO 2025 here. Stay informed on the latest news in breast cancer research and download the Outcomes4Me app. Enable notifications so we can send you important updates.
