AI Tools Show Promise in Improving HER2 Breast Cancer Classification and Treatment Eligibility
Key Insights
A multinational study demonstrates that AI assistance significantly improves pathologists' accuracy in HER2 (search) breast cancer (search) scoring, with accuracy rates increasing from 89.1% to 96.1% when AI tools were utilized.
The Digital PATH Project, involving 31 partners including pharmaceutical companies and academic centers, validated AI diagnostic technologies using 1,100 breast cancer (search) tissue samples to assess consistency in HER2 (search) expression identification.
AI tools particularly enhanced detection of HER2 (search)-low and HER2-ultralow expression levels, reducing misclassification by 24.4% and potentially expanding treatment eligibility for antibody-drug conjugates (search) to patients previously classified as HER2-negative.
A groundbreaking multinational study has demonstrated that artificial intelligence tools can significantly enhance the accuracy of HER2 (search) breast cancer (search) classification, potentially expanding treatment options for thousands of patients previously ineligible for targeted therapies.
Researchers found that pathologists using AI assistance achieved a 96.1% accuracy rate in HER2 (search) scoring compared to 89.1% without AI support. The improvement was particularly notable for patients with low and ultralow levels of HER2 expression, according to findings presented ahead of the 2025 ASCO Annual Meeting.
"Roughly 65% of breast tumors once called HER2 (search)-negative actually demonstrate some level of HER2 expression and belong to subgroups now classified as HER2-low or HER2-ultralow breast cancers," explained lead study author Marina De Brot, MD, PhD, Associate Pathologist at A.C. Camargo Cancer Center in São Paolo, Brazil. "Some of these tumors could be treated with HER2-targeted drugs (search), but only if we detect their HER2 expression levels."
Digital PATH Project Validates AI Diagnostic Tools
Complementing these findings, the Digital PATH Project—one of the most comprehensive evaluations of AI diagnostic technologies in oncology to date—has validated the consistency of AI tools in identifying HER2 (search) expression across approximately 1,100 breast cancer (search) tissue samples.
The initiative, involving 31 contributing partners including pharmaceutical companies, academic centers, and government agencies, assessed variability between different digital pathology tools. Jeff Allen, president and CEO of Friends of Cancer Research (search), who will present the study's results at the upcoming Next Generation Dx Summit, emphasized that the goal wasn't to rank technologies but to evaluate their consistency and accuracy.
"We're very interested in exploring what the policy implications may be and how the use of independent reference sets could support the validation of these types of technologies for evaluating other biomarkers in the future," Allen stated.
Clinical Significance for Breast Cancer Treatment
The research comes at a pivotal time in breast cancer (search) treatment. The recent clinical recognition of "HER2 (search)-low" breast cancer has expanded treatment possibilities, with three antibody-drug conjugates (search) now approved for patients previously classified as HER2-negative.
In the multinational study, researchers implemented an AI-integrated digital training platform called ComPath Academy to teach 105 pathologists from 10 countries about HER2 (search) scoring. The pathologists assessed 20 digital breast cancer (search) cases across three exams, with AI assistance available only during the final exam.
Results showed that concordance among pathologists improved dramatically from 0.506 without AI to 0.798 with AI assistance. Most importantly, the rate of misclassification of HER2 (search)-low or HER2-ultralow cases as HER2-null was reduced by 24.4% with AI assistance—a critical improvement that could prevent patients from being incorrectly deemed ineligible for HER2-targeted therapies.
Challenges in Low Expression Detection
The Digital PATH Project found that while AI tools consistently matched pathologists' assessments for high HER2 (search) expression, they showed greater variability when evaluating low expression levels. Allen noted this variability wasn't unexpected, as these tools were trained before widespread recognition of the need to score HER2 at low levels because it was not yet an "actionable classification."
This challenge highlights the importance of continued refinement of AI diagnostic tools, particularly as treatment paradigms evolve to target more subtle molecular alterations.
Future Implications for Precision Medicine
Julian Hong, MD, MS, Associate Professor and Medical Director of Radiation Oncology Informatics at the University of California, San Francisco, and an ASCO expert in AI, emphasized the collaborative potential of AI in oncology: "These findings shed light on the promising role for AI in oncology, not as a replacement for the physician, but as a powerful tool to help us work smarter and faster to deliver high-quality, more personalized care."
The research methodology—using a standardized reference set across multiple platforms—demonstrates a potential pathway for efficient clinical validation of AI diagnostic tools. The speed with which the samples were evaluated—"in a matter of days and weeks"—highlights how this approach could accelerate regulatory approval processes.
Following this successful initiative, Friends of Cancer Research (search) has already launched a similar project focused on AI-enabled radiographic imaging tools that measure tumor changes following treatment, suggesting a broader application of AI validation approaches across different diagnostic modalities in oncology.