A New AI Tool Assists Pathologists in Assessing HER2 0 vs 1+ in Breast Cancer
Pathologists supported by a new artificial intelligence tool outperformed Standard of Care/Gold standard in determining between HER2 0 from 1+ - low expressing cases.
3D rendering of breast cancer: © Giovanni Cancemi - stock.adobe.com
Pathologists supported by a new artificial intelligence (AI) tool outperformed Standard of Care/Gold standard in determining between HER2 0 from 1+ - low expressing cases, according to a team of investigators who published their findings in JCO Precision Oncology.1 As targeted therapies increasingly differentiate treatment effectiveness based on HER2 status, accurately determining HER2 status and keeping abreast of the latest guidelines are becoming essential for patient care, making this AI solution a vital tool for both pathologists and clinicians in this cancer space. 2
“There are other solutions out there and other published articles, but very few that support this new categorization [for HER2 status], which is a very important specification. In addition to development, we also have clinical validation with pathologists using this AI solution to interpret and score HER2 in the standard categories of 0, 1+, 2+, and 3+,” Manuela Vecsler, PhD, told Targeted Therapies in Oncology in an interview.
Vecsler is vice president of clinical and scientific affairs at Ibex Medical Analytics in Tel Aviv, Israel. The company, which supported the study, focuses on AI-driven solutions for cancer detection across various tissue types, catering to the global market.
AI vs Pathologists
In this 2-arm multireader study, 120 HER2 immunohistochemistry slides were assessed by 5 expert pathologists to establish the reference HER2 score (ground truth) and 4 “reader” pathologists, both with and without the help of an AI tool.1 The overall interobserver agreement among the expert pathologists was 72.4% for all slides. Complete agreement was reached on 44.2% of slides, and an additional 32.5% of slides had agreement from 4 of 5 pathologists. For slides with high-confidence ground truth (where at least 4 out of 5 pathologists agreed), the AI tool showed 92.1% agreement.
Interobserver agreement of the reader pathologists increased from 75.0% to 83.7%, and scoring accuracy improved from 85.3% to 88.0% with the assistance of AI vs manual review. For distinguishing HER2 0 from HER2 1+ cases, use of the AI tool coupled with pathologists’ expertise significantly improved both agreement (69.8% without AI vs 87.4% with AI) and accuracy (81.9% without AI vs 88.8% with AI).
“We were surprised to find that the interobserver agreement among the pathologists was only about 72%, which is quite low,” Vecsler said. “In the end, what the study showed was that by using AI we can improve this interobserver agreement and consequently improve consistency and standardization of scoring.”
The high-confidence ground truth was established by agreement of 4 of 5 breast pathology subspecialists according to American Society of Clinical Oncology and College of American Pathologists 2018 and 2023 guidelines.2
The breast cancer slides were from 4 pathology laboratories in the US, France, and Israel and were randomly assigned retrospective cases from years 2021 to 2022.1
A confirmatory study has been completed involving 2000 patients, Vescler added, and these data were presented at the 2024 San Antonio Breast Cancer Symposium. This larger global study involved 14 different medical centers across US, Europe, UK, and Brazil.
“Although we showed in this small cohort that the AI solution is robust—performing well across different labs in different staining and preanalytical conditions and scanners and so on—in the larger study, it is more evident because it spans many countries, centers, and pathologists. Thus, it was important for us to show this on a larger scale,” Vecsler said.
Next steps include submitting data to the various regulatory bodies depending on the country. Also, of note, Vecsler addressed the practical application of adapting these types of AI tools into the clinical setting. “There are pathologists who are using digital pathology. For these pathologists, adoption will be very fast because it is a ‘plug-in and play’ application,” Vecsler explained. However, for those not acclimated to digital pathology the transition may take longer and “our pathologists are available to help guide through this process,” Vecsler said. There are training programs that can be customized so everyone can effectively utilize the tool.
“The main purpose of this AI-solution is to provide the support for pathologists for determining HER2 0 vs HER2 1+, not only in terms of interobserver agreement and consistency, but also for improved accuracy. In the end, it means that more patients are identified and will receive the right treatment, which is a very important goal,” Vecsler concluded.
REFERENCES
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 Precision Oncology. Published online October 11, 2024. doi.10.1200/po.24.00353
Wolff AC, Somerfield MR, Dowsett M, et al. Human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology–College of American Pathologists guideline update. Arch Pathol Lab Med. 2023;147(9):993-1000. doi.10.5858/arpa.2023-0950-sa
Advancing Neoadjuvant Therapy for HER2+ Breast Cancer Through ctDNA Monitoring
December 19th 2024In an interview with Targeted Oncology, Adrienne Waks, MD, provided insights into the significance of the findings from the DAPHNe trial and their clinical implications for patients with HER2-positive breast cancer.
Read More
Real-World Data and AI Help Improve ILD Management in HER2+ Breast Cancer
December 13th 2024Project EVOLVE findings show how AI-driven real-world data analysis, combined with targeted educational interventions, can enhance adverse event monitoring and improve patient outcomes for those with HER2+ breast cancer.
Read More
