MedNet.nl: A study involving 105 pathologists from 10 countries has shown that artificial intelligence (AI) support improves the assessment of HER2 expression on breast tumors. The accuracy of categorizing tumors as HER2-positive, low, ultralow, and negative increased from 90.1% without AI assistance to 95.0% with AI support.
With the development of new HER2-targeted antibody–drug conjugates, it has become relevant to identify tumors with relatively low HER2 expression in addition to classical HER2 expression. However, the analysis of so-called HER2-low or HER2-ultralow tumors is challenging due to the low HER2 expression, which leads to a proportion of tumors being incorrectly classified as HER2-negative. A group of international researchers investigated whether AI can support pathologists in correctly classifying tumors. To do this, the pathologists used an AI-supported online training platform from the German company Mindpeak.
Participants were 105 pathologists from 10 countries in Asia, Africa, and South America. They assessed 20 digital HER2-stained tumor samples, supported in part by AI. Among other things, AI support led to higher scores on an exam of samples previously assessed by independent experts. Without AI, accuracy was 89.1% and with AI, it was 96.1%. The use of AI also led to greater concordance among pathologists; it increased from 0.506 to 0.798. In particular, there was a relatively large effect on classifying tumors with low or ultra-low HER2 expression. The use of AI reduced misclassification of these tumors by 24.4%.
According to the researchers, the findings underscore the value of AI in assessing markers such as HER2. In a follow-up study, they aim to determine the clinical effects of more accurately identifying patients with HER2-low or HER2-ultralow breast cancer.
This article was originally published in Dutch on MedNet.nl