Our study evaluated how an online training platform integrated with artificial intelligence could improve sensitivity and concordance amongst pathologists in identifying IBC-HER2 score and also HER2 clinical categorization. So we developed this platform with AI integration to train pathologists in HER2 masterclasses, totaling 105 pathologists from 10 countries. Each pathologist assessed 20 HER2-IDC breast cancer cases in three exams with AI assistance versus without AI assistance...
Our study evaluated how an online training platform integrated with artificial intelligence could improve sensitivity and concordance amongst pathologists in identifying IBC-HER2 score and also HER2 clinical categorization. So we developed this platform with AI integration to train pathologists in HER2 masterclasses, totaling 105 pathologists from 10 countries. Each pathologist assessed 20 HER2-IDC breast cancer cases in three exams with AI assistance versus without AI assistance. And we had a gold standard in HER2-IHC score, which was a reference center score comprised by a group of expert pathologists. So then all pathologists evaluated these 20 cases, giving IHC scores with AI assistance versus without AI assistance. What we detected amongst almost 20,000 readings was an increase in sensitivity amongst pathologists compared to the reference center score in detecting HER2-IHC score and also a decrease in discordance amongst pathologists. We also detected a higher sensitivity, higher accuracy, and higher concordance amongst pathologists in determining HER2 clinical categories. So by instance, manual scoring without AI assistance, the sensitivity was the lowest in detecting HER2-NU and HER2-Ultra-Low breast cancers. With AI support, we detected an increase in sensitivity in detecting HER2-NU and HER2-Ultra-Low breast cancers from close to 50% to higher than 90% in accuracy and also an increase in kappa levels in concordance amongst pathologists. So this is a huge improvement in accuracy and precision in detecting HER2 clinical categories. Also for HER2-low breast cancer, sensitivity in manual scoring was around 78% compared to higher than 90% with the use of AI assistants. And finally we saw a decrease by 24% in misinterpretation of cases as HER2-null breast cancer. What does this mean clinically? Is that we decrease by 24% patients who would miss the opportunity to have effective treatment, we should call them HER2-null. When using AI, we could call these patients HER2-ultra-low, HER2-low, compared to a reference center score and then these patients could have more access to affected IDC treatment.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
