Introduction
Expectations surrounding artificial intelligence (AI) in pathology are high. In contrast, little is known about the impact of AI on clinical practice, especially in tedious tasks such as mitotic activity scoring in breast cancer. Mitotic activity is a critical feature in the Nottingham grading system. Despite its high prognostic value, mitosis scoring suffers from interobserver variability and is often considered time-consuming, making it a good candidate to benefit from AI support. In this large-scale international validation study, we measured the impact of an AI algorithm (Aiosyn Mitosis Breast) aimed at detecting mitoses in breast cancer slides on mitotic scoring. The effect of AI was evaluated on time consumption, accuracy (compared to a panel of 3 experts), and interobserver variability.
Material and methods
The study involved 28 pathologists from 9 countries. A total of 210 whole slide images (WSIs) of biopsies and resections stained with H&E and originating from 8 international centers were selected. Participants were divided into 2 groups following a split-plot, each reading 105 WSIs twice in a randomized order: once with AI and once without in two sessions separated by a washout period of 3 weeks.
Results and discussion
We observed a time gain of 10.4% with AI (p< 0.001), a 15.5% increase in productivity on resections, and a decrease in interobserver variability (p<0.001). Accuracy did not decrease with AI (p< 0.001, non-inferiority).
Conclusion
Our study suggests that readily available AI can play a significant role in pathology globally by decreasing time consumption and improving interobserver reproducibility and, therefore, patient care.