Artificial Intelligence-Based Mitosis Scoring in Breast Cancer: Clinical Application

Area of Science:

• Computational pathology
• Breast cancer diagnostics
• Artificial intelligence in oncology

Background:

• Artificial intelligence (AI) excels at mitosis identification and quantification in cancer.
• Clinical implementation of AI for breast cancer (BC) requires evaluation against established methods.
• Accurate mitotic figure scoring is crucial for BC grading and prognosis.

Purpose of the Study:

• To assess the optimal method for AI-based mitotic figure scoring in breast cancer (BC).
• To compare AI-derived mitotic counts (mitotic count per tumor area [MCT], mitotic index [MI], mitotic activity index [MAI]) with visual scoring and clinical outcomes.

Main Methods:

• Utilized whole slide images from large BC cohorts (Nottingham, TCGA-BRCA).
• Employed automated mitosis detection to calculate MCT, MI, and MAI.
• Evaluated AI metrics against the Nottingham grading system's visual scoring, Ki67 scores, clinicopathologic parameters, and patient survival.

Main Results:

• All AI-based mitotic scores (MCT, MI, MAI) correlated with clinicopathologic characteristics and survival (P < .001).
• Only MAI and MCT showed positive correlation with the gold standard visual scoring (r=0.8, r=0.7) and Ki67 scores (r=0.69, r=0.55).
• MAI was the sole independent predictor of survival in multivariate analysis (P < .05).

Conclusions:

• The optimal AI method for BC mitosis scoring needs careful consideration for clinical application.
• Mitotic activity index (MAI) using AI provides reliable, reproducible, and accurate quantification of mitotic figures in breast cancer.
• MAI demonstrates superior performance as an independent prognostic marker in breast cancer.