Accuracy and efficiency of an artificial intelligence tool when counting breast mitoses

Area of Science:

• Digital pathology
• Computational pathology
• Oncology research

Background:

• Mitotic count is a key prognostic indicator in breast carcinoma.
• Manual mitotic figure counting suffers from significant inter- and intra-laboratory variability.
• Artificial intelligence (AI) with whole slide imaging presents a solution to standardize mitotic counting.

Purpose of the Study:

• To evaluate an AI tool designed for quantifying mitotic figures in invasive breast ductal carcinoma whole slide images.
• To assess the impact of AI assistance on pathologist accuracy and efficiency in mitotic counting.

Main Methods:

• Whole slide images of 320 invasive breast ductal carcinoma cases were scanned.
• An AI algorithm was trained and validated using expert pathologist annotations.
• 24 readers with varying expertise counted mitotic figures with and without AI support on 140 high-power fields.

Main Results:

• AI support improved accuracy, precision, and sensitivity for mitotic figure counting across all experience levels.
• The majority of readers (87.5%) identified more mitoses with AI, and 54.2% reduced false positives.
• AI assistance led to an overall time saving of 27.8% in the counting task.

Conclusions:

• AI tools enhance pathologist accuracy and efficiency in quantifying mitotic figures in digital breast cancer images.
• AI assistance promotes higher inter-pathologist agreement, potentially standardizing diagnostic practices.
• There is considerable enthusiasm within pathology for integrating AI into routine tasks like mitotic counting.