A Comprehensive AI-Based Approach in Classifying Breast Lesions: Focusing on Improving Pathologists' Accuracy and...

Area of Science:

• Digital pathology
• Computational pathology
• Medical artificial intelligence

Background:

• Accurate breast lesion classification is crucial for effective clinical decisions and patient management.
• Whole slide images (WSIs) are increasingly used in pathology, necessitating advanced analytical tools.
• Artificial intelligence (AI) offers potential for automating and improving diagnostic accuracy in pathology.

Purpose of the Study:

• To evaluate the performance of an AI solution for classifying breast lesions using WSIs.
• To compare the AI's classification accuracy against the ground truth established by expert breast pathologists.
• To assess the impact of AI assistance on pathologists' diagnostic accuracy and efficiency.

Main Methods:

• A cohort of 104 breast cases was analyzed, comprising invasive carcinomas, microinvasive carcinomas, ductal carcinoma in situ (DCIS), and lobular neoplasia/benign cases.
• The AI's classification performance was evaluated using metrics such as area under the curve (AUC), sensitivity, and specificity.
• Pathologists' performance with and without AI support was compared, along with changes in review time and ancillary test usage.

Main Results:

• The AI achieved high AUCs for invasive carcinoma (0.976), DCIS (0.976), and lobular neoplasm (0.953), with excellent sensitivity and specificity across categories.
• AI demonstrated strong performance in microcalcification detection (AUC 0.925, sensitivity 95%).
• AI support improved pathologists' accuracy from 97.1% to 100%, reduced review time by 16.5%, and decreased immunohistochemistry usage by 33%.

Conclusions:

• The AI solution demonstrates significant potential for accurate and efficient breast lesion classification from WSIs.
• The AI tool exhibits high sensitivity and specificity, comparable to or exceeding expert performance in certain aspects.
• Integration of this AI into routine pathology practice is supported, offering benefits in diagnostic accuracy, efficiency, and resource utilization.