Deep learning performance for detection and classification of microcalcifications on mammography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Mammography is vital for early breast cancer detection, but radiologist capacity is limited.
• Artificial intelligence (AI) offers a solution for evaluating microcalcifications in mammograms.
• This study focused on developing and testing an AI model for microcalcification analysis.

Purpose of the Study:

• To develop and evaluate an AI model for the localization and characterization of microcalcifications on mammograms.
• To assess the performance of different neural networks (AlexNet, ResNet18, ResNet34) in detecting and classifying microcalcifications.
• To determine the potential of AI to support radiologists in mammogram interpretation.

Main Methods:

• A dataset of 1,986 mammograms from 1,000 patients was annotated by expert radiologists.
• Three neural networks (AlexNet, ResNet18, ResNet34) were trained, validated, and tested.
• Performance was evaluated using metrics such as sensitivity, specificity, and Area Under the Curve (AUC).

Main Results:

• AlexNet achieved the highest performance with 0.98 AUC for detection and 0.94 AUC for classification.
• ResNet18 and ResNet34 also demonstrated high accuracy in detection (0.98 AUC) and classification (0.92 AUC).
• The AI models showed high sensitivity and specificity in identifying malignant and benign microcalcifications.

Conclusions:

• The developed AI models accurately detect and characterize microcalcifications on mammography.
• AI-based systems can significantly assist radiologists in interpreting mammograms.
• This research underscores the importance of reliable deep learning models for breast cancer screening.