Deep convolutional neural networks for mammography: advances, challenges and applications

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Traditional computer-aided detection (CAD) for mammography has limitations.
• Early breast cancer detection is crucial, and accurate diagnosis impacts patient outcomes.
• Deep learning (DL), particularly convolutional neural networks (CNNs), shows promise in medical imaging analysis.

Purpose of the Study:

• To review recent convolutional neural network (CNN) applications in mammography image analysis.
• To identify best practices for improving diagnostic accuracy using CNNs.
• To provide insights into CNN architectures and publicly available mammography datasets.

Main Methods:

• A detailed review of 83 research studies on CNNs in mammography.
• Analysis of CNN architectures for various mammography tasks.
• Summary of common publicly available mammography image repositories.

Main Results:

• Identified strengths, limitations, and performance of recent CNN applications.
• Detailed insights into CNN architectures used for lesion detection, risk assessment, and classification.
• Comparison of publicly available mammography datasets.

Conclusions:

• The survey serves as a resource for mammography research, guiding database selection.
• Best practices like image pre-processing and multi-view imaging enhance CNN performance.
• Techniques such as transfer learning, data augmentation, and dropout combat overfitting and improve model generalization.