Self-supervised learning and transformer-based technologies in breast cancer imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer is a leading global malignancy in women.
• Medical imaging is crucial for early detection, diagnosis, and treatment planning.
• Artificial intelligence (AI) offers novel approaches to breast image analysis.

Purpose of the Study:

• To provide a comprehensive overview of recent developments in AI for breast imaging.
• To highlight the application of self-supervised learning (SSL) and transformer models in breast lesion analysis.
• To discuss the advantages, limitations, and future directions of these AI technologies.

Main Methods:

• Review of recent studies utilizing self-supervised learning (SSL) for breast image analysis.
• Analysis of transformer-based architectures, including Vision Transformers, for capturing global contextual information in breast images.
• Examination of AI models for breast lesion segmentation, detection, and classification.

Main Results:

• SSL demonstrates label-efficient strategies and strong performance in breast image analysis.
• Transformer models effectively capture long-range dependencies and global context, complementing traditional methods.
• Representative studies showcase advancements in AI-driven breast lesion segmentation, detection, and classification.

Conclusions:

• AI, particularly SSL and transformer models, shows significant promise for improving breast cancer imaging.
• Clinical translation requires multi-institutional datasets, external validation, and interpretable AI models.
• Future research should focus on generalizability, real-world performance validation, and clinician trust for safe deployment.