Neural Network-Based Mammography Analysis: Augmentation Techniques for Enhanced Cancer Diagnosis-A Review

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Biomedical Data Science

Background:

• Machine learning (ML) advancements in breast cancer detection are driven by annotated mammography datasets.
• Key datasets like CBIS-DDSM, VinDr-Mammo, and CSAW-CC are crucial for training classification and detection models.
• Improving breast cancer screening relies on robust ML models and well-designed datasets.

Purpose of the Study:

• To review mammography studies applying ML for breast cancer detection.
• To analyze data augmentation techniques and their impact on abnormality detection.
• To discuss challenges like dataset imbalance and explore solutions for enhanced screening.

Main Methods:

• Review of existing mammography studies utilizing prominent datasets (CBIS-DDSM, VinDr-Mammo, CSAW-CC).
• Analysis of data augmentation strategies and their performance in detecting breast tissue abnormalities.
• Examination of techniques for addressing dataset imbalances, including synthetic data generation and Generative Adversarial Network (GAN) augmentation.

Main Results:

• Data augmentation techniques show potential for improving the performance of ML models in mammography.
• Methods like synthetic data generation and GAN augmentation offer solutions for dataset imbalance challenges.
• The design of datasets, quality of annotations, and choice of ML architectures significantly impact screening model efficacy.

Conclusions:

• ML techniques, coupled with effective data augmentation and strategies for imbalance, are vital for advancing breast cancer detection.
• Well-designed datasets with detailed annotations are fundamental for developing reliable breast cancer screening tools.
• Future research should focus on refining augmentation, tackling imbalance, and improving model interpretability (e.g., using Grad-CAM) for clinical translation.