Data augmentation for medical imaging: A systematic literature review

Area of Science:

• Medical Imaging
• Deep Learning
• Computer Vision

Background:

• Collecting large medical imaging datasets is challenging due to privacy and cost.
• Deep learning models require extensive data for optimal performance.
• Data augmentation offers a solution to expand training datasets artificially.

Approach:

• Systematic literature review analyzing over 300 articles (2018-2022).
• Investigated various data augmentation strategies in the medical domain.
• Assessed the impact of augmentation on clinical tasks like classification, segmentation, and lesion detection.

Key Points:

• Data augmentation techniques range from simple transformations to complex generative models.
• Specific augmentation strategies are crucial for generating plausible medical data and regularizing deep neural networks.
• Augmentation effectively addresses underrepresented classes, such as generating artificial lesions.

Conclusions:

• Data augmentation is highly effective across diverse organs, imaging modalities, and tasks.
• Augmentation improves performance regardless of dataset size.
• The review identifies promising future research directions in medical data augmentation.