[Medical image segmentation data augmentation method based on channel weight and data-efficient features].

Area of Science:

• Computer-aided medical diagnosis
• Deep learning for medical imaging
• Medical image segmentation

Context:

• Acquiring labeled medical image data is costly.
• Deep learning models often require extensive data and lack interpretability.
• High demand exists for interpretable AI in medical diagnosis.

Purpose:

• Propose a novel data augmentation method for medical image segmentation.
• Address challenges of data scarcity and lack of interpretability in deep learning models.
• Enhance model performance, data efficiency, and interpretability.

Summary:

• Utilizes gradient-weighted class activation mapping for data-efficient feature extraction, fused with original images.
• Introduces a channel weight feature extractor to learn inter-channel weights.
• Achieves non-destructive data augmentation, improving Intersection over Union (IoU) and Dice scores on Hyper-Kvasir and ISIC-Archive datasets.

Impact:

• Demonstrates strong data efficiency, maintaining 95% performance with 70% of training data.
• Enhances model interpretability through built-in interpretable features.
• Offers excellent universality and plug-and-play integration with existing segmentation methods without network modification.