DCSAU-Net: A deeper and more compact split-attention U-Net for medical image segmentation

Area of Science:

• Computer Vision
• Medical Image Analysis
• Deep Learning

Background:

• Convolutional Neural Networks (CNNs) and U-Net architectures have advanced biomedical image segmentation.
• Existing U-Net models face limitations in extracting diverse features from varying depths due to uniform encoder design.
• Increasing complexity in medical images necessitates improved segmentation techniques.

Purpose of the Study:

• To propose a novel, deeper, and more compact split-attention U-shaped network (DCSAU-Net) for enhanced medical image segmentation.
• To efficiently leverage both low-level and high-level semantic information.
• To address the limitations of current methods in handling complex medical image data.

Main Methods:

• Developed a novel U-Net architecture incorporating a compact split-attention block.
• Implemented primary feature conservation frameworks to improve information utilization.
• Evaluated the model on diverse datasets: CVC-ClinicDB, 2018 Data Science Bowl, ISIC-2018, SegPC-2021, and BraTS-2021.

Main Results:

• The proposed DCSAU-Net demonstrated superior performance over state-of-the-art methods.
• Achieved higher mean intersection over union (mIoU) and Dice coefficient scores.
• Showcased excellent segmentation accuracy, particularly on challenging medical images.

Conclusions:

• The novel DCSAU-Net effectively extracts multi-level features for improved medical image segmentation.
• The model offers a significant advancement in handling complex medical imaging data.
• The proposed architecture provides a robust solution for various biomedical segmentation tasks.