Recurrent residual U-Net for medical image segmentation.

Area of Science:

• Medical image analysis
• Deep learning
• Computer vision

Background:

• Deep learning (DL) excels in medical image analysis tasks like segmentation.
• U-Net is a popular DL architecture for these applications.
• Existing models face challenges in deep architecture training and feature representation.

Purpose of the Study:

• To propose novel recurrent U-Net models (RU-Net and R2U-Net) for improved medical image segmentation.
• To leverage residual networks and recurrent convolutional neural networks for enhanced feature representation.
• To achieve better performance in medical image segmentation with comparable network parameters.

Main Methods:

• Developed Recurrent U-Net (RU-Net) and Recurrent Residual U-Net (R2U-Net) architectures.
• Integrated residual units for training deep architectures.
• Employed recurrent residual convolutional layers for feature accumulation and representation.
• Validated models on blood vessel, skin cancer, and lung lesion segmentation datasets.

Main Results:

• Proposed RU-Net and R2U-Net models demonstrated superior segmentation performance.
• Achieved better feature representation through recurrent residual convolutional layers.
• Outperformed established models like SegNet and U-Net on benchmark datasets.
• Enabled design of more effective U-Net architectures with similar parameter counts.

Conclusions:

• RU-Net and R2U-Net offer significant improvements in medical image segmentation.
• The proposed architectures provide enhanced feature representation and training stability for deep networks.
• These models represent a promising advancement for various medical image segmentation applications.