[Paper Review] Recurrent Residual Convolutional Neural Network based on U-Net (R2U-Net) for Medical Image Segmentation

Md Zahangir Alom, Md. Mahmudul Hasan+3 more|arXiv (Cornell University)|Feb 19, 2018

Brain Tumor Detection and ClassificationNeuroscience681 citations

TL;DR

This paper introduces R2U-Net, a recurrent residual CNN based on U-Net for medical image segmentation, combining residual units and recurrent convolution to improve feature representation.

ABSTRACT

Deep learning (DL) based semantic segmentation methods have been providing state-of-the-art performance in the last few years. More specifically, these techniques have been successfully applied to medical image classification, segmentation, and detection tasks. One deep learning technique, U-Net, has become one of the most popular for these applications. In this paper, we propose a Recurrent Convolutional Neural Network (RCNN) based on U-Net as well as a Recurrent Residual Convolutional Neural Network (RRCNN) based on U-Net models, which are named RU-Net and R2U-Net respectively. The proposed models utilize the power of U-Net, Residual Network, as well as RCNN. There are several advantages of these proposed architectures for segmentation tasks. First, a residual unit helps when training deep architecture. Second, feature accumulation with recurrent residual convolutional layers ensures better feature representation for segmentation tasks. Third, it allows us to design better U-Net architecture with same number of network parameters with better performance for medical image segmentation. The proposed models are tested on three benchmark datasets such as blood vessel segmentation in retina images, skin cancer segmentation, and lung lesion segmentation. The experimental results show superior performance on segmentation tasks compared to equivalent models including U-Net and residual U-Net (ResU-Net).

Motivation & Objective

• Motivate improving medical image segmentation performance beyond standard U-Net.
• Propose architectures that blend U-Net with residual learning and recurrent convolutions.
• Evaluate RU-Net and R2U-Net on diverse medical imaging tasks to demonstrate enhanced segmentation quality.

Proposed method

• Introduce RU-Net (Recurrent U-Net) integrating recurrent convolutional layers into the U-Net framework.
• Develop R2U-Net (Recurrent Residual U-Net) by incorporating residual connections with recurrent units within U-Net.
• Leverage feature accumulation from recurrent residual layers to enhance representations without increasing parameter count.
• Compare against baseline U-Net and ResU-Net on standard medical segmentation benchmarks.

Experimental results

Research questions

• RQ1 Do recurrent residual convolutional layers within a U-Net backbone improve medical image segmentation performance compared to standard U-Net and ResU-Net?
• RQ2 Does RU-Net/R2U-Net offer better feature representation and training dynamics for retinal vessel, skin lesion, and lung lesion segmentation?

Key findings

• RU-Net and R2U-Net leverage residual and recurrent convolutions to improve segmentation quality.
• Proposed models show superior performance over corresponding U-Net and ResU-Net baselines on benchmark medical datasets.
• Feature accumulation via recurrent residual layers yields richer representations for segmentation tasks.