DUNet: A deformable network for retinal vessel segmentation (1811.01206v1)

Abstract: Automatic segmentation of retinal vessels in fundus images plays an important role in the diagnosis of some diseases such as diabetes and hypertension. In this paper, we propose Deformable U-Net (DUNet), which exploits the retinal vessels' local features with a U-shape architecture, in an end to end manner for retinal vessel segmentation. Inspired by the recently introduced deformable convolutional networks, we integrate the deformable convolution into the proposed network. The DUNet, with upsampling operators to increase the output resolution, is designed to extract context information and enable precise localization by combining low-level feature maps with high-level ones. Furthermore, DUNet captures the retinal vessels at various shapes and scales by adaptively adjusting the receptive fields according to vessels' scales and shapes. Three public datasets DRIVE, STARE and CHASE_DB1 are used to train and test our model. Detailed comparisons between the proposed network and the deformable neural network, U-Net are provided in our study. Results show that more detailed vessels are extracted by DUNet and it exhibits state-of-the-art performance for retinal vessel segmentation with a global accuracy of 0.9697/0.9722/0.9724 and AUC of 0.9856/0.9868/0.9863 on DRIVE, STARE and CHASE_DB1 respectively. Moreover, to show the generalization ability of the DUNet, we used another two retinal vessel data sets, one is named WIDE and the other is a synthetic data set with diverse styles, named SYNTHE, to qualitatively and quantitatively analyzed and compared with other methods. Results indicates that DUNet outperforms other state-of-the-arts.

• The paper presents DUNet, which integrates deformable convolutions into U-Net to adapt receptive fields for enhanced vessel segmentation.
• It achieves state-of-the-art performance with global accuracies of 96.97%, 97.29%, and 97.24% on the DRIVE, STARE, and CHASE_DB1 datasets.
• The innovative approach improves early diagnosis of retinal and systemic diseases and offers potential for extension to other medical imaging tasks.

An Overview of DUNet: A Deformable Network for Retinal Vessel Segmentation

Abstract and Objective

The paper "DUNet: A Deformable Network for Retinal Vessel Segmentation" presents a novel approach to segmenting retinal vessels in fundus images, a critical process in diagnosing diseases like diabetes and hypertension. The authors introduce the Deformable U-Net (DUNet), which leverages local features of retinal vessels in a U-shaped architecture, aimed at enhancing automatic retinal vessel segmentation.

Methodology

DUNet incorporates deformable convolutional networks into the U-Net architecture, allowing the model to adaptively adjust receptive fields according to vessel shapes and scales. This adaptation is achieved by adding learned offsets to the sampling grid locations, as proposed in deformable convolutions. Consequently, DUNet captures context information and enables precise localization by integrating low-level and high-level feature maps.

Datasets and Experimental Setup

The authors utilize three public datasets: DRIVE, STARE, and CHASE_DB1 for training and testing DUNet, comparing its performance against existing networks like Deformable-ConvNet and the conventional U-Net. Additionally, the paper references WIDE and SYNTHE datasets to assess the generalization capability of DUNet.

Results

Empirical results indicate that DUNet outperforms existing approaches in various metrics. The network achieves state-of-the-art performance with global accuracy scores of 0.9697, 0.9729, and 0.9724 on DRIVE, STARE, and CHASE_DB1 respectively. The Area Under Curve (AUC) scores further validate its efficacy—0.9856, 0.9868, and 0.9863 respectively.

Implications

The integration of deformable convolution within U-Net demonstrates substantial improvement in handling vessel segmentation tasks, particularly in capturing vessels of varying shapes and scales. This advancement offers potential for more accurate early diagnosis of retinal and systemic diseases, enhancing clinical decision-making processes. Furthermore, DUNet's adaptability to synthetic datasets, as indicated by performance on WIDE and SYNTHE, suggests promising applications across diverse imaging environments.

Future Directions

The paper sets the stage for extending the deformable network's application to other medical imaging tasks. The paper indicates potential research avenues in expanding DUNet to three-dimensional frameworks, promising more comprehensive analysis capabilities in medical imaging.

In conclusion, DUNet represents a significant methodological advancement in retinal vessel segmentation, comprising an effective blend of deformable convolutional concepts with the established U-Net structure. It provides a framework that can efficiently tackle the challenges posed by variable vessel morphologies within complex retinal images.