Multi-dimensional cascades neural network models for the segmentation of retinal vessels in colour fundus images

Deep learning has recently received attention as one of the most popular methods for boosting performance in different sectors, including medical image analysis, pattern recognition and classification. Diabetic retinopathy becomes an increasingly popular cause of vision loss in diabetic patients.. Retinal vascular status in fundus images is a reliable biomarker for diabetes, hypertension and many ophthalmic diseases. Therefore, accurate segmentation of retinal vessels is of great significance for the diagnosis of many diseases. However, due to the inherent complexity of the retina itself and the lack of data, it is difficult to obtain the ideal accuracy of the segmentation results of the vascular end. To solve this problem, we propose an innovative multi-dimensional deep convolutional Neural network (MDUNet) to segment the retinal vessels in fundus images. The fusion of cross-dimensional transformation makes full use of the relevance of information between different dimensions. Meanwhile, the self-attention calculation method of cross-window is applied to effectively reduce the computational complexity. MDUNet is proposed to provide a research basis for the application of Transformer structure in the field of medical image segmentation. The proposed method is evaluated on different evaluation metrics such as sensitivity, specificity, and accuracy. Experimental results on six public datasets show that the proposed work MDUNet achieves better vessel segmentation accuracy with a smaller number of parameters compared with classical models such as U-Net, SegNet, and DeepLabv3+.