Retinal Blood Vessel Segmentation Based On Improved Pulse Coupled Neural Network Combined With U-Net

Pathological changes of retina are closely related to many human diseases,such as hypertension and diabetes.In clinical medicine,the pathological conditions of retinal blood vessels are usually used to diagnose various related diseases in human body.Retinal vascular segmentation is the basis of this medical diagnosis and plays an important role in the screening and diagnosis of related diseases.Because the whole tubular structure of fundus blood vessels is complex,the contrast between the target pixel and the background pixel in the blood vessel image is low,and it is difficult to identify and accurately segment the tiny blood vessels,so it is very difficult to segment the fundus blood vessels.In order to solve the problems such as the loss of tiny details of blood vessels and poor connectivity of blood vessels during segmentation.In this thesis,a new segmentation algorithm is proposed.The algorithm uses the dense deformable convolution U-Net of the dual-pool SE(Squeeze-and-Excitation)module,and combines the simplified pulse coupled neural network(SPCNN)model with adaptive threshold to segment blood vessels.The purpose is to ensure the accuracy,keep more details of tiny blood vessels,reduce blood vessel breakage and improve the robustness and sensitivity of the segmentation network.The main contents of this thesis are as follows:(1)An image enhancement algorithm of fundus blood vessels based on U-Net is designed.The network in this algorithm is mainly based on U-Net coding and decoding network structure,and dense deformable convolution module and SE module with double pooling mechanism are introduced to extract blood vessel features at multiple scales,pay more attention to global information,improve the accuracy of capillary segmentation and keep more details of tiny blood vessels.(2)Based on the enhancement of fundus blood vessel image,the enhanced image is combined with the original image,and an A-SPCNN network is designed to segment the fused image.The purpose is to improve the performance of retinal blood vessel segmentation by combining the advantages of supervised learning and unsupervised learning.The adaptive threshold makes the segmentation more targeted,and the coupling characteristics of PCNN can connect the broken blood vessels while retaining more tiny blood vessels,ensuring the connectivity of blood vessels.The segmentation effect is tested on DRIVE and STARE data sets.The sensitivity of the model is 0.8335 and 0.8315 on DRIVE and STARE data sets,the specificity is 0.9767 and 0.9777 respectively,and the accuracy is 0.9749 and 0.9570 respectively.The overall index is better than the existing algorithms.