Research And Implementation Of Granular Computing U-net For Retinal Vessels Segmentation

At present,the use of deep neural networks to achieve intelligent and accurate segmentation of retinal vessels has become a hot research topic in the field.However,Retinal vessels have ambiguous boundaries,various thickness,and are even obscured by lesion areas.These phenomena will cause deep neural networks to face two types of uncertain information when segmenting retinal vessels,that is uncertainty in feature channels and boundary labels.The uncertainty in feature channels will affect the channel attention coefficient,so that the deep neural network cannot pay attention to the detailed features of retinal vessels.The uncertainty in boundary labels will lead to incorrect back propagation information,making it impossible to train an effective segmentation model.In order to solve the influence of the above uncertainty information on the segmentation results,this paper combines deep neural networks with the theory related to granular computing to design robust retinal vessels segmentation models and implement an auxiliary diagnosis application system of retinal vessel intelligent segmentation.The main work contents and contributions are as follows:(1)For the uncertainty of feature channels,this paper proposes a retinal vessels segmentation method based on a rough channel attention mechanism.Firstly,the method integrates the ability of deep neural networks in learning complex features and the ability of rough sets in handling uncertainty to design rough neurons.Secondly,a rough channel attention mechanism module is constructed based on rough neurons,and it is embedded in the skip connection of the U-Net model to achieve the fusion of high-level and low-level features.Then the residual connections are added to transfer low-level features to high-level features,which not only enriches the network feature extraction but also helps to back-propagate the gradient when training the model.Finally,the effectiveness of the method is verified on three retinal vessels datasets.The experimental results show that the proposed method has obvious improvement in both accuracy and sensitivity.(2)For the uncertainty of boundary labels,a U-Net retinal vessels segmentation method based on the three-way loss function is proposed.First of all,the method uses dilation and erosion operators to construct the upper and lower bounds of the uncertain boundary respectively,and maps the vascular boundary with uncertain information into a range.Then the uncertainty of the boundary is united with the network loss function,and the three-way loss function is proposed.The loss function calculates the cross-entropy loss between the predicted boundary and the manual segmentation of the gold standard map,the upper boundary of the uncertain boundary and the lower boundary of the uncertain boundary,and generates the final loss by weighted fusion.The total loss of the three-way loss function is used to train the network parameters using stochastic gradient descent algorithm.Finally,the effectiveness of the method is verified on three retinal vessels datasets.The experimental results show that the proposed three-way loss function can obviously improve the accuracy of segmentation.(3)On the basis of the above research,an auxiliary diagnostic application system for intelligent retinal vessels segmentation is designed and implemented.The data acquisition module is responsible for the input of the basic information of the patient and the upload of the color fundus retinal blood vessel images taken by fundus camera.The retinal vessels intelligent segmentation module uses the U-Net model based on the rough channel attention mechanism and the U-Net model based on the three-way loss function proposed in this paper to realize the intelligent segmentation of color fundus images.In the auxiliary diagnosis module,the doctor analyzes the patient’s condition and fills in the diagnosis opinion based on the basic information of the patient,the color retinal vessels images captured by the fundus camera and the corresponding retinal vessels segmentation result maps.The application system has good practicability for auxiliary diagnosis of ophthalmologists and cardiologists.This paper combines the ability of granular computing to handle uncertain information and the efficient feature learning capability of deep learning models to achieve accurate segmentation of fine-grained retinal vessels in fundus images,and develops an auxiliary diagnosis application system based on the proposed two types of models.The relevant research results have good research significance and practical value for the intelligent segmentation of retinal vessels.