Research And Implementation Of Neural Segmentation Based On Deep Learning Method

At present,ultrasound images become the most common images in the medical field,which can assist doctor diagnose patient's health condition.Ultrasound images also play an important role in clinical surgery.Nowadays,local anesthesia is the most common anesthetic method using in the clinical surgery.The successful implementation of the surgery requires accurate judegment of the nerve area in the ultrasound image.Traditional image segmentation methods rely on artificial means to extract and select the information such as the edge,color and texture from the image.However,these methods not only consume a lot of personnel and energy,but also requires a certain amount of professional knowledge.Sometimes,it can easily cause a large amount of expenditure but did not get useful feature information.As one of the most efficient image segmentation methods,convolutional neural network has achieved good results in the field of image segmentation.Convolutional neural network does not need to manually extract image features,it can automatically extract feature information from the image through the image convolution operation.Using convolutional neural network to process the image,can not only saves energy,but also extracts more useful features,and the segmentation effect is more accurate.Utrasound images,as one of the most complicated images,should not only judge the semantic information but also get the location of the information at the same time.This need analyzes ultrasound images pixel-by-pixel.This paper focus on solving ultrasound nerve segmentation problem using deep learning method.Based on the real medical ultrasound image data,we propose some new deep learning models to better solve this problem.The main works are as follow;1.Research the traditional deep learning model,including Fully Convolutional networks,U-NET networks,and SegNet networks.Build these three models and use them to solve the ultrasound nerve segmentation problem.The real medical ultrasound image data is used to verify the experimental results.The experimental results are compared to find the shortcomings of the three models.2.Study the neural segmentation method based on the Residual U-shaped networks.U-NET network has a wide range of applications in medical image segmentation,but compared to other deep learning models,U-NET's structure is not deep enough.Therefore,we propose a new neural segmentation method,residual U-shaped network.Experiment shows that the segmentation effect of the residual U-shaped network is better than the other three traditional network structures.3.Study the neural segmentation method based on the Residual Segmentation network.SegNet has a wide range of applications in road recognition,but its network structure is too complicated and its training time is too long for the ultrasound image segmentation.Therefore,another neural segmentation method,Residual Segmentation networks are proposed in this paper.Experiment shows that Residual Segmentation network has better segmentation effect and shorter training time comapared with the SegNet network.