Research On Medical Image Segmentation Technology Based On U-Net

Medical Image segmentation is an important foundation of modern medical image analysis.The cross-fusion of deep learning and medicine aims to assist doctors to extract human tissue or focus area,and analyze large amount of medical image data efficiently and accurately.Therefore,using the computer learning ability to assist doctors in medical image analysis can save the time and energy of doctors in analysis and balance the medical resources in different development areas.Nowadays,with the improvement of computer performance and the increase of medical data,the traditional method of manually extracting image features has certain limitations.Deep learning can extract the local detail information and high-level semantic abstract information of the image through the superposition of multi-layer neural network.This paper deeply studies the medical image segmentation technology based on deep learning,and proposes two new medical image segmentation algorithms based on u-net.Finally,the segmentation model proposed in this paper shows good performance on multiple medical image segmentation public data sets,and obtains better visual segmentation results.The following is the main research work in this paper:First of all,This paper improves the classical U-Net,proposes the RAU-Net which is a medical image segmentation algorithm based on residual network and attention mechanism,and verifies the performance of RAU-Net on the data set on lung infection caused by COVID-19.Because the U-Net network has insolvenous problems with undersal segmentation and feature extraction,RAU-Net enhances the ability of the model extraction characteristics by using the residual block in the codec path,thereby more efficiently extracting the lesion region characteristics.RAU-Net processes the encoded feature of skipping connection by paying the door AG,making the network model more attention to the COVID-19 infected lesion zone,inhibiting the characteristic response of the background area in the CT image,and effectively reducing the problem of blurring the boundary of the disease.The next,R2AU-Net is proposed based on recurrent neural network,and the performance of R2AU-Net is verified on three medical image segmentation tasks.RAUNet has good segmentation performance for COVID-19 lung infection region,but it ignores the correlation of pixels in ROI,and has some incoherence for some medical image segmentation results.R2AU-Net uses recurrent residuals convolution blocks for feature extraction to enhance the model’s ability to integrate contextual information.For different forms of medical images,recurrent residual convolution blocks can be used to extract critical shallow features,and then attention gates can be used to aggregate shallow details and high-level abstract semantic information.Finally,this paper has adopted different training strategies for different medical images.Using the thoughts of random regimen,improve the Re LU activation function,further refine the segmentation results of skin lesions,and reintegrate data enhancement of retinal vascular data,and re-production of pulmonary labels.The experimental results also prove the effectiveness of different strategies for different segmentation tasks.