Research On CT Image Segmentation Of Liver Tumor Based On Deep Learning

Liver tumor is a great threat to human health.Hepatic malignancy,also known as liver cancer,is one of the most common internal malignancies in the world and one of the leading causes of cancer death.Early detection,diagnosis and accurate staging of liver cancer is an important part of imaging research.Even with a rapidly growing understanding of the anatomy and pathology of the patient’s liver,and with technical assistance in surgery,planning liver intervention remains a challenge for surgeons.In fact,few medical decisions are made without the use of imaging technology,among which CT has the advantages of convenient,rapid and safe examination,high spatial resolution,and solving the problem of overlapping traditional X-ray images.In CT scans,manual segmentation is tedious and prohibitively time-consuming for clinical Settings.Therefore,the research of automatic segmentation method has high clinical application value.Aiming at the current research hotspot,this paper studies the problem of difficult segmentation of CT images of liver tumors based on deep learning technology,because small targets account for a very small proportion of the background.Liver tumor are small targets and usually make up a very small proportion of the background.In order to achieve the accurate segmentation of liver tumors,this paper takes two steps to segment liver tumors.Firstly,the larger target liver is segmented.Based on a U-Net network structure,which is the most popular in medical image segmentation,this paper uses dilated convolution and ordinary convolution crossover to carry out downsampling in the encoding stage to increase the receptive field.Correspondingly,in the decoding stage,transposed convolution is used to replace the usual bilinear interpolation as the up-sampling method,which can automatically optimize the weight.In the second step,on the basis of the liver segmentation in the first step,in order to reduce the accidental noise interference in the target area as far as possible,the region outside the liver was cropped and the processed image was unified in size.Similarly,based on the U-Net framework,the hybrid dilated convolution is used to increase the image sensing field,and the Tversky Loss function is further used to effectively improve the quasi-imbalance in the segmentation of liver tumors.The experimental results show that the use of network segmentation can be well for the liver and liver tumor area of the image segmentation.The proposed method is superior to the traditional image segmentation method,as the segmentation of liver and liver tumor dice similarity coefficient has achieved 0.92 and 0.68 respectively,thus having a certain clinical reference value.