Research On Segmentation Algorithm Of Liver CT Image Based On Improved U-Net Architecture

The liver is the human body’s largest digestive gland with two blood supply systems.It is mainly located in the right quarter rib and upper abdomen.Among the most common causes of malignant tumors and tumor deaths in the world,liver cancer ranks fifth,and the fatality rate ranks three,and early diagnosis and treatment have a greater effect on reducing the incidence and mortality of liver cancer.The liver segmentation task refers to the task of segmenting the liver from a given abdominal CT image.The precise division of liver contours can help doctors grasp the health of the liver,determine the area of liver lesions,and help doctors formulate surgical planning before surgery to ensure surgery try to preserve the healthy part of the liver while removing the diseased area.Due to the pathological changes of the liver in abdominal CT images,which result in different liver shapes,low contrast between the liver and neighboring organs,and unstable image quality,the task of liver segmentation for abdominal CT images faces many challenges.With the continuous development of computer vision algorithms and deep learning methods,convolutional neural networks have been more and more widely used in biomedical image segmentation.Among them,the U-Net network model composed of the contraction path for obtaining context information and the expansion path for precise positioning achieves better results in medical image segmentation.However,the original U-Net network model still has a lot of room for improvement in segmentation performance,but if conventional methods such as increasing the number of network layers by increasing the convolution kernel to improve the model effect are adopted,the parameter amount of the model and the network will be increased.The amount of calculations will cause the training and reasoning of the network model to become more difficultly,and it also puts higher requirements on the mathematical operation of the computer.In this paper,in order to improve the effect of the U-Net network model in liver CT image segmentation,that is,to improve the segmentation evaluation of DICE similarity index,recall rate and accuracy rate in liver segmentation,avoid adopting routines such as increasing the amount of network parameters or improving computer computing performance.Approach and propose a new solution.The first is to use the limited contrast adaptive equalization algorithm to enhance the specific area in the abdominal CT image,thereby improving the liver contrast in the region of interest,and providing better data for the network model and the effective recognition and segmentation of the human eye,the second is to detect and segment for the problem of small targets at the proximal and distal ends of the liver,the attention mechanism algorithm is introduced to improve and optimize the convolution module of the network to improve the detection and recognition ability of small targets in the imbalance of the network processing sample,finally,in order to improve the network model in the boundary processing on the performance of upper and location restoration,the dense connection mechanism is introduced to optimize and improve the network model,so as to solve the problems of poor edge processing ability,edge loss and unclear boundary in liver segmentation.Experimental results show that the improved U-Net network model can better adapt to liver segmentation of abdominal CT images with different pathological degrees.This method can effectively improve the main performance of liver segmentation.The DICE similarity coefficient has been increased from 86.43%to 89.43%.The recall rate has increased from 88.65%to 92.15%,and the accuracy rate has been increased from 92.43%to 93.64%.There are cerrtain improvements in the main evaluation indicators.In order to effectively use the liver segmentation results of abdominal CT images based on the improved U-Net network,this paper conducts certain research on the application of the segmentation results,and provides a technical route for the effective use of liver segmentation results.The research mainly includes methods such as liver extraction in abdominal CT images,redundant information cropping,and focus information pixel adjustment,which improves the detection accuracy of unobvious liver focus segmentation in liver focus segmentation.The experimental results prove that the two-step method of segmenting the liver first and then segmenting the lesion can effectively cope with the problem of inconspicuous targets in lesion segmentation,which provides practical preparation for in-depth study of liver lesion segmentation.