Method Research Of Liver CT Image Automatic Segmentation Based On Optimal Search Strategy

China has a high incidence of liver cancer,the liver surgery is the main treatment.However,manual interpretation of pathological images before operation is inefficient and prone to misjudgement.Computer-aided diagnosis technology plays an important role in guiding liver surgery.Accurate segmentation of liver from CT images is an important step in computer-aided diagnosis.Because of the low gray difference with adjacent tissues in CT images and similar shape between liver,liver segmentation is difficult.At present,the clinical system is mainly through manual segmentation,which is not only time-consuming and labor-intensive,but also prone to deviations.For the existing methods of three-dimensional CT image liver segmentation,manual participation is the majority,but there are defects.On the basis of previous work,the paper introduces the voting scoring strategy to the automatic liver segmentation method to realizes the optimal search of liver segmentation sets,and completes the accurate liver segmentation.The main work of this paper is as follows:(1)In the preprocessing stage of the image,a series of preprocessing steps designed,including image enhancement and image sharpening.Image enhancement use logarithmic transformation to optimize the three-dimensional image whose effect is to make the liver contour clearer and can be easily distinguished from other tissues.Imagesharpening use Laplacian method to highlight the boundary of the liver.Aiming at the problem of obtaining seed points by using three-dimensional region growing method,a method of automatically obtaining seed points from liver parts of three-dimensional CT images is proposed.At present,seed points are mainly selected by human-computer interaction,which not only increases the workload of people,but also tends to produce deviations.Based on the statistical analysis of CT values on three-dimensional CT images,the automatic acquisition of seed points in liver is realized.(2)In the stage of determining the optimal initial contour of the liver,an improved three-dimensional region growing method for image segmentation is proposed.Methods of Curvature-driven stream filter was used to filter the preprocessed image,and a series of liver initial contour segmentation sets are obtained by two voting criteria: average-cut and maxmin-cut.The experimental results show that the three-dimensional region growing method incorporate curvature-driven stream filter filtering and voting scoring strategy,and the segmentation results are significantly improved.(3)The accurate liver segmentation model includes two stages:image correction and accurate liver segmentation.In the phase of image correction,an image correction method is proposed in this paper.According to the method of the relationship between gray histogram ofliver region and neighborhood pixels,the segmentation results were corrected.In the stage of accurate liver segmentation,the optimal segmentation result is selected by voting scoring strategy on the level set of geodesic active contour and the two segmentation results generated by three-dimensional region growth method.In the method of geodesic active contour level set segmentation method,the selection of the initial level set is a difficult problem.In this paper,the optimal initial contour of the liver is used as the initial level set,which avoids increasing the computational complexity from the global energy function calculation and improves the efficiency of segmentation.The segmentation results are analyzed and evaluated,and compared with the methods in other literatures,show method is more accurate and robustness.