Research On Interactive Medical Image Segmentation And Application

The medical image segmentation method extracts the medical object region for subsequent analysis.It is a very important and fundamental research branch of medical image analysis.Many image segmentation algorithms have been proposed.According to the amount of user interaction,we can broadly class these approaches into three categories:manual delineation,automatic segmentation algorithms and interactive segmentation algorithms.However,manual delineation is time-consuming and laborious for radiologists,and the subjective differences between radiologists cannot be ignored.The automatic segmentation methods often can’t obtain sufficiently accurate and robust segmentation performance when it used to segment medical images with poor image quality.Moreover,the surgeon’s interactive input is essential in the application of clinical surgical guidance and target area delineation.The interactive segmentation methods are obtained accurate and robust segmentation performance,as integrating the user’s knowledge can take into account the application requirements and make it easier to distinguish different tissues.In contrast to the automatic and manual segmentation methods,the interactive image segmentation balanced the accuracy and efficiency,and has a widely applications in clinic.A good interactive segmentation method should require as little user time as possible to reduce the user’s burden.In this paper,two medical image segmentation application problems are researched through interactive method without user’s input and a small amount of user’s input.The shape and orientation prior knowledge of ulna and radius(UR)in forearm X ray DR images can be used to simplify automatic seed point detect.Therefore,we simply detected four seed points as user’s input,and the interactive UR segmentation problem is transformed into automatic UR segmentation.In studying the 3D medical image segmentation problem,an interactive segmentation method based on 2D image layer propagation is adopted.The main contributions of this dissertation include:(1)Segmentation of UR in forearm radiographs.Accurate segmentation of UR in forearm radiographs is a necessary step for single X-ray absorptiometry bone mineral density measurement and diagnosis of osteoporosis.Accurate and robust segmentation of UR is difficult given the variation in forearms between patients and the non-uniformity intensity in forearm radiographs.Inspired by the intelligent scissors algorithm use dynamic programming(DP)to search the minimum cumulative cost path as object edge.The one-dimensional dynamic programming is used to trace the minimum cost path as the target boundary on the synthesized horizontal and vertical cost maps from the positioned seed points located on the target boundary contour,along the horizontal and vertical directions.The intelligent scissors algorithm is insensitive to the non-uniformity intensity and pseudo boundary in forearm radiographs.In this work,we proposed a practical automatic UR segmentation method through DP algorithm to trace UR contours.The proposed method is quantitatively evaluated using 37 forearm radiographs with manual segmentation results.The average Dice similarity coefficient of our method reached 0.9449.The average absolute distance between the contour of the segmentation and the outline drawn by the doctor is 5.04 pixels,and the segmentation computation time of the DR forearm image with a down-sample ratio of 0.7 takes an average of 1.54 seconds.The proposed algorithm can segment the ulna and bone efficiently,accurately and robustly.(2)Interactive segmentation of 3D medical images.There are large number of 3D medical images in the radiomics’researches that need to accurately delineate the region of interest.To reduce the burden of radiologists,this paper proposed a 3D interactive segmentation method based on layer propagation.The FC-DenseNet is used to interactively segment the 2D image.The interactive input of the first layer is user input,and the interactive input of the subsequent layer is obtained by the segmentation result of the adjacent layer.The 3D interactive segmentation methods by 2D layer propagation just need user to select a 2D layer to delineate,which greatly reduced the burden of radiologists.In this work,we use our method solve the brain tumors field segmentation in 3D MRI images.