Superpixel Based Random Walk Method For Efficient Liver Organ Image Segmentation And Its Application

Liver segmentation is one of the most fundamental tasks in computer aided diagnosis(CAD)system for lung diseases.The precise segmentation of medical organ CT images have always been a challenging task.On one hand,the ever-changing shape and large amount of count of CT image increases the difficulty of segmentation accurately.On the other hand,the existing symptoms of a patient(such as a tumor)can greatly interfere with the anatomy of the organ.In order to divide the abdominal CT image into different tissues,many different methods have been developed by the industry.In recent years,the development trend of CT image segmentation technology can be summarized as follows:(1)Interactive methods such as Graph Cut and Random Walk algorithm;(2)Automatic methods such as probability maps,statistical shape models,artificial neural networks.In this paper,our research focuses on the improvement and application of the classical method of organ segmentation.The main contents of this paper include the following:Propose a more efficient random walk(RW)algorithm.The traditional three-dimensional segmentation method based on random walk mainly has the following problems:the number of three-dimensional images is large,it takes a lot of manual marking time,and computational burden is heavy,the segmentation result is not accurate enough.This paper propose a random walk algorithm based on prior knowledge of slice-by-slice partitioning,which greatly reduces the time cost of manual marking,but the algorithm still has the problem of low segmenting efficiency.In this paper,based on the existing three-dimensional pixel-based random walk study,we propose a super-pixel-based random walk algorithm,and compared with the traditional random walk algorithm by comparative experiments,the results show that the proposed algorithm is more efficiently.When segmenting 20 groups of 3D liver images,the average DSC coefficients and processing time of the proposed method are 0.956 and 18.28 seconds,respectively.Compared with the traditional random walk algorithm of 0.901 and 928.12 seconds respectively.Design and implementation of a multi-phase medical liver CT image retrieval system.In clinical CT liver images according to different contrast agent injection time,CT images can be divided into different phases.Because of the different tumor types,different injection times of contrast agent,tumor images of individual phases are not clear enough to be cut directly,we propose a registration method combining rigid body and non-rigid body transformation,which can be registered with clear liver and tumor Obtaining unclear tumor,combined with the previous proposed watershed super-pixel-based random walk algorithm,designed and implemented multiphase medical CT image retrieval system.The method of this article is mainly used in the system pre-processing module that is segmentation and registration module.The system has established a case image database,which can be used to assist doctors in clinical diagnosis of cases.In this system,the physician can quickly and automatically segment the liver and tumor regions of the three phases simply by manually marking several key areas of the image.And then through the search method of the predecessor,retrieves the case information similar to the new case,in order to assist the doctor to complete the diagnostic report.