Reserach On Liver And Lung Tumor Detection Methods Based On CT Image

Computer-aided detection and segmentation plays an important role in the early diagnosis and treatment of vital organs,and other surgical planning and navigation.Because of its high resolution,low cost and other characteristics,CT medical imaging technology are widely used in medical diagnosis of important human organs.However,the important human glandular organ— liver,due to the low contrast,irregular shape,uneven gray,interference factors adjacent tissue,its CT imaging often lead to difficulties in computeraided segmentation and detection.And the important human respiratory organ —lung,although the high contrast of its CT imaging,the existence of vessel,bronchus,and the nearby pleural abnormalities caused great impact on the accuracy of segmentation and detection.To solve these problems,this paper is aiming at accurate segmentation of liver and lung in accompany with the tumor detection.Using high resolution spiral CT images,exploring the segmentation method of the liver,and the liver tumor detection technology;and also exploring the segmentation method of the lung,and the lung tumor detection technology.As the research purpose of this study,is to improve the accuracy of automatic segmentation and detection,and automation of these two vital organs,so as to help radiologists improve diagnosis efficiency,thus reduce risk of death from liver cancer and lung cancer.For liver segmentation process,considering the segmentation problem that each of the abdominal organs adjacent to each other,and low contrast of the imaging,we propose a probabilistic atlas and level set with the prior based hybrid automatic segmentation framework.Firstly,by calculating the similarity between the training pattern and the test image,based on the establishment of a specific patient's liver map,generating a maximum likelihood liver region;Secondly,according to the histogram analysis of the liver maximum a posterior probability regions classified by the posterior probability map,obtaining coarse segmentation of the liver,while cutting off small errors;Finally,based on the shape-gray priori model of level set evolution technology,the coarse segmentation contour is optimized,and ultimately the liver parenchyma refinement is achieved.The new method is applied to liver segmentation for the first time.Experiments show that the new algorithm has high precision of automatic segmentation,and can handle segmentation situations effectively that close to neighboring organs and liver containing cancer.The complexity of liver segmentation and diversity of liver disease,is a major obstacle to the development of automatic detection technology of liver tumor.In view of liver tumors of spherical form,this paper,on the basis of preliminary study of automatic segmentation of the liver,proposes an automatic detection method based on a variable quoit filter.Firstly,based on the level set and atlas mixing pattern,the liver segmentation contour is obtained;Secondly,by using the adaptive radius of the quoit filter,the suspicious tumor area is detected,and the gray weight based conversion function is applied in the seed area for contrast enhancement;and finally,the support vector machine based classifier is used on the characteristics of candidate tumor,to obtain the final location of the tumor and category information.In this paper,variable quoit filter is the first time applied to the detection of nearly spherical liver tumors.Experiments show that the proposed method has a high detection sensitivity for approximate spherical liver tumors of different radius sizes.As pulmonary segmentation,in consideration of the over-segmentation caused by the attached pleural tumors,in this paper,we propose a new lung segmentation technique based on boundary reconstruction and concave region correction.Firstly,by means of morphological filtering and connected component analysis technique,the chest area mask is obtained,and followed by chest extraction;Secondly,based on the diagonal boundary tracking technology,the initial lung contour segmentation is completed;Thirdly,by using the maximum cost path,the lung separation is achieved,while by applying the boundary reconstruction technology based on re-sampling technology,the jagged arc edges of the lung outline is smoothed;and finally,concave-convex judgment function is proposed to detect and repair the possible concave pleural region.By decreasing the possible oversegmentation error caused by the attached pleural tumor,the segmentation accuracy of lung parenchyma is improved.Experiments show that the proposed method is able to reduce the over-segmentation errors efficiently that caused by juxta-pleural tumors,while maintain good accuracy and complexity.Lung segmentation technology is relatively mature,which led to the rapid development of automatic lung tumors detection technology.Many scholars have put forward a variety of detection methods based on all kinds of classifier.However,on the aspects of automatic classification,performance in detection accuracy,false positive,the system speed and other cases,lung tumor detection technologies are still not high enough to be directly applied to clinical.On the basis of the preliminary research of lung segmentation,in this paper,automatic detection method for lung tumors based on an improved fuzzy clustering technique is proposed.Firstly,through a variable quoit filter,suspicious near-spherical candidate tumor is extracted,and by use of a ball-degree threshold,the approximate tubular structure is screened,in which way,the false positives is reduced to a certain extent.Secondly,with regard to the candidate lung tumor,the feature selection and feature calculation are implemented,and followed by training and testing clustering operation using improved adaptive fuzzy clustering technique.By this way,the final lung tumor classification information is obtained.The proposed method can improve the adaptability of classification parameters,and achieve the similar classification level of the mainstream algorithms with limited features training.In short,this paper is aiming at the new approach of liver and lung segmentation accompany with their tumor detection.Experiments and analysis are implemented on datasets provided by the hospital,and also some experimental method is participated in the international competition with high score.On the whole,the proposed methods achieved the expected results,and proved a good practical prospect in computer-aided segmentation and detection.