Research On Technology For Left Ventricle Automatic Segmentation In Cardiac MRI With Intensity Inhomogeneities

According to the news report from the WHO and cardiovascular disease in China,cardiovascular disease has become one of the most deadly diseases in humans,and the incidence is growing.Hence,the prevention and diagnosis of cardiovascular disease are imminent.At present,medical imaging technology has become an auxiliary method for diagnosis and treatment of heart disease.Because Cardiac Magnetic Resonance Images possesses some advantages such as noninvasive and suitable for soft tissue.etc,it has become the main means of clinical diagnosis and treatment of heart disease.Clinicians can perform qualitative and semiquantitative diagnosis by segmenting left ventricle(LV)on CMRI.However,in practice,LV myocardium is manually segmented by experienced doctors,which is subjective and time-consuming.Therefore,the automatic segmentation of left ventricle based on MRI sequences has become hot area of research in medical image processing.In this dissertation,I focus on the automatic segmentation of the left ventricle based on short axis CMR images.The following three aspects are studied,1)the extraction of region of interest(ROI),2)the automatic segmentation of the LV endocardium,and 3)the extraction of LV epicardium.The purpose of this dissertation is to design an accurate and stable LV automatic segmentation model to overcome the shortcomings of the existed LV segmentation and the segmentation methods.The main results and conclusions are in follow:1.In order to save the computing space and time,and ensure the subsequent segmentation accuracy,we mainly studied the extraction of region of interest.In this dissertation,we proposed an automatic extraction algorithm of ROI based on support vector machine(SVM).Firstly,the algorithm selects some CMR images with great difference in MICCAI 2009 different cases database.Secondly,the single CMR image is extended.Thirdly,the extended image is cutted into 9 pieces of the size of the 107×107 image.Fourthly,40 positive and 40 negative samples are selected in the segmented image blocks,and then the features of positive and negative samples are extracted to train the classifier.Finally,the trained classifier is used to automatically extract the ROI of the pre-segmented image.The results showed that the proposed algorithm can extract ROI with an accuracy of 100%.2.In order to solve the problem that the traditional level set algorithm can not segment the CMR image with intensity inhomogeneity,this dissertation proposed a new segmentation model based on improved level set and morphology.This model used an image segmentation method from the middle layer to the top layer,and to the base layer.We used different post-processing methods for the segmentation of the primary image with and without LVOT phenomenon.For the single layer image,in order to solve the problem of intensity inhomogeneity in the CMR image,the improved level set is used to remove the bias field.The specific processes of this model are: Firstly,set the initial contour in ROI.Secondly,the improved level set are used to the endometrial contour evolution.Thirdly,the location of the bleeding pool is determined based on the prior knowledge and morphological operation.Finally,according to the existence of the LVOT phenomenon,the following segmentation is performed in two cases.This method is tested in different MICCAI2009 case database.The test results showed that the algorithm has high accuracy,the segmentation of CMRI with intensity inhomogeneity is very effective,stable segmentation results and good universality.3.In order to solve the problem of the FCM algorithm can not get accurate clustering results in the original CMR image,and can not accurately segment the epicardium of LV,this dissertation proposed a clustering algorithm based on ROI with bias correction.In this method,FCM clustering is used on basis of ROI with bias correction,and three kinds of correct results are obtained: myocardium,blood pool and background.Then,the epicardium delineation was obtained based on prior knowledge and morphological operations.The results showed that the new model can be used to the segmentation of CMRI with intensity inhomogeneity based on FCM,which has high accuracy.