Research On Despeckling And Segmentation Of Left Ventricular Echocardiography

Cardiovascular diseases have been a serious disaster to human health. It is also oneof those leading causes of the death in industrialized countries. With the medicalimaging technology, doctors can analyze the functional changes of the heartquantitatively and qualitatively, so as to establish its relation with cardiovasculardiseases and auxiliary clinical treatment. Among all the imaging technologies, medicalultrasound imaging is widely used in diagnosis and treatment due to its characteristic ofharmless, real-time, low cost, and easy repetition, etc.2D echocardiography displaysreal-time images of human heart, and it can be used to observe the form and motion ofheart. Since visual diagnosis will inevitably have a doctor’s subjectivity, some cardiacfunction indexes are usually needed in clinical for objective diagnosis. The base ofcomputing the index is to obtain the contours of human heart in different time of cardiaccycle which can be solved by image segmentation. Unfortunately, the quality andresolution of ultrasound images are badly affected by its intrinsic noise known asspeckle. Its existence causes trouble for segmentation. A better version of an ultrasoundimage for segmentation can be obtained by despeckling. Therefore, the study ofultrasound image despeckling and segmentation has great significance for thedevelopment of medical ultrasound imaging.Taking the practical application of the medical diagnosis into consideration,despeckling and segmentation algorithms of2D ultrasound echocardiography arediscussed in this thesis. The research includes the following four parts:First of all, according to the summary of theory of ultrasound imaging, thecharacteristic of speckle is discussed. Moreover, two methods are used to simulateultrasound images.Then a time series non-local means speckle filter is proposed to reduce the noise inultrasound images. The traditional non-local means filter is improved from thefollowing aspects. A distance measure relevant to speckle model is introduced to takeplace of the Gaussian weighted Euclidian distance according to a Bayesian formulation.By taking the information along the temporal axis into account, the non-local meansfilter is extended from single frame to image time series. To lighten the computationalburden, a blockwise approach and a pre-classification process are used to accelerate the algorithm.Then the level set method is used for segmentation of echocardiography.Considering the high cost of re-initialization, the variational level set method is used,and the energy function is designed to make the zero level set shrink or expand atdifferent locations so that the endocardium of left ventricular. Then an area constraintenergy function is designed based on the incompressible of myocardium to overcomethe problem of the segmentation of involucurm for its indistinct contour.At last, the segmentation results of2D echocardiography from an entire cardiaccycle are used for the calculation of indicators of left ventricular function. The resultsare compared with the actual data from M-mode echocardiography.