Hard Plaque Characteristics Based Media-adventitia Border Detection In Intravascular Ultrasound Images

Coronary artery disease (CAD) is one of the main diseases threatening human beings. In recent years, CAD’s morbidity is increasing and the number of people who died from CAD is rising. In our country, with the improvement of living standards and the increased number of elder citizens, CAD has become the leading cause of death, and more and more young people suffer this disease. Therefore, the prevention and treatment of CAD get extensive attention, and have become the most important task for researchers.Nowadays, Coronary angiography (CAG) and Intravascular ultrasound (IVUS) are the most widely used technologies in clinical diagnosis and treatment. CAG is regarded as a reliable imaging modality and the "gold standard" for assessment of CAD. CAG could show the position and the severity of coronary narrow by displaying the contour of coronary lumen. But CAG cannot provide the detailed information of vascular wall, such as the composition of atherosclerosis plaques. Because of the limitation of evaluating plaque, it is difficult to diagnose some special pathological changes of coronary.Intravascular ultrasound (IVUS) is a new technique of CAD diagnosis, which place the ultrasound probe into vascular and scans the vascular by360degree with high frequency ultrasound. IVUS not only shows the lumen of vascular, but also show the information of vascular wall, especially the atherosclerosis plaques. This information could assist clinicians in diagnosing the disease and designing the therapy program. With the rapid development of IVUS, the quality of IVUS image has been significantly improved, and more and more novel IVUS-based technologies are now available to clinical application. IVUS has become the indispensable coronary imaging modality.Due to the huge quantity of IVUS images, manual processing is often tedious and time-consuming. And the correction of manual processing results is especially difficult to guarantee. The computer-aided image processing technology is an effective way to process IVUS images, and is successfully used in many fields of IVUS images, such as border detection, plaque recognition, registration,3D reconstruction, etc. It is playing a significant role in the advancement of CAD therapeutic level and clinician working efficiency.The main purpose of this paper is to design a media-adventitia border detection algorithm which could detect the media-adventitia border accurately and rapidly. Media-adventitia is the basis of CAD diagnosis and IVUS image processing. There is a problem of current media-adventitia border detection approaches that the existence of hard plaques in IVUS images could deteriorate the results of media-adventitia border. A novel hard plaque characteristics based media-adventitia border detection approach is presented in this paper to overcome this problem. The IVUS image is first enhanced by a multiscale based adaptive bilateral filter and converted from Cartesian coordinate into Polar coordinate. Then, the hard plaques are recognized based on the intensity variation within different regions that were obtained using the κ-means algorithm. In the next step, a cost matrix representing the probability of the media-adventitia border is generated by combining image gradient, plaque location and image intensity. A heuristic graph-searching is applied to find the media-adventitia border from the cost matrix. In this approach, there are three key algorithms:(1) image denoising algorithm,(2) hard plaque recognition algorithm,(3) media-adventitia border detection algorithm.(1) In high frequency IVUS images, speckle is a primary factor that degrades the contrast resolution and masks valuable texture information. Therefore we present a novel denoising method to solve such problem. We design the multiscale based adaptive bilateral filter by adding the characteristic of multiplicative noise into bilateral filter, and researching the change of scale-space from perspective of multiscale theory. This filter could eliminate the speckle and keep the important texture information.(2) The atherosclerotic plaque information is of great significance for CAD diagnosis and therapy program designing, and the hard plaque could affect the searching of media-adventitia border. Therefore it is necessary to recognize hard plaque. The κ-means algorithm is adopted to segment the image and the judgment formulas are presented according to the intensity variation within different kinds of plaque region. This algorithm could recognize the type, position and size of hard plaques.(3) The heuristic graph-searching algorithm is adopted to detect the media-adventitia border. The key technology for this method is to design a suitable cost function. The cost for every pixel is calculated by this function, and the cost represents the probability of a pixel as a border point. We define a cost function by three sections:(1) image gradient;(2) enhanced cost of hard plaque rear region;(3) eliminated cost of hard plaque region. This cost matrix contains enough border information and could detect the correct media-adventitia border.The dataset for testing and validation consists of654frames of IVUS image which were provided by the department of radiology, Leiden University Medical Center, the Netherlands. The automated detected results were quantitative analyzed and reviewed by clinicians. The total hard plaque recognition correct rate is89.94%. Among them, the correct rate of images with calcified plaque recognition is92.16%; the correct rate of images with fibrous plaque recognition is91.55%; the correct rate of images with mixed plaque recognition is71.43%. The result of media-adventitia border detection approach presented by this paper is very close to the result of doctor-drawn. The total media-adventitia border detection correct rate is95.57%. Among them, the correct rate of images without hard plaque is88.32%; the correct rate of images with hard plaque is97.49%. The results of this paper approach are obviously advanced when compare with other two approaches:GVF-snake and traditional graph-searching. Finally, we design experiments to prove the necessity and effectiveness of every part of the cost matrix.In summary, this paper presents denoising, hard plaque recognition and media-adventitia border detection algorithm for IVUS images, and the experiment results of huge number of clinical IVUS images certificate the accuracy and effectiveness of this paper approaches.