Design And Implementation Of Automated Retrieval System For IVUS Image Sequences

Intravascular ultrasound(IVUS), an interventional imaging modality, combines a non-invasive ultrasound technique and invasive catheter-based technique. A catheter with an ultrasonic probe mounted on its tip is directly inserted into the vascular lumen under the direction of X-ray angiograms. Then, the catheter is slowly pulled back during with a series of tomographic images of the vessel is acquired. IVUS imaging has been widely used in the clinical diagnosis of coronary artery diseases.Clinically acquired IVUS image sequences are heavily polluted by noises and several kinds of artifacts. There are a lot of frames without diagnostic value in the tomographic image sequence due to the fast pullback of the catheter. It is a tedious work to check and analyze the images manually. The results are not repeatable enough and highly depend on the clinical experiences and professional knowledge of the operator.This dissertation includes two parts of automated retrieval of key frames and detection of stents and vascular bifurcation from routinely acquired IVUS gray-scale image sequences. Regarding the first issue, two methods are presented. One is based on the vascular morphological description. The vessel wall contours are firstly extracted and feature vectors including curvature characteristics of profiles were then constructed. The key frames based on the adaptive thresholding algorithm are finally detected according the Mahalanobis distances between successive frames. The other method is based on image regional gray-scale features. Those key frames are detected through comparing the Bhattacharyya distances between corresponding regional gray level histograms of successive frames. Regarding the second issue, Haar-like and LBP are used to extract texture features of IVUS images. Then, Gentle Adaboost, Modest Adaboost and Real Adaboost classifiers are employed to classify the texture features in order to detect the presence of stents and bifurcation. The validity of the methods has been demonstrated with clinically acquired image data. The accuracy has been quantitatively evaluated according to the experimental results.