Research On Intravascular Ultrasound Tissue Characterization

Intravascular ultrasound (IVUS) represents a unique imaging tool to analyze themorphological vessel structures and make decisions about plaque presence. It utilizes thetiny ultrasound probe embedded in the top of the cardiac catheterization to obtain sectionimages from the lumen of the endovascular and display the vascular cross-sectionincluding full adventitia and medial borders, namely plaque burden between the luminaland intima.Currently, plaques classification in the IVUS images is mainly done by manual, notonly the huge workload, but the results of strong subjectivity, poor repeatability,vulnerable to the influence of clinical experience and professional knowledge of theoperator. Therefore, make use of digital image processing technology to detect andclassify the plaque tissue in IVUS images automatically has important clinical value.As IVUS image contains a large number of texture information and the texturedifferences between normal tissue and lesions are significant, texture information can beused as an important basis for IVUS image characterization. There are three classicimage texture feature extraction methods to extract the texture features of IVUS image inthe paper. In GLCM (gray level co-occurrence matrix), the two order statistics of wholeimage are calculated to describe the image texture; LBP (local binary pattern) is agrayscale texture measurement; Gabor filter is a kind of the multi-resolution analysistools and describe the local details of image texture effectively. After feature extraction,the paper use principal component analysis, Fisher linear discriminate analysis and MDSto reduce the dimension of feature data, and design SVM (support vector machine),Adaboost and random forest classifier to classify image feature data, then optimizeclassifier parameters to improve the classification accuracy. At last, choose a moreappropriate classifier for characterization and complete the classification of the differenttissues in IVUS images. Using clinical images data to validate the feasibility of themethods, then compare and discuss the advantages and disadvantages of each method.