Research On Image Process Key Technologies For Intravascular Ultrasound Images

Intravascular ultrasound (IVUS) imaging technology has been developed for years and gradually been widely applied to diagnosis and therapy for cardiovascular disease. By computerized image process and analysis for IVUS images, it can be a powerful assistant when doctors diagnose the state of an illness and make surgical plan. It also provides accurate pathological support for pre-surgery and aft surgery as well. Under the demands of cardiovascular clinic, relevant theories and key technologies of IVUS image sequences processing and analyzing are studied deeply in this dissertation. And the main contents and innovative achievements are as follows:1. To get IVUS image sequences that have the same spatio-temporal attribute from raw data, a new method of automated phasic registration and resampling is proposed. It is practical and can be done without any hardware. The heart cycle is resumed according to the different physical characteristics between systole and diastole period, of which the image data at the same phase is resampled to make up new volume data. And it will be the powerful guarantee for subsequent image process.2. To remarkably decrease the blood speckle noise and increase the contrast between arterial wall and lumen, a new noise reduction algorithm is introduced. It is based on that blood echo speckles have higher spatio-temporal variation than the arterial wall. The energy spectrum of organism and blood is firstly calculated by two- dimensional Fourier transform algorithm. Afterwards a scale parameter defined by the radio of high frequency and low frequency is used to determine whether the signal is to be organism or blood speckle. Then those organism signals are reserve and speckles are discarded.3. To overcome time-consuming and the subjective difference by manual segmentation, an algorithm of automatic contour detection based on fast active contour model is promoted to detect luminal border and medial-adventitial border. In the experiment, many improved techniques are introduced to the model. We introduce the growth period of Snake curve, replace gray gradient by edge contrast gradient, design two methods for contour initialization and so on. In order to correct the local error introduced by two-dimensional algorithm, a three dimensional contour detection is designed by extending the fast active contour to three dimensions. Experimental results show that the new algorithms not only have good reliability and veracity, but also save much time than manual detection. Besides, three-dimensional method is much better in detection effect and efficiency when tracing multi-frame image sequences.4. To classify and identify the plaques tissues accurately, a technique of combination of Co-occurrence Matrix and fractal dimension textural methods is proposed. Firstly, a textural characteristic space vector is composed of four characteristic values defined from Co-occurrence Matrix and three from fractal dimension. Secondly, the vector is projected to two new characteristic spaces respectively by principal component analysis and Fisher linear discriminant analysis methods. Finally, the classification is completed in those new vector spaces. Experimental results indicate that it is more effective and accurate than other methods.5. To display the three-dimensional structure of vessel and do accurate volume measurement, a theory of three-dimensional modeling for IVUS image sequences is established. By combination of biplane angiography and intravascular ultrasound techniques, theoretic solutions are given to those problems on detection of vessel centerline, computation of relative orientation for adjacent images and torsional orientation of image itself.