Research On Several Techniques Of Intravascular Ultrasound Image Processing And Analysis

Intravascular ultrasound (IVUS) imaging provides direct coronary anatomy and allows precise assessment of human coronary pathology in vivo, so it is becoming increasingly accepted as an imaging modality in diagnosis and treatment of coronary heart disease (CHD).However, predecessors as to the quantitative analysis of IVUS images got few of achievements because of low signal-noise-ratio and complex features of IVUS images. There are many technical difficulties such as blood speckle denoising, contrast enhancement, contour extraction and automated classification of plaques. IVUS images processing and the measurement for vessel parameters are important to the clinical application and scientific research. To meet the requirement of clinical application, several key IVUS images processing techniques such as denoising, enhancement, contours extracting and texture classification are studied in this dissertation. The main works are as follows:(1)After analyzing the property of blood speckle noise, logarithm transform is performed and then a threshold denoising technique based on wavelet transform is applied to process intravascular ultrasound images to discard blood speckle noise in IVUS images. Since threshold denoising algorithms based on wavelet transform need to estimate noise variance, a noise variance estimation method is proposed according to characters of IVUS images, then different thresholds selection ways for discrete wavelet and dyadic wavelet are studied, afterwards, the virtues and shortcomings of this two types wavelet denoising algorithms are discussed. Experimental data shows that denoising technique based on wavelet transform can achieve signal-noise-ratio and keep boundaries well.(2)By connecting contrast of IVUS images with gradient of multiscale boundary of images, a contrast enhancement algorithm based on multiscale boundary representation of images and Hermite polynomial interpolation is proposed. Experimental results indicate that this algorithm can resolve the conflict between amplifying noise with enhancing weak information existed in available technique.(3)For contours extracting, an algorithm which combined border detection based on dyadic wavelet transform and fast active contour model is advanced. To achieve proper initial contour which is needed in fast active model, an initial contour extraction method based on dyadic wavelet transform is applied. The initial contour obtained by this method is close to the object contour, which greatly speeds up deform velocity and can obtain a proper object contour. Experimental results show that the contours collected by the algorithm are near the ideal contours and the tracing error is smaller than inter-observer error in most cases.(4)A resample technique based on emulating ECG signal is promoted when extracting contours of serial IVUS images. According to the relationship between heart motion and image distance, the cardiac cycle length of each frame is calculated. A set of frames from the pullback sequence are picked at the point in the cycle when the heart is maximally motionless. This resample technique can remove motion artifacts and improve accuracy when tracing contours in pullback sequences. Experiments demonstrate that images in new image set obtained by the resample technique have higher correlation and better stability.(5)A classifying algorithm combining textural feature selecting and support vector machine is studied. A textural characteristic vector is composed by several characteristic values defined by statistical methods and support vector machine is used as a classifier when the plaque tissues are automatically identified. Experimental data indicate that the algorithm is effective and the average recognition rate of plaques is 93.3%.The research project of this paper is an intersectant research area, which is based on the development of rapid information discipline and biomedical technology. Research results solved the key problems in the IVUS images processing, realized qualitative and quantitative analyses and improved precision and efficiency of clinical diagnosis.