Algorithm Analysis Of Attribute Vector-based Image Registration Method And Its Application In Medical Ultrasound Images

Medical image registration is an important technique in the field of medical image processing, and is becoming more and more important for clinical diagnosis and treatment. Modern medical research requires integrated analysis of multiple images to get more information and improve the effectiveness and accuracy of clinical diagnosis. Medical image registration is recovering the geometric relationship between corresponding points in multiple images of the same scene. The results make all anatomical points, or at least the points which have diagnostic significance matched. Medical ultrasound imaging has been widely used as its unique real-time, good repeatability, high sensitivity and ease of use. It has an enormous potential in the quantitative analysis of the organs and internal tissues of the body, real-time monitoring and treatment planning. However the existence of speckle noises reduces the quality of images and increases the difficulty of image analysis. How to improve the registration speed and accuracy on the premise of decreasing influence of speckle noise has strong practical significance and research value.This thesis briefly introduces relevant knowledge of image registration and reviews the current methods firstly. After comparing the advantages and disadvantages of similarity-based and feature-based registration methods, this dissertation focuses on the algorithm based on attribute vectors.The main work of this dissertation includes:(1) The flow and frame of the algorithm based on attribute vector is implemented and different features applying to medical image registration of different organs and different modality are selected, such as MR brain images, DTI brain images and CT pelvis images.(2) Due to the existence and corruption of speckle noises, the most salient parameter features in ultrasound images can be extracted based on the statistical modeling of speckle noises and the basic knowledge of information theory. This method has a strong feature response and is much less distracted by the speckle compared with the LoG and Canny operator.(3) Aiming to the shortage of modified HAMMER, an attribute vector is calculated as the intensity, the magnitudes of the gradient, the second order derivative and parameter feature which is extracted based on speckle noise models at each pixel in an ultrasound image. With mutual information as the evaluation criteria, different feature combinations are experimented to find the optimal one.According to the effectiveness and importance of different characteristics, mass of experimental data tell us the best ultrasound image registration result can be obtained by choosing the feature combination of intensity, second order derivative and parameter feature. The results show that the method cuts down the number of leading points effectively then reduces the time of image registration and raises the registration precision.The new method for ultrasound image registration is robust and flexible and is expected to be applied to other new medical imaging technology, such as 3D ultrasound and ultrasound CT. This idea can be extended to multi-modal medical image registration between ultrasound image, CT images and MRI images by researching on image property of different modalities and seeking the appropriate attributes based on some optimization strategy or clinical recommendations of doctors, which can monitor intra-operative changes of anatomic regions in a real-time manner and easily and do not give rise to additional damage and pain to patients.