Research On Key Technologies For Computer Aid Diagnosis Of Cervical Lymph Nodes Ultrasound Images

Lymph node is one of the important human immune system organs. The evaluation of cervical lymph nodes is meaningful for the disease diagnosis. For cervical lymph nodes are the cores of the human lymph immune system. Medical ultrasound is the preferred technique for the evaluation of cervical lymph nodes, due to its irreplaceable advantages:safe, valid, portable and cost effective. However, the son graphic examination of cervical nodes by radiologist is very subjective. For the image quality of ultrasound is low with its imaging mechanism. Therefore, how to obtain high quality ultrasound images, accurate contours of cervical lymph nodes and quantified son graphic features for the radiologists to diagnose the cervical lymph nodes more objectively have been urgent to be solved. This paper mainly research some related aspects of computer aided diagnosis including ultrasound image denoising, segmentation and features selection. Detail works are listed as below:1. A new adaptive ultrasound images speckle reduction model based on texture structure is proposed to improve TV (Total Variation, TV) model. First, the speckle's characteristic of ultrasound images is described by textural information in this new model. The value of homogeneity is defined by texture in the ultrasound images, and the gray-scale domain of ultrasound images is mapped to the homogeneity domain. Second, the threshold is obtained by two-dimensional histogram of homogeneity, which is used to partition different pixel set of homogeneity set and not homogeneity set. Finally, the different model with different norm is chosen by different set. This model removes the speckle noise effectively while keeping lots of details information of original images.2. Generally, cluster ensemble for image segmentation hold two drawbacks:high time complexity and local optimum. Therefore, a novel spectral cluster ensemble image segmentation algorithm is proposed. The computation complexity is low in improved spectral cluster with avoiding the eigenvalue decomposition problem of large scaled matrixes by solving the eigenvalue decomposition of small matrixes. The algorithm gets the global optimum by employing the improved spectral cluster to integrate the cell clusters with lower computation complexity.3. An improved active contour ultrasound image segmentation algorithm based on relevance vector machine energy representation. The relevance vector machine has the advantages of supervised learning classification and the global region distribution information can be exploited to enhance the performance. A new region based image energy term is made by the output of relevance vector machine. The algorithm is more robust than other active contour for it takes into account the image segmentation knowledge of human. The relevance vector machine is also employed to obtain initial contour firstly for improving the segmentation speed.4. In most of discrete differential evolution (DDE) algorithm, iterated greedy algorithm is employed for local search, which is time-consuming and easy to lead to a premature convergence. In this paper, a novel discrete differential evolution (DDE) algorithm with virus-evolutionary is presented to solve the cervical lymph nodes features selection problem and named as virus-evolutionary discrete differential evolution (VEDDE) algorithm. Biological virus mechanism and the infection-based operation between host and virus are introduced in the DDE. In the co-evolutionary process, the virus propagates partial genetic information in the DDE by virus infection operators which enhance the individual diversity and local search capability in solution space.