超声甲状腺结节图像分割算法研究

发布时间：2018-03-28 11:58

本文选题：超声图像　切入点：甲状腺结节　出处：《长春工业大学》2017年硕士论文

【摘要】：近年来,甲状腺结节发病率呈现出逐年增长趋势,已成为危害人类健康重要疾病之一。超声医学图像以其廉价性、无辐射、费用低等优点,被广泛应用于临床诊断中。利用图像处理技术准确分割出结节区域,给医生的诊断提供重要的辅助信息,对指导临床治疗有重要价值。由于超声甲状腺图像分辨率及对比度较低,且斑点噪声严重,不同的人群结节大小、位置及形状差异度较大,大大影响了结节分割的准确性。本文以超声甲状腺图像为研究内容,以主动轮廓水平集法作为分割基础,提出两种甲状腺结节分割算法。首先提出了基于全局与局部结合的水平集分割模型。该算法在传统局部LBF模型中加入了具有全局指向性的面积能量项,解决了LBF模型对初始轮廓敏感及容易陷入局部极小的缺点。利用LBF模型的调整项调节本文水平集的长度和光滑性,最小化能量泛函后得到演化水平集方程。通过与传统的CV模型、LBF模型进行分割对比,实验表明本文算法分割准确度更高。其次,融合模糊核聚类与改进的距离正则化水平集模型,提出了第二种分割算法,解决了距离正则化水平集模型对初始轮廓敏感、演化参数需要人工设定及分割弱边缘能力较差的问题。算法先利用模糊核聚类算法粗分割出结节区域并二值化作为水平集初始演化轮廓,然后利用初始演化轮廓围成的区域自动计算出水平集演化参数,最后采用高斯正则化规则演化水平集分割出甲状腺结节区域。与CV模型、LBF模型与DRLSE模型进行对比实验,实验结果表明本文提出的两种算法能更准确地提取甲状腺结节区域。
[Abstract]:In recent years, the incidence of thyroid nodules has been increasing year by year and has become one of the most important diseases harmful to human health. It is widely used in clinical diagnosis. Image processing technology is used to segment the nodule area accurately, which provides important auxiliary information for the doctor's diagnosis. It has important value in guiding clinical treatment. Because of low resolution and contrast of ultrasound thyroid image and serious speckle noise, the size, position and shape of nodules in different population are different. The accuracy of nodule segmentation is greatly affected. In this paper, the ultrasonic thyroid image is used as the research content, and the active contour level set method is used as the basis of segmentation. Two kinds of thyroid nodule segmentation algorithms are proposed. Firstly, a level set segmentation model based on global and local combination is proposed, which adds an area energy term with global directivity to the traditional local LBF model. The disadvantages of LBF model being sensitive to initial contour and easily falling into local minima are solved. The length and smoothness of the level set of this paper are adjusted by the adjustment term of LBF model. After minimizing the energy functional, the evolutional level set equation is obtained. Compared with the traditional CV model and LBF model, the experimental results show that the segmentation accuracy of this algorithm is higher than that of the traditional CV model and LBF model. Secondly, Combining fuzzy kernel clustering with the improved distance regularization level set model, a second segmentation algorithm is proposed, which solves the sensitivity of the distance regularization level set model to the initial contour. The evolution parameters need to be manually set and the ability of weak edge segmentation is poor. Firstly the fuzzy kernel clustering algorithm is used to roughly segment the nodule region and the binary is converted into the initial evolutionary contour of the level set. Then the evolution parameters of the level set are calculated automatically by using the region surrounded by the initial evolution contour. Finally, the thyroid nodule region is segmented by Gao Si regularization rule evolution level set, and compared with the CV model and the DRLSE model and the LBF model. The experimental results show that the two algorithms proposed in this paper can extract thyroid nodule more accurately.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R581;TP391.41

【参考文献】