文章摘要

文章信息/Info

Title:: CT image feature association location based on deep cooperative representation

Author(s):: WANG Baojing; QIN Quanbo; MAO Yisheng; KONG Ling; DING Danhui; ZHANG Guofu; Department of Magnetic Imaging, Xuchang ,Xuchang 461000, China

Keywords:: Deep cooperative expression; CT images; Characteristic; Similarity; Associated; Positioning

摘要:: 单层协作表达算法重建CT图像超分辨率效果较差，不能创造CT图像关联特征定位的有利条件，因此，采用深度协作表达算法实现CT图像特征关联定位。该算法逐层迭代更新初始化字典，更新对应每层的最优表达权重系数，将最后表达层全部重建图像块合为高分辨率CT图像。采用距离法计算特征相似性，特征间欧氏距离越小、相似性越大，与选定特征距离最小的特征为关联特征，采用Sobel算子模板构建活动轮廓线套索融合模型，提取CT图像边缘点灰度值与关联特征点灰度值求取灰度势能差异性特征，完成CT图像特征关联定位。研究表明，定位算法用于肺部CT图像诊断时错误率小、定位病灶位置精准，可在医疗诊断中推广使用。

Abstract:: Single-layer cooperative expression algorithm has poor super-resolution effect in reconstructing CT image, and can not create favorable conditions for CT image correlation feature location. Therefore, deep cooperative expression algorithm was used to realize CT image feature correlation location. The algorithm iteratively updated the initialization dictionary layer by layer, updated the optimal expression weight coefficients corresponding to each layer, and combined all reconstructed image blocks of the final expression layer into high-resolution CT images. The distance method was used to calculate feature similarity, the smaller the Euclidean distance between features was, the greater the similarity was，and the smallest feature distance was selected from the feature as the correlation feature. Sobel operator template was used to construct the active contour lasso fusion model, extract the gray value of edge points and the gray value of correlation feature points to obtain the difference of gray potential energy, and complete the correlation location of CT image features. The research shows that the localization algorithm can be widely used in medical diagnosis because of its low error rate and precise location in pulmonary CT image diagnosis.