欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 医学图像与设备 • 上一篇    下一篇

基于非下采样轮廓波变换和压缩感知的PET/CT 像素级融合算法

  

  1. 1. 宁夏医科大学理学院，宁夏 银川 750004；2. 宁夏医科大学公共卫生与管理学院，宁夏 银川 750004；
    3. 西北工业大学计算机学院，陕西 西安 710100；4. 宁夏医科大学总医院泌尿外科，宁夏 银川 750004
  • 出版日期:2017-12-30 发布日期:2018-01-11
  • 基金资助:
    国家自然科学基金项目(81160183，61561040)；宁夏自然科学基金项目(NZ16067)；宁夏高教项目(NGY2016084)

PET/CT Image Pixel-Level Fusion Algorithm Based on NSCT and Compressed Sensing

  1. 1. School of Science, Ningxia Medical University, Yinchuan Ningxia 750004, China;
    2. School of Public Health and Management, Ningxia Medical University, Yinchuan Ningxia 750004, China;
    3. School of Computer Science, Northwestern Polytechnical University, Xi’an Shaanxi 710100, China;
    4. Urinary Surgey, General Hospital of Ningxia Medical University, Yinchuan Ningxia 750004, China
  • Online:2017-12-30 Published:2018-01-11

PDF (PC)

160

摘要： Piella 多分辨率图像融合框架包括4 种融合规则的构造方法，围绕该框架下的第1
种融合规则提出了基于非下采样轮廓波变换(NSCT)和压缩感知的PET/CT 像素级融合算法。首
先，对已配准的PET 和CT 源图像进行NSCT 变换；然后，为突出低频图像的病灶部位，采用
对特征和区域敏感性更高的脉冲耦合神经网络融合规则；其次，选择高斯随机矩阵对高频子带
进行压缩采样得到测量值，基于Piella 融合框架以高频子带分块计算的直方图距离作为匹配测
度，以高频子带的区域能量作为活性测度，用匹配测度和活性测度构造出自适应的决策模型计
算融合因子d，根据融合因子对高频测量值进行融合，再利用正交匹配追踪算法重构出高频融
合图像；再次，对融合后的低频图像和重构后的高频图像同时进行NSCT 逆变换得到最终的融
合图像；最后，进行了融合算法比较、活性测度比较和匹配测度比较的实验。实验结果表明该
算法可以更好地呈现病灶信息，从主观效果和客观评价指标均验证了其有效性。

关键词: 非下采样轮廓波变换, 脉冲耦合神经网, 压缩感知, Piella 框架

Abstract: There are four construction methods of fusion rules in Piella framework, according to the
first method, a pixel-level fusion algorithm of PET/CT based on non-subsampled contourlet transform
(NSCT) and compressed sensing is proposed. Firstly, NSCT is used to decompose the registered PET
and CT image. Secondly, in order to highlight the focal region of the low frequency image, the low
frequency components are fused by pulse coupled neural network (PCNN) which more high
sensitivity to the characteristics and regions. Thirdly, in high frequency, the Gauss random matrix is used
for compression measurement, in sub-block of high frequency, histogram distance is utilized in match measure, the regional energy is regarded as the activity measure. The fusion factor ‘d’ is calculated by
adaptive decision model using matching measure and activity measure. According to the fusion factor, the
high frequency measurement values are fused, and the high frequency fusion image are reconstructed by
using the orthogonal matching pursuit algorithm. Fourthly, the final fusing images are acquired through
the NSCT inverse transformation of low frequency fusion image and high frequency fusion image. Finally,
three experiments are done, comparison experiment with other image fusion algorithms, comparison with
different activity measures and different matching measures. The experimental results shown that the
algorithm can retain and show the edge and texture information of focus better. And the algorithm are
superior to other fusion algorithms from subjective and objective evaluation.

Key words: non-subsampled contourlet transform, pulse coupled neural network, compressed sensing, framework of Piella