基于掩膜优化的多模态医学图像刚性配准

Rigid registration of multimodal medical images based on mask optimization

目的:基于配准开源平台ITK和开源计算机显示视觉库OpenCV开发刚性配准程序,并集成到DeepPlan计划系统中,实现快速准确的多模态刚性配准。方法:基于形态学开运算初步去除图像中无需关注的细小区域和部分扫描床,使用最大类间方差法(Otsu)突出感兴趣的图像部位,Canny算子用于提取富含信息区域的边界信息。使用像素填充技术得到图像配准需要的掩膜,并采用OpenMP并行技术加速掩膜计算过程。最终在配准过程中将掩膜作用于参考图像或浮动图像。结果:测试了多组不同模态和部位的算例,实验结果表明基于掩膜优化的多模态医学图像刚性配准方法可以自动去除绝大部分背景图像和扫描床板,节约图像配准中一半以上时间,且图像配准质量并无下降;在1 min内可以完成两组100张左右的图像配准。且本方法以动态链接库的形式成功集成在治疗计划系统DeepPlan中。结论:在保证配准结果准确的基础上,基于掩膜优化的多模态医学图像刚性配准方法显著提高了图像配准速度,且算法稳定性能高,有很好的临床应用前景。

Objective To develop a rigid registration program based on ITK and OpenCV and make it compatible with DeepPlan treatment planning system for achieving rapid and precise multimodality image registration.Methods Mathematical morphology analysis was used to erasure insignificant and tiny structures and partial scanning couch;Otsu filter was applied to further highlight the regions of interest;and Canny operators were utilized to extract edge information from regions rich in information.Then the mask required in image registration was obtained by pixel filling algorithm,and the pre-processing process was accelerated with OpenMP parallel strategy.Finally,the mask acted on the fixed image and moving image during registration.Results The proposed method was validated on clinical images datasets with different modalities and body parts,and the results revealed that the most interferential background image and scanning couch were removed by rigid registration of multimodal medical image based on mask optimization,saving more than more than half of the time while without degrading the image registration quality.The registration of two groups of about 100 images could be finished within 1 minute.The relative codes were also successfully integrated in DeepPlan treatment planning system in the form of dynamic-link library.Conclusion Rigid registration of multimodal medical image based on mask optimization improves the efficiency of image registration,without affecting registration accuracy,and the proposed algorithm is highly stable.Therefore,the proposed method has a good prospect in clinical application.