基于解剖标志点约束的脑部PET/CT图像图谱配准与脑区分割

Brain parcellation from PET/CT images based on anatomical landmark constraint atlas registration

正电子发射断层扫描/计算机断层扫描(positron emission tomography/computed tomography,PET/CT)图像在脑部疾病的诊断和治疗中已得到广泛应用.目前的脑部图谱配准方法大多采用全局优化策略,不能精细勾画大脑的解剖感兴趣区域,尤其对于深层脑区的分割精度有限.为此提出了一种结合解剖标志点的PET/CT图像脑区分割方法,将图像灰度特征和局部解剖标志点相结合实现对局部脑区的精确配准,并融合PET和CT两种图像特征进行双模态联合配准.采用针对中国人群的脑部数字解剖图谱作为配准模板,以适应中国人群的脑区形态分布特征.通过20幅临床PET/CT头部图像对比了所提方法和传统的基于灰度的配准方法,结果表明所提方法能获得更加准确的脑区划分,各脑区相似系数(dice similarity coefficient,DSC)为79.91%,平均表面距离(average surface distance,ASD)达到PET图像的亚像素级精度0.95 mm,器官体积恢复系数(recovery coefficient,RC)接近1.

Positron emission tomography/computed tomography(PET/CT)images have been widely used in recent years with the goal of helping diagnosis and treatment in different brain diseases.The limitation of the existing brain atlas registration methods is that they mostly adopt the global optimization strategy,and the parcellation results might not be particularly accurate for the delineation of the anatomical volume of interest(VOI)of brain,especially in the deep brain region.Therefore,a brain parcellation method based on PET/CT images with anatomical landmark is presented,which combines the gray property of image with the local anatomical landmark to realize the accurate registration in local brain region,and uses image features of PET and CT to conduct dual-modality registration.In order to adapt to the distribution characteristics of brain regions in Chinese population,digital anatomical atlas of Chinese population brain is used as registration template.The proposed method is compared with the representative intensity-based registration methods in 20 clinical PET/CT images of head.The results show that the proposed method can obtain accurate delineated VOIs of brain,yielding parcellation accuracy of 79.91%in dice similarity coefficient(DSC),the average surface distance(ASD)obtains sub-pixel level accuracy of 0.95 mm in PET image and the value of recovery coefficient(RC)of organ volume is close to 1.