基于多分辨率GMI Demons算法的^(18)F-FDG PET-CT图像配准在食道癌病例中的应用

Multi-Resolution GMI Demons Algorithm Based ^(18)F-FDG PET and CT Images Registration with Application in Esophageal Cancer

18F-FDG PET和CT图像的精确配准在肿瘤的放射治疗中具有重要的临床研究意义,本研究采用全局刚性粗配准对食道癌病例中的PET和CT图像进行预处理,尽可能地减小摆位误差,然后使用基于互信息梯度的Demons算法(GMI Demons)进行局部形变配准,有效弥补内部器官误差,另外为了加快配准过程,保持图像的鲁棒性的同时避免局部极值,在形变配准前使用多分辨率图像金字塔结构。通过对10例食道癌病例的定量分析,最大互信息值结果说明经GMI Demons算法配准之后的图像精度比基于MI算法要提高8.046%±0.041%,配准前后临床上肿瘤靶区(GTV)大小的变化,说明经GMI Demons算法配准之后的GTV大小比基于MI算法配准之后的精度提高8.022%±0.044%。两种定量结果的一致性和通过对图像的定性分析,说明该配准策略可以快速地精确肿瘤靶区位置,在制定精确的放疗计划和实际的临床应用中具有研究意义。

Accurate Registration of 18F_FDG PET and CT image has important clinical significance in radiation oneology. In this paper, global rigid coarse registration was firstly used to preproeess PET and CT images with esophageal cancer, minimizing the setup margin errors. We utilized the gradient of mutual information based Demons algorithm （GMI Demons） to achieve local deformable registration, thus effectively reduced errors between internal organs. In order to speed up the registration process, maintain its robustness and avoid the local extremum, multi-resolution image pyramid structure was used before deformable registration. By quantitatively analyzing ten eases of esophageal cancer, the result of maximization of mutual information values indicate that PET and CT images accuracy after GMI Demons-based registration was improved 8. 046% ＋ 0. 041% than MI-based registration. The changes of clinical Gross Tumor Volume （GTV） indicated that the accuracy of GTVs after GMI Demons-based registration was improved 8. 022% ＋ 0. 044% than MI-based registration. The consistency of two quantitative results and by qualitative analysis indicated that the registration scheme proposed in this paper is of significance for accurately position tumor target precisely in clinical radiation therapy appliation.