中文摘要：

目的通过基于四维CT（4DCT）图像改进的demons算法和基于B样条的FFD算法比较研究，探讨呼吸运动对原发性肝癌肝脏及靶区形态的影响。方法对8例原发性肝癌患者行4DCT模拟定位，并将4DCT图像依据呼吸曲线均分为10个呼吸时相。采用改进的demons算法和基于B样条的FFD算法配准10个时相的图像。从归一化互信息、Hausdorff距离以及配准时间等方面比较两种方法差异。结果改进的demons算法较基于B样条的FFD算法的归一化互信息平均提高了4．75％（P=0．002），Hausdorff距离平均降低了15．2％（P=0．020），配准时间上也具有显著优势（P=0．036）。结论呼吸运动造成的正常肝脏及肿瘤靶区位置和形态变化是非常显著的，两种方法均可实现4DCT图像间的形变配准，改进的demons算法在配准精度和速度方面更具优势。

英文摘要：

Objective To study the morphology of normal liver and tumors by breathing motion of hepatocellular carcinoma patients, through comparing the modified demons algorithm and FFD algorithm based on B-spline, and combing four-dimensional computed tomography （4DCT）. Methods The 4DCT images of 8 HCC patients were segmented into 10-series which were named CT0, CT10... CT80, CT90 according to the respiratory phases, CT0 and CT50 are defined to be end-inhale and end-exhale respectively. CT50 was chosen as the reference image. We used the modified demons algorithm and FFD algorithm based on B-spline to deform the images. Linear interpolation was used in both mode 1 and mode 2. The normalized mutual information （NMI） , Haasdorff distance （dH） and registration speed were used to verify the registration performance. Results The average NMI for the end-inhale and end-exhale images of 8 HCC patients after demons registration in mode 1 improved 4. 75% with FFD algorithm based on B-spline（P =0. 002）. And the difference of dH after demons reduced 15.2% comparing with FFD model algorithm （ P = 0. 02）. In addition, demons algorithm has the absolute advantage in registration speed（P = 0. 036 ）. Conclusions The breathing movement for deformation of normal liver and tumor targets is significant. These two algorithms can achieve the registration of 4DCT images and the modified demons registration can deform 4DCT images effectively.