中文摘要：

针对多投影图像的特点，提出一种利用同一病人多投影图像中相近位置的候选结节互信息的配准算法，由此来减少检测结果中假阳性结节的数目。通过对多投影图像中候选结节的初始检测与精确分割、特征提取与分类，完成候选结节的检测。此时，在敏感性为65％条件下，平均每张图像检测到的假阳性结节数目为11．3。再使用互信息对检测到的候选结节进行配准，并利用配准信息进一步去掉假阳性结节，平均每张图像检测到假阳性结节的数目降为I．9。即使实验数据大部分为小结节，且图像噪声大，对比度低，此检测结果仍然令人满意。因此，提出的多投影图像肺结节配准算法能有效提高结节的检测性能。

英文摘要：

Aiming at the characteristics of multi-projection images, the paper proposes a registration algorithm, which makes use of the mutual information of the nodule candidates near to each other in the multi-projection images of the same patient to reduce the number of false positive nodules. The nodule candidates were determined in the steps of initial detection as well as precise identification and segmentation, feature extraction and classification. At the sensi- tivity of 65% , 11.3 false positive nodules per image were determined in average. Then, the detected nodule candi- dates were registered using the mutual information, and the number of false positive nodules was further reduced to 1.9 per image in average. Even though the noise level in the chest radiography was high, the nodule size was small and the contrast of most nodules was low, a satisfactory performance was still achieved. Therefore, the proposed lung nodule registration algorithm can effectively improve the performance of nodule detection.