中文摘要：

针对CT图像中肺结节与血管粘连导致分割困难的问题,提出了一种基于平均密度投影和平移高斯模型的肺结节检测与分割算法。首先通过对二维CT序列图像作平均密度投影（AIP）,融合局部三维特征生成AIP图像,然后利用阈值分割和形态学方法对结节轮廓进行粗分割,最后通过建立平移高斯模型来拟合肺结节,从而实现对肺结节的精确分割。对30个血管粘连性肺结节CT图像的实验结果表明,本文算法与专业医师标记区域的面积交迭度达到91%,能够实现对粘连型肺结节的有效分割,但对于灰度较弱且体积较小的肺结节仍存在漏检的风险,需要后续进一步研究。

英文摘要：

Aiming at the problem that in CT image the blood vessels and nodules may connected that leads to segmentation difficulty, a novel nodule segmentation algorithm is proposed based on average intensity projection（AIP） and shift Gaussian model. Firstly, the AIP method is employed on a 2D CT image to generate an AIP image with local 3D features. Secondly, the region of interest（ROI） is obtained by performing threshold segmentation and morphology transformation on the AIP image. Finally, through establishing the shift Gaussian model the nodules are accurately extracted. The experiments on 30 CT images with blood vessel connection were conducted, the results show that the area overlap measure（AOM） reaches 91% between the proposed algorithm and the professional doctor manual segmentation; the proposed algorithm can achieve effect segmentation of lung nodules with blood vessel connection; however, there is still a risk of miss detection for the lung nodules with weak grey scale and small volume, which needs to be further studied.