基于ASM和GrabCut的DR图像肺实质分割

Lung Field Segmentation Based on ASM and GrabCut on DR Image

针对DR图像肺实质分割中分割精度与自动分割难以两全的情况,提出一种由粗到精的肺实质分割方法。该方法增强DR图像边界和组织的对比度;使用训练好的ASM模型匹配得到粗略肺部轮廓;使用模板匹配检测出左右肋膈角、左右心膈角和左右肺尖的实际位置,并根据该实际位置线性调整肺部轮廓;根据调整后的肺部轮廓作为Grab Cut的目标输入,迭代分割,得到肺实质精确分割结果。通过对100张DR图像进行测试,分割重叠率达到97.8%,平均边界距离低至1.68。

In order to improve accuracy and ensure automatic segmentation at the same time on DR image, presents a novel segmentation scheme for lung field segmentation. Improves the contrast at the lung boundary and on tissue surface. Trains active shape model to fit the lung con- tour to output coarse lung contour. Then obtains actual landmarks of apex pulmonis, costophrenic angle cardiodiaphragmatic angle ac- cording to template matching, and adjust lung contour with these actual landmarks. Based on the adjusted lung mask, proposes GrabCut to iterate segment lung field accurately. Applies the proposed approach for the analysis of 100 DRs, and the results show that be able to get overlap greater than 97.8% and mean absolute contour distance less than 1.65, and the approach provides a reliable lung segmenta- tion scheme on DR images.