摘要
目的胸部CT图像的肺实质自动分割是肺部疾病计算机辅助检测的重要基础。为提高分割速度,本文提出并实现了一种基于重采样的分割算法。方法首先对数据重采样,提取部分(1/8)体数据。再基于重采样体数据,通过阈值分割、胸腔提取、气管剔除、血管填充、左右肺分离和肺壁结节填充等步骤,得到初步分割结果。然后将该结果还原到完整数据体上,形态学平滑后即完成最终分割。最后将算法应用于20例患者数据(2556个断层),并与放射科医生手动分割结果进行比较。结果本文算法对20例患者数据均能取得优异结果,与放射科医生手动分割的平均面积重叠率达99.02%,且适用于左右肺相连、肺壁存在结节、视野不完整等异常情况。通过数据重采样极大缩短分割时间,一般可缩短50%,一帧图像平均耗时小于0.25s。结论本文算法能够实现胸部CT图像肺实质的自动分割,结果准确可靠,鲁棒性好,速度快,基本满足实际临床需求。
Objective Automatic lung parenchyma segmentation is one of the most important steps in the computer aided diagnosis （CAD） of the lung. To increase segmentation speed, an algorithm based on resampling of the image data is proposed and implemented. Methods The algorithm firstly resamples and extracts a small part （1/8） of the original CT images data. Several steps are implemented to get preliminary segmentation with the resampled data, which include simple threshold segmentation, body region elimination, trachea extraction, removal of interior cavities, left-right lung separation and lung nodule filling. The final results are obtained after projecting the preliminary segmentation to the original dataset and morphology smoothing. The proposed algorithm is applied to 20 patients＇ data （2556 slices） , and the results are compared to the manual segmentations. Results The algorithm can get accurate results with an average area overlapped ratio 99.02% to the manual segmentation by the radiologist, and works well for the abnormal cases （right-left connected, with nodules and uncompleted views ） . Through resampling, the time consumption of the algorithm is shortened significantly, typically by 50%, and the processing for one slice image is less than 0.25 s. Conclusions The proposed automatic lung parenchyma segmentation algorithm with excellent robustness and high speed, can get accurate result and satisfy the requirements of current clinical applications.
出处
《北京生物医学工程》
2012年第4期349-355,共7页
Beijing Biomedical Engineering
基金
国家自然科学基金(61071213
51006021)资助
关键词
肺实质
重采样
CT图像
分割
计算机辅助诊断
lung parenchyma
resampling
CT image
segmentation
computer aided diagnosis
