一种基于 DICOM 序列影像的肺结节 ROI 自动检测方法

Pulmonary Nodules ROI Automatic Detection Based on DICOM Sequence Images

本文是以DICOM标准的肺部序列影像为研究对象,将CT图像序列分割提取获得肺实质模块,再获取种子区域进行优化分割,最后通过ROI检测提取肺部特征信息并进行分类,从而达到肺结节ROI自动检测的目的。实验结果表明本文算法对微小结节特别是3mm以下的结节敏感性不高,而直径大于5mm的结节检出较为准确。实验中出现假阳性结节的个数较多,说明所选特征向量与判别分类标准比较严格,分类器的一些参数需要进一步优化,以达到更高的检出率及更低的漏检率。

This paper is based on lung image sequence of DICOM standard as the research object, segmenting a sequence of CT image for lung parenchyma, then getting the seed region were again optimized segmentation, and finaly being detected by ROI extraction and classification of lung feature information, so achieve the automatic detection of ROI lung nodules. Experimental results show that for the proposed algorithm nodules sensitivity is not high, especially less than 3mm tiny nodules, and the detection of the diameter greater than 5mm nodules is more accurate, the number of false-positive nodules in the experiment are larger, indicating the selected feature vectors and discriminate classification criteria is stricter, and the classifier needs to optimize some parameters in order to achieve a higher detection rate and lower missed rate.