一种肺部肿瘤CT图象序列的自动分割方法

An Automatic Segmentation Approach for CT Serial Images of Lung Tumors

肺部肿瘤序列图象的自动分割是计算机肺部肿瘤三维辅助诊断系统的关键技术之一,肿瘤与周围组织关系的复杂性造成分割困难.为了给医生提供准确的肺部肿瘤影像,运用纹理分析和径向基神经网络实现了肺部肿瘤CT图象序列的自动分割,并根据相邻层肿瘤图象灰度、位置的相关性,提出了一种自动获取多层肿瘤区域神经网络训练样本的阈值分割算法.该算法首先计算图象纹理统计参数,以组成特征矢量空间,然后利用自适应径向基神经网络对特征矢量进行分类来实现肿瘤序列图象的自动分割.实验结果表明,与基于灰度的区域增长法和基于梯度算子和形状算子的最优阈值的分割方法相比较,该方法不仅能充分利用肺部肿瘤序列图象的三维信息,还可最大限度地减少人工干预,且分割结果较好地表现了肿瘤形态特征,经临床医生评估,具有较好的临床指导价值.

The segmentation of lung tumor serial images is one of the key techniques of Computer Lung Tumor Three-dimensional Aided Diagnosis System. The complex relation between tumor and its adjacent tissue makes it difficult to get good result. For providing doctors more accuate lung image, an automatic segmentation of lung tumor in CT serial images is presented based on texture analysis and radial basis function(RBF) neural network. With the correlation of tumor's gray level and position in sequential slices, we got the training swatch of tumor region automatically. Some second-order statistical texture parameters were computed for composing feature space; a classification procedure based on RBF neural network was applied to this space to segment the tumor. Compared with region growth algorithm and the multi-criterion segmentation algorithm, the experiment demonstrates that the proposed method can make full use of three-dimensional information of tumor serial images, and reduce manual intervention as possible. The segment results also confirm the validity and the clinical value of the proposed method.