基于Local Jet变换空间纹理特征的肺结节分类方法

发布时间：2019-02-13 01:31

【摘要】：为了在纹理特征下改善肺结节良、恶性的模式识别,提出一种基于local jet变换空间的纹理特征提取方法。首先利用二维高斯函数的前三阶偏微分函数将结节原图像变换到local jet纹理图像空间,然后利用纹理描述子在该空间提取特征参数。以灰度值的前四阶矩和基于灰度共生矩阵的特征参数作为纹理描述子,分别提取结节原图像和变换后纹理图像的特征参数,以BP神经网络作为分类器,对同一纹理描述子下的2个不同图像空间的经核主成分分析优化后的特征参数集进行结节良、恶性分类。以157个肺结节(51个良性,106个恶性)作为实验数据进行对比实验,结果显示:两种纹理描述子基于local jet变换空间提取的特征参数分别获得82.69%和86.54%的分类正确率,较原图像空间提高6%~8%,同时AUC值提高约10%。实验结果表明,基于local jet变换空间提取的纹理特征可以有效地改善肺结节良、恶性的模式识别。
[Abstract]:In order to improve the pattern recognition of benign and malignant pulmonary nodules under texture features, a texture feature extraction method based on local jet transform space is proposed. Firstly, the first three order partial differential function of Gao Si function is used to transform the nodule original image into the local jet texture image space, and then the texture descriptor is used to extract the feature parameters in the space. The first four moments of gray value and the feature parameters based on gray level co-occurrence matrix are used as texture descriptors to extract the feature parameters of the original nodular image and the transformed texture image, respectively. The BP neural network is used as the classifier. The feature parameter sets of two different image spaces under the same texture descriptor are classified into benign and malignant nodules by kernel principal component analysis (KPCA). 157 pulmonary nodules (51 benign and 106 malignant) were used as experimental data. The results showed that the classification accuracy of the two texture descriptors was 82.69% and 86.54%, respectively. Compared with the original image space, the value of AUC is increased by about 10. The experimental results show that the texture features extracted from local jet transform space can effectively improve the pattern recognition of benign and malignant pulmonary nodules.
【作者单位】：天津大学精密仪器与光电子工程学院;天津医科大学附属肿瘤医院;天津市生物医学检测技术与仪器重点实验室;
【分类号】：R563;TP391.41

【相似文献】