北京生物医学工程2017，Vol.36Issue(3)：257-261,5.DOI:10.3969/j.issn.1002-3208.2017.03.06.
基于PET/CT图像纹理参数的肺结节诊断模型
Diagnostic models of pulmonary nodules based on texture features of PET/CT images
摘要
Abstract
Objective To establish diagnostic models for pulmonary nodules based on texture features of PET/CT images and improve the identification of lung cancer.Methods We selected 52 patients who underwent 18F-PET/CT scan from the Department of Nuclear Medicine of Capital Medical University Xuanwu Hospital,then the PET/CT images and the information of demography and medical imaging were collected.The PET/CT texture parameters were extracted by Contourlet conversion and co-occurrence matrix arithmetic.Based on which we built the SVM (support vector machine) diagnosis models and got the results of classification for each pulmonary nodule.To improve the effectiveness of diagnostic models,we also built SVM models with the parameters of nodule margin,maximum standard uptake value and halo sign.At last we evaluated the effectiveness of the diagnostic models by the index of sensitivity,specificity,accuracy,etc.Results The diagnostic model of texture features had the sensitivity of 90.7% and specificity of 93.5%.The diagnostic model based on texture plus medical imaging information had the sensitivity of 95.7% and specificity of 100.0%.Conclusions This diagnosis models based on texture features of PET/CT images have good classification of benign and malignant nodules.关键词
PET/CT/肺结节/图像纹理/诊断模型/支持向量机Key words
PET/CT/pulmonary nodule/texture feature/diagnostic models/support vector machine分类
医药卫生引用本文复制引用
马圆,陈斯鹏,田思佳,梁志刚,崔春蕾,郭秀花..基于PET/CT图像纹理参数的肺结节诊断模型[J].北京生物医学工程,2017,36(3):257-261,5.基金项目
国家自然科学基金(81530087、81373099)资助 （81530087、81373099）
