基于HRCT图像的纹理分析技术在鉴别良恶性肺部病变中的价值

Differentiating of benign and malignant lesions of the lung using texture analysis based on high-resolution computed tomography

目的:探讨基于HRCT图像的纹理分析技术在鉴别良恶性孤立性肺实性病变中的价值。方法:收集69例经本院病理证实的孤立性肺实性病变(良性28例,恶性41例)患者的CT资料。所有患者术前行胸部HRCT平扫及双期增强扫描,经后处理获得5组图像,包括骨和软组织重建算法平扫肺窗图像及平扫、动脉期和静脉期软组织重建算法纵隔窗图像。采用MaZda软件通过手动勾画ROI的方式提取病变的纹理特征参数,纹理特征选择方法包括Fisher系数、分类错误概率联合平均相关系数(POE+ACC)、交互信息(MI)及上述3种方法联合法(FPM)。纹理特征分类统计方法包括原始数据分析(RDA)、主成分分析(PCA)、线性分类分析(LDA)和非线性分类分析(NDA)。各纹理特征参数的诊断结果以误判率形式表示。结果:5组HRCT图像中,对良恶性肺部病变有鉴别诊断意义的纹理特征参数主要来自软组织重建算法肺窗及静脉期纵隔窗图像。特征选择方法中,Fisher系数、POE+ACC和MI及FPM鉴别两类病变的误判率分别为10.14%～44.93%、10.14%～52.17%、10.14%～42.03%及7.25%～50.72%。特征统计方法中,NDA对两种病变的误判率(7.25%～17.39%)较RDA(21.74%～50.72%)、PCA(27.54%～52.17%)和LDA(8.70%～36.23%)均低。结论:基于HRCT图像的纹理分析方法能为孤立性肺实性良恶性病变的诊断提供更多有价值的信息,有助于提高HRCT对肺部实性病变的诊断准确性。

Objective:To investigate the diagnostic value of texture analysis derived from high-resolution computed tomography(HRCT)images in differentiating benign and malignant solid lesions of lung.Methods:Clinical and imaging data of forty-one patients with malignant pulmonary lesion and twenty-eight patients with benign pulmonary lesion were retrospectively analyzed in our study.All patients underwent plain and dual phase contrast-enhanced HRCT scan.Five types of images were obtained from post-processing,including lung window plain images with bone algorithm and soft tissue algorithm,mediastinal window images of non-enhanced,arterial phase and venous phase with soft tissue algorithm.The ROI of each lesion was manually drawn out,and then texture features were calculated using MaZda software.The methods for feature selection included Fishers coefficient,classification error probability combined with average correlation coefficients(POE+ACC),mutual information(MI)and the combination of the three methods(FPM)mentioned above.The statistical methods including raw data analysis(RDA),principal component analysis(PCA),linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)were used to distinguish malignant from benign pulmonary lesions.The diagnostic results were shown using the index of misclassification rate.Results:The texture features for differentiating malignant and benign pulmonary lesions were mainly derived from soft tissue algorithm lung window and venous phase mediastinal window images.The misclassification rates of the feature selection methods were similar in MI(10.14%to 42.03%),Fisher coefficient(10.14%to 44.93%)and POE+ACC(10.14%to 52.17%),while the misclassification rate of FPM(7.25%to 50.72%)was the lowest.In the statistical methods,NDA(7.25%to 17.39%)had lower misclassification rate than RDA(21.74%to 50.72%),PCA(27.54%to 52.17%)and LDA(8.70%to 36.23%).Conclusion:Texture analysis based on HRCT images can provide more reliable and valuable evidence for differentiating benign from malignant pulmonary lesions,thus can increase the diagnostic accuracy of HRCT in pulmonary solid lesions.