纹理分析对CT引导下经皮肺穿刺活检术出血并发症的预测价值初探

A Preliminary Study on the Value of Texture Analysis in Predicting Bleeding Complications of CT-guided Percutaneous Lung Biopsy

目的:探讨基于CT图像的纹理分析对CT引导下经皮肺穿刺出血并发症的预测价值。方法:回顾性分析130例行CT引导下经皮肺穿刺活检患者的术前平扫图像,术区有大于或等于2级肺出血者为有出血组,0级或1级肺出血者为无/少量出血组。首先随机选取100例作为训练组,采用MaZda软件,分别手动勾画出平扫肺窗图像上预穿刺路径周边的肺野作为感兴趣区(ROI),分别通过Fisher系数、分类错误概率联合平均相关系数(POE+ACC)、交互信息(MI)法筛选出区分出血组及无/少量出血组最具有价值的纹理特征,然后分别采用原始数据分析(RDA)、主要成分分析(PCA)、线性分类分析(LDA)和非线性分类分析(NDA)四种特征分类统计方法进行判断,结果以错判率形式表示;最后再根据得到的最优纹理参数及特征分类方法分别对另外30例图像加以验证。结果:以穿刺路径周边的肺野作为ROI时,最低错判率为11.00%(11/100),该结果出现在特征选择方法采用POE+ACC或MI,特征分类统计方法采用NDA时,以此结果进行验证的错判率分别为13.33%(4/30)和16.67%(5/30),两者差异无统计学意义。结论:分析预穿刺路径周边肺野的纹理特征有助于预测CT引导下肺穿刺并发出血的风险,为选择合适的穿刺路径以减少肺出血并发症提供依据。

Objective: In this paper we intends to explore the value of texture analysis based on CT image in predicting the complications of percutaneous pulmonary puncture hemorrhage under the guidance of CT. Methods:The preoperative plain scan images of 130 patients who underwent CT-guided percutaneous lung biopsy were analyzed retrospectively. The patients with pulmonary hemorrhage greater than or equal to grade 2 in the operative area were assined into the bleeding group while the patients with grade 0 or grade 1 pulmonary hemorrhage were assined into the no/small bleeding group. 100 cases were randomly selected as the training group, and the lung field around the pre-puncture path on the plain scan lung window image was manually drawn as the region of interest(ROI) by using MaZda software. The most valuable texture features were selected by methods of Fisher coefficient, classification error probability joint average correlation coefficient(POE + ACC) and interactive information(MI) to distinguish between bleeding group and no/small amount of bleeding group. Then, the four following feature classification statistical methods;raw data analysis(RDA), principal component analysis(PCA),linear classification analysis(LDA) and nonlinear classification analysis(NDA), were used for judgement, and the results were shown by way of error rate. Finally, the other 30 cases were verified according to the optimal texture parameters and feature classification method. Results: The lowest error rate was 11.00%(11/100) when the lung field around the puncture path was used as ROI. The error rates were respectively 13.33%(4/30) and 16.67%(5/30), when the feature selection method was POE + ACC or MI, and the feature classification statistical method was NDA, there was no significant difference between the two groups. Conclusion: The analysis of the texture characteristics of the lung field around the puncture path is helpful in predicting the risk of pulmonary puncture complicated with hemorrhage under the guidance of CT, and can provide certain basis for selecting a suitable puncture path to reduce the complications of pulmonary hemorrhage.