临床观察
西李,伟华董,湘生肖
2019, 43(11): 0-0.
摘要
目的：
评估CT图像纹理特征对≥6mm纯磨玻璃密度肺腺癌中浸润性病变的鉴别价值。
材料和方法：
回顾性分析2013年9月至2015年10月期间所有高分辨率CT图像表现为≥6mm肺部纯磨玻璃结节（Pure ground glass nodules, pGGNs）且病理结果明确的患者，并收集患者基线临床资料。共收集来自91例患者的91个pGGNs，依照病理结果将患者分为浸润前组（n=39）和浸润性组（n=52）。对每一个pGGN沿分界进行半自动分割并且提取该区域的图像纹理特征参数。对两组患者的临床数据和图像参数纹理特征参数分析比较。使用二元logistic回归分析浸润性病变的独立鉴别指标，使用接受者操作特性（receiver operating characteristic, ROC）曲线分析各独立指标及回归模型的鉴别诊断效能。
结果：
平均CT值、最大CT值、最大有效长径、表面积、体积、质量及逆差距在两组中存在统计学差异（P<0.05）。二元logistic回归结果证明逆差距(调整优势比,adjusted odds ratio [OR]= 0.559)、最大有效长径(adjusted OR=1.305)及平均CT值(adjusted OR=1.009)为浸润性病变的独立鉴别指标(P<0.05)。使用二元logistic回归模型鉴别浸润性病变的ROC曲线下面积（area under the curve，AUC）为0.809，鉴别效能优于单独使用逆差距（AUC=0.672）、平均CT值（AUC=0.660）及最大有效长径（AUC=0.704）。
结论：图像纹理参数分析能够较为准确的鉴别≥6mm的pGGNs中的浸润性病变，逆差距、平均CT值及最大有效长径为独立鉴别指标。
关键词：体层摄影术,X线计算机; 磨玻璃密度; 肺腺癌; 纹理分析
Computerized texture analysis differentiate lung invasive adenocarcinoma manifest as pure ground-glass nodules 6 mm or larger
Li Xi, XIAO Xiangsheng, DONG Weihua
(Department of Interventional Radiology, Changzheng Hospital, the Second Military Medical University, Shanghai 200003, China)
Abstract
Purpose:
To evaluate the performance of computerized texture analysis in diagnose invasive adenocarcinoma manifest as pure ground-glass nodules 6 mm or larger.
Material and Methods:
All patients with pathologically confirmed pure ground-glass nodules (pGGNs) 6 mm or larger between September 2013 and October 2015 were retrospectively reviewed, baseline demographic were collected. A total of 91 pGGNs from 91 patients were included. Patients were divided into preinvasive group (n=39) and invasive group (n=52) according to the pathology results. All PGGNs were semiautomatic segmented on images and computerized texture features were extracted. Patients’ demographics and computerized texture features were compared between groups. Binary logistic regression analysis were used to investigate the differentiating factors of invasive lesions from invasive lesions. Discriminating performance of the texture features were evaluated by ROC curve analysis.
Results:
Between two groups, mean CT value, maximum CT value, maximum diameter, volume, surface area, mass and inverse difference moment (IDM) showed significantly different (P < 0.05). Binary logistic regression analysis revealed that IDM (adjusted OR=0.559), maximum diameter (adjusted OR=1.305) and mean CT value (adjusted OR=1.009) were independent predictors of invasive lesions. A combination of these features showed excellent differentiating performance (AUC=0.809), which is better than IDM (AUC=0.672), mean CT value (AUC=0.660), or maximum diameter (AUC=0.704) alone.
Conclusion:
Computerized texture analysis can differentiate lung invasive adenocarcinoma manifest as pGGNs with diameter ≥6mm. IDM, mean CT value and maximum diameter are independent differentiators.
Keywords: Computed tomography, ground-glass opacity, lung adenocarcinoma, texture analysis
