基于人工智能CT定量分析对长径≤10 mm肺磨玻璃结节浸润程度的预测价值  

The predictive value in infiltration degree of ≤10mm pulmonary ground-glass nodules based on artificial intelligence CT quantitative analysis

作　　者：邓琦[1] 徐志锋[1] 周涛[1] 李勤祥[1] 段锐 DENG Qi;XU Zhifeng;ZHOU Tao;LI Qinxiang;DUAN Rui(Department of Medical Imaging,Foshan First People's Hospital,Foshan,Guangdong Province 528000,China)

机构地区：[1]佛山市第一人民医院医学影像科,广东佛山528000

出　　处：《实用放射学杂志》2023年第7期1088-1092,共5页Journal of Practical Radiology

摘　　要：目的 探讨基于人工智能(AI)CT定量分析对长径≤10 mm肺磨玻璃结节(GGN)浸润程度的预测价值.方法 回顾性选取经胸部高分辨率CT(HRCT)扫描长径≤10 mm的GGN且手术病理证实为早期肺腺癌患者109例(共113个GGN)根据病理浸润程度分为前驱腺体病变组[不典型腺瘤样增生(AAH)+原位腺癌(AIS)]32个、微浸润腺癌(MIA)组43个、浸润性腺癌(IAC)组38个.利用SPSS 24.0软件对AI相关定量参数(长径、实性体积、实性体积占比、实性质量、实性质量占比、最大CT值、最小CT值、平均CT值、中位数CT值、标准差、峰度、偏度、熵)进行统计学分析.结果 实性体积、实性体积占比、实性质量、实性质量占比、平均CT值、熵在不同浸润程度GGN的组间差异均有统计学意义(P<0.05).Bayes逐步判别分析显示上述统计学差异变量的判别效果较好,总体诊断准确率81.4%.有序logistic回归分析筛选出实性质量[95%置信区间(CI)0.011~0.274,P=0.033]、实性质量占比(95%CI 0.061~0.364,P=0.006)、熵(95%CI 2.298~5.829,P<0.001)是GGN浸润程度的独立预测因子.独立预测参数的受试者工作特征(ROC)曲线结果显示AAH+AIS与MIA+IAC的鉴别诊断效能由高到低分别为熵[曲线下面积(AUC)0.841]、实性质量(AUC 0.781)、实性质量占比(AUC 0.772),MIA与IAC的鉴别诊断效能由高到低分别为实性质量占比(AUC 0.784)、熵(AUC 0.771)、实性质量(AUC 0.722).结论 基于AI的CT定量分析能较好地预测长径≤10 mm的GGN浸润程度,特别是实性质量、实性质量占比、熵有助于鉴别诊断.Objective To explore the value of artificial intelligence(AI)CT quantitative analysis in predicting infiltration degree of pulmonary ground-glass nodules(GGN)with length diameter≤10 mm.Methods A total of 109 patients(113 GGN)with length diameter<10 mm scanned by chest high-resolution CT(HRCT)and confirmed by surgery and pathology as early-stage lung adenocarcinoma were selected retrospectively.According to the degree of pathological invasion,they were divided into prodromal gland lesions groupLatypical adenomatous hyperplasia(AAH)+adenocarcinoma in situ(AIS)J(32 cases),minimally invasive adenocarcinoma(MIA)group(43 cases)and invasive adenocarcinoma(IAC)group(38 cases).SPSS 24.0 software was used for statistical analysis of AI-related quantitative parameters(length diameter,solid volume,solid volume ratio,solid mass,solid mass ratio,maximum CT value,minimum CT value,average CT value,median CT value,standard deviation,kurtosis,skewness,and entropy).Resultss The solid volume,solid volume ratio,solid mass,solid mass ratio,average CT value,and entropy were statistically significant in the differences between groups of GGN with different infiltration degrees(P<0.05).Bayes stepwise discriminant analysis showed that the above statistical difference variables had better discriminant effect,and the overall diagnostic accuracy was 81.4%.Solid mass[95%confidence interval(CI)0.011-0.274,P=0.033],solid mass ratio(95%CI 0.061-0.364,P=0.006),and entropy(95%CI 2.298-5.829,P<0.001)were independent predictors for the degree of GGN infiltration screened out by ordinal logistic regression analysis.The receiver operating characteristic(ROC)curve results of the independent predictive parameters showed that the differential diagnostic efficacy of AAH+AIS and MIA+IAC from high to low were entropy Larea under the curve(AUC)o.841],solid mass(AUC 0.781),and solid mass ratio(AUC 0.772),the differential diagnostic efficacy of MIA and IAC from high to low were solid mass ratio(AUC 0.784),entropy(AUC 0.771),and solid mass(AUC 0.722).Conclusio

关 键 词：人工智能 定量分析 磨玻璃结节 计算机体层成像 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R563[自动化与计算机技术—控制科学与工程] R814.42[医药卫生—呼吸系统]