摘要
目的:探讨人工智能(AI)定量参数在肺磨玻璃结节(GGN)病理类型中的预测价值。方法:回顾收集2022年1月至2023年4月经手术病理证实的肺GGN患者65例,按照2021版WHO肺肿瘤新分类标准分成三组:前驱腺体病变组(不典型腺瘤样增生AAH+原位癌AIS组)、微浸润性腺癌组(MIA组)、浸润性腺癌组(IAC组);另根据有无浸润性分为两组:非浸润性腺癌组(AAH+AIS组)及浸润性腺癌组(MIA+IAC组);所有患者均执行胸部CT扫描及肺窗薄层重建,并通过AI方法测得GGN的7个相关量化参数:长径(mm)、短径(mm)、体积(mm3)、平均CT值(Hu)、最大CT值(Hu)、最小CT值(Hu)、实性占比(%)。统计分析不同组间AI定量参数的差异性,使用ROC曲线评价参数预测肺GGN病理类型的效能。结果:前驱腺体病变22例,微浸润性腺癌21例,浸润性腺癌22例。7个AI定量参数在三组间比较均有统计学差异(P<0.05),而且显示7个参数值由前驱腺体病变至浸润性腺癌呈递增趋势。非浸润性腺癌与浸润性腺癌组间比较,长径、短径、体积、最大CT值显示具有统计学差异(P<0.05),而平均CT值、最小CT值、实性占比在两组间比较无统计学意义(P>0.05)。通过ROC曲线分析,取GGN的长径界值8.5mm,产生曲线下面积(AUC)0.758,敏感性67.4%,特异性68.2%。短径界值6.5mm, AUC=0.736,敏感性69.8%,特异性59.1%。体积界值194.47mm3,AUC=0.734,敏感性76.7%,特异性59.1%。CT最大值界值-138.5HU,AUC=0.723,敏感性79.1%,特异性63.6%。结论:AI定量参数能在术前有效预测肺GGN的病理类型,尤其GGN的长径预测效能最大,随着长径的增加,为浸润性腺癌的机会就越大,这为临床及时干预治疗及预后评估提供重要的参考依据,值得广泛推广应用。
Objective:To explorethepredictivevalueof quantitativeparametersof artificial intelligence(AI)in the pathological types of ground glass nodule(GGN)of lung.Methods:Sixty-five patients with lung GGN confirmed by surgery and pathology from January 2022 to April 2023 were retrospectively collected and divided into three groups according to the new classification standard of lung tumors of WHO in 2021:precursor gland lesion group(atypical adenomatous hyperplasia AAH+carcinoma in situ AIS group),micro-invasive adenocarcinoma group(MIA group)andinvasive adenocarcinoma group(IAC group);In addition,they were divided into two groups according to whether they were invasive:non-invasive adenocarcinoma group(AAH+AIS group)and invasive adenocarcinoma group(MIA+IAC group);All patients underwent chest CT scanning and thin-layer reconstruction of lung window,and seven related quantitative parameters of CGN were measured by AI method:long diameter(mm),short diameter(mm),volume(mm'),average CT value(Hu),maximumCT value(Hu),minimumCT value(Hu)and solid ratio(%).ThedifferencesofAI quantitative parameters among different groups were statistically analyzed,and the effectiveness of parameters in predicting pathological types of lung GGN was evaluated by ROC curve.Results:There were 22 cases of precursor gland lesions,21 cases of micro-invasive adenocarcinoma and 22 cases of invasive adenocarcinoma.Seven quantitative parameters of AI were statistically different among the three groups(P<0.05),and the values of seven parameters showed an increasing trend from precursor gland lesions to invasive adenocarcinoma.Compared with the invasive adenocarcinoma group,the long diameter,short diameter,volume and maximum CT value showed statistical dfferences(P<0.05),while the average CT value,minimum CT value and solid ratio were not statistically significant(P>0.05).According to the analysis of ROC curve,the cutoff value of length and diameter of GGN is 8.5mm,and the area under curve(AUC)is0.758,with sensitivity of 67.4%and specificity of 68.2%.The critical value of short diameter is 6.5mm,AUC=0.736,sensitivity is 69.8%,and specificity is 59.1%.The volume boundary value was 194.47mm^(3),AUC=0.734,sensitivity was 76.7%,and specificity was 59.1%.The maximum value of CT was-138.5HU,AUC=0.723,sensitivity was 79.1%,and specificity was 63.6%.Conclusion:AI quantitative parameters can effectively predict the pathological types of lung GCN before operation,especially the length and diameter of GGN have the greatest prediction efficiency.With the increase of the length and diameter,the chances for invasive adenocarcinoma are greater,which provides important reference for timely clinical intervention and prognosis evaluation,and is worthy of wide application.
作者
杨井
钱娟
钱林清
顾伟光
吴晓钢
Yang Jing;Qian Juan;Qian Linqing;Gu Weiguang;Wu Xiaogang(Department of Radiology,Wuzhong People's Hospital,Suzhou,Jiangsu 215128)
出处
《现代医用影像学》
2024年第1期33-37,共5页
Modern Medical Imageology
基金
苏州市科学技术局指令性项目,项目名称:亚实性肺结节CT表征联合智能量化参数预测肺腺癌病理亚型应用基础研究,项目编号:SKY2023106。
关键词
肺磨玻璃结节
人工智能
病理类型
鉴别诊断
预测
pulmonary ground-glass nodules
artificial intelligence
pathological types
differential diagnosis
prediction
