摘要
目的探讨胸部CT人工智能技术联合CT征象诊断肺磨玻璃结节(GGN)良恶性及侵袭性的价值。方法选取2021年6月至2023年6月收治的肺GGN患者121例,根据手术病理诊断结果分为良性组24例与恶性组97例。患者均行胸部CT检查,分析影像医师阅片CT征象、人工智能定量参数对肺GGN良恶性的诊断价值。并将恶性组根据侵袭性分为非浸润亚组58例与浸润亚组39例,对比2亚组胸部CT人工智能定量参数,分析各参数与恶性肺GGN侵袭性的关系。结果恶性组分叶征、毛刺征、支气管充气征、空泡征、胸膜凹陷征、血管集束征比例高于良性组,结节长径、最大面积、体积、平均CT值、心胸比率(CTR)大于良性组(P<0.05)。受试者工作特征曲线分析,CT征象分叶征、毛刺征、空泡征、支气管充气征、血管集束征、胸膜凹陷征诊断肺GGN良恶性的曲线下面积(AUC)分别为0.694、0.663、0.669、0.669、0.642、0.756,联合诊断AUC为0.874;人工智能定量参数结节长径、最大面积、体积、平均CT值、CTR诊断肺GGN良恶性的AUC分别为0.787、0.792、0.751、0.770、0.789,联合诊断AUC为0.882;CT征象与人工智能定量参数联合诊断肺GGN良恶性的AUC为0.951,诊断价值最优。浸润亚组结节长径、结节短径、最大面积、体积、平均CT值、CTR大于非浸润亚组(P<0.05)。结节长径、结节短径、最大面积、体积、平均CT值、CTR与恶性肺GGN侵袭性呈正相关(P<0.01)。结论胸部CT人工智能技术联合CT征象可有效提高肺GGN良恶性的诊断准确率,可通过二者联合诊断早期鉴别肺GGN良恶性以及恶性肺GGN侵袭性,以制订针对性干预方案。
Objective To explore the value of artificial intelligence(AI)-assisted chest CT combined with CT signs in the diagnosis of the benign,malignant,and invasive nature of ground-glass nodules(GGN).Methods A total of 121 patients with pulmonary ground-glass nodules(GGN)from June 2021 to June 2023 were selected and divided into the benign group(n=24)and the malignant group(n=97)based on results of surgical pathology diagnosis.All patients underwent chest CT scans,and the diagnostic value of radiologists'interpretation of CT signs and AI-derived quantitative parameters for the benign and malignant nature of pulmonary GGN was analyzed.The malignant group was further divided into a non-invasive subgroup(n=58)and an invasive subgroup(n=39)according to invasiveness.The AI-derived quantitative parameters of chest CT scans in the two subgroups were compared,and the relationship between each parameter and the invasiveness of malignant pulmonary GGN was analyzed.Results The malignant group had a higher proportion of lobulated sign,spiculated sign,air bronchogram sign,vacuolar sign,pleural indentation sign,and vessel convergence sign than the benign group,and the nodule length,maximum area,volume,average CT value,and cardiothoracic ratio(CTR)were also greater in the malignant group than in the benign group(P<0.05).According to the receiver operating characteristic(ROC)curve analysis,the area under the curve(AUC)of CT signs including lobulated sign,spiculated sign,vacuolar sign,air bronchogram sign,vessel convergence sign,and pleural indentation sign in diagnosing the benign or malignant nature of pulmonary GGN was 0.694,0.663,0.669,0.669,0.642,and 0.756,respectively,with a combined AUC of 0.874.The AUC for AI-derived quantitative parameters including nodule length,maximum area,volume,average CT value,and CTR in diagnosing the benign or malignant nature of pulmonary GGN was 0.787,0.792,0.751,0.770,and 0.789,respectively,with a combined AUC of 0.882.The combined diagnosis of CT signs and AI-derived quantitative parameters achieved an AUC of 0.951,indicating the optimal diagnostic value.In the infiltrative subgroup,long and short diameters of nodules,maximum area,volume,average CT value,and CTR were greater than those in the non-infiltrative subgroup(P<0.05).Long and short diameters of nodules,maximum area,volume,average CT value,and CTR were positively correlated with the invasiveness of malignant pulmonary GGN(P<0.01).Conclusion The combination of AI-assisted chest CT and CT signs diagnosis can effectively improve the diagnostic accuracy of the benign or malignant nature of pulmonary GGN.This combined approach allows for early differentiation between benign and malignant nature of pulmonary GGN,as well as the invasiveness of malignant pulmonary GGN,enabling the development of targeted intervention plans.
作者
王荣平
陈尚岳
WANG Rongping;CHEN Shangyue(Department of Radiology,Beijing Xiaotangshan Hospital,Beijing 102211,China;Department of Pharmacy,Beijing Xiaotangshan Hospital,Beijing 102211,China)
出处
《临床误诊误治》
2025年第4期37-42,共6页
Clinical Misdiagnosis & Mistherapy
基金
北京小汤山医院科研项目(汤2021-12)。
关键词
肺磨玻璃结节
胸部CT
人工智能技术
良恶性
侵袭性
浸润
诊断价值
受试者工作特征曲线
Pulmonary ground-glass nodule
Chest CT
Artificial intelligence technology
Benign and malignant nature
Invasiveness
Infiltration
Diagnostic value
Receiver operating characteristic curve
