摘要
目的:探究人工智能(AI)提取肺结节大小、密度以及纹理特征和胸部HRCT影像特征对鉴别肺亚实性结节浸润性病变的价值。方法:收集2018年1月—2021年8月我院就诊的166例患者(共166个肺亚实性结节)的临床资料。以术后病理结果为金标准,根据有无肺间质组织浸润,将肺结节分为非浸润与浸润组。比较AI与人工鉴别肺结节浸润性的结果。比较两组间AI提取特征及HRCT影像特征。用逐步Logistic回归分析分别建立AI、HRCT及两者联合的预测模型,比较三个模型的预测效能。结果:病理结果显示非浸润组46例,浸润组120例。AI与影像医师人工鉴别肺结节浸润性病变准确度分别78.92%、88.55%,两组准确度具有统计学差异(P<0.001)。AI提取的17个特征中,除了偏度外,其余16个特征在两组间有统计学差异(P均<0.05)。HRCT影像特征中最大截面长短径、边缘、形态、细支气管改变、胸膜凹陷征、血管集束征及实性成分在两组间有统计学差异(P均<0.05)。ROC曲线显示,AI、HRCT及联合模型鉴别肺结节浸润性的AUC值分别为0.867.0.900.0.922,联合模型的诊断效能最高,其诊断敏感度81.67%、特异度93.48%、阳性预测值97.03%、阴性预测值66.15%。DeLong检验表明:联合模型的AUC高于AI和HRCT模型(Z=-2.564、-2.066,P=0.010、0.039),但AI与HRCT模型间的AUC无统计学差异(Z=-1.155,P=0.248)。结论:AI、HRCT及联合模型对鉴别肺结节浸润性病变均有较高价值,联合模型预测效能明显高于AI及HRCT模型,说明AI辅助影像医生提高肺结节诊断和评估的准确性。
Objective:To explore the value of artificial intelligence(Al)in extracting the size,density,and texture features of pulmonary nodules,as well as the imaging features of thoracic HRCT in differentiating invasive lesions of subsolid pulmonary nodules.Methods:Clinical data of 166 patients(total of 166 CGN)who received treatment at our hospital from January 2018 to August 2021 were collected.Post-operative pathological findings were regarded as the gold standard.Based on the presence or absence of tissue invasion,the pulmonary nodules were classified into the non-invasive group and the invasive group.The outcomes of Al-based and manual discrimination of the invasiveness of pulmonary nodules were compared.Additionally,AI-related parameters and HRCT imaging characteristics were contrasted between the two groups.Prediction models were constructed separately for AI parameters,HRCT imaging features,and their combination.Subsequently,the predictive performances of these three models were evaluated and compared.Results:The pathological findings indicated that there were 46 cases in the non-invasive group and 120 cases in the invasive group.The accuracies of AI and radiologists in differentiating invasive lesions of pulmonary nodules were 78.92% and 88.55%respectively.A statistically significant difference in accuracy was observed between the two groups(P<0.001).Among the 17 features extracted by AI,16 features,with the exception of skewness,exhibited statistically significant differences between the two groups(all P<0.05).The differences in the imaging features of pulmonary nodules in terms of nodule maximum cross-sectional length and diameter,margin,morphology,fine bronchial changes,pleural depression sign,vascular cluster sign and solid component were statistically significant between the two groups(all P<0.05).The ROC curve analysis demonstrated that the area under the curve(AUC)values for the AI model,HRCT model,and the combined model of them in differentiating the invasiveness of pulmonary nodules were 0.867,0.900,and 0.922 respectively.The combined model exhibited the highest diagnostic efficacy for differentiating the invasiveness of pulmonary nodules,with a diagnostic sensitivity of 81.67%,a specificity of 93.48%,a positive predictive value of 97.03%,and a negative predictive value of 66.15%.The DeLong test showed that the AUC of the combined model was higher than that of the AI and HRCT models(Z=-2.564,-2.066,P=0.010,0.039),but the difference in AUC between the AI and HRCT models was not statistically significant(Z=-1.155,P=0.248).Conclusion:The AI model,HRCT model,and the combined model all exhibit high utility in the identification of infiltrative lesions of pulmonary nodules.The predictive efficacy of the combined model is significantly higher than that of the AI and HRCT models.It indicates that AI-assisted radiologists can improve the accuracy of the diagnosis and assessment of pulmonary nodules.
作者
阮惠萍
陈萍
郑德春
RUAN Hui-ping;CHEN Ping;ZHENG De-chun(Department of Radiology,Clinical Oncology School of Fujian Medical University,Fujian Cancer Hospital,Fuzhou 350014,China)
出处
《中国临床医学影像杂志》
北大核心
2025年第11期784-788,共5页
Journal of China Clinic Medical Imaging
基金
福建省卫生健康科技计划项目(编号2023GGA055)。
