CT图像人工智能辅助诊断系统鉴别肺结节良恶性的价值  

Value of artificial intelligence-assisted diagnostic system for CT image interpretation in differential diagnosis of benign and malignant pulmonary nodules

作　　者：沈小钦 梁红 朱晓琼 SHEN Xiaoqin;LIANG Hong;ZHU Xiaoqiong(Department of Respiratory Medicine,Anyue County People’s Hospital,Ziyang 642350 China;Department of Radiology,Anyue County People’s Hospital,Ziyang 642350 China)

机构地区：[1]四川省安岳县人民医院呼吸内科,四川资阳642350 [2]四川省安岳县人民医院放射科,四川资阳642350

出　　处：《中国辐射卫生》2024年第5期578-583,共6页Chinese Journal of Radiological Health

摘　　要：目的比较CT扫描图像人工智能辅助诊断系统和传统人工阅片用于阳性肺结节的检出效果以及良恶性鉴别诊断的价值,从而为人工智能用于肺癌临床筛查提供参考依据。方法以2019年3月—2023年12月在接受肺结节胸部CT扫描的病例为研究对象,肺结节CT扫描图像分别进行人工智能分析和传统人工阅片分析。以肺部病灶病理检查结果为金标准,比较人工智能分析和传统人工阅片用于阳性肺结节检出率和肺结节良恶性鉴别诊断效能。结果207例病例累计检出真阳性肺结节327个,人工智能分析对阳性肺结节检出率显著高于传统人工阅片(95.72%vs.86.85%;χ^(2)=16.16,P<0.01)。此外,人工智能分析对实性(χ^(2)=7.71,P<0.01)和磨玻璃肺结节的检出率(χ^(2)=5.80,P<0.05)均显著高于传统人工阅片,对<1(χ^(2)=4.97,P<0.05)、1~<2 cm(χ^(2)=7.04,P<0.01)和2~<3 cm肺结节的检出率(χ^(2)=4.91,P<0.05)亦显著高于传统人工阅片。人工智能分析鉴别肺结节良恶性的灵敏度、特异度、阳性预测值、阴性预测值和准确度分别为98.08%、91.53%、95.33%、96.43%和95.71%,而传统人工阅片鉴别肺结节良恶性的灵敏度、特异度、阳性预测值、阴性预测值和准确度分别为91.34%、77.97%、87.96%、32.62%和86.50%,人工智能分析鉴别肺结节良恶性的灵敏度(χ^(2)=4.70,P<0.05)、特异度(χ^(2)=4.20,P<0.05)、阴性预测值(χ^(2)=65.28,P<0.01)和准确度(χ^(2)=8.52,P<0.01)均显著高于传统人工阅片,但两者阳性预测值差异无统计性意义(χ^(2)=3.80,P>0.05)。结论CT扫描图像人工智能辅助诊断系统可较传统CT图像人工阅片显著提高阳性肺结节检出率,并可以提高肺结节良恶性鉴别诊断效能,值得在健康体检、肺癌早期筛查中进一步推广应用。Objective To compare artificial intelligence-assisted diagnostic system and conventional manual CT image interpretation for detection of positive pulmonary nodules and differential diagnosis of benign and malignant pulmonary nodules,and to provide a reference for the application of artificial intelligence in clinical screening for lung cancer.Methods Patients who underwent chest CT scans for pulmonary nodules from March 2019 to December 2023 were en-rolled.The CT images were subjected to artificial intelligence-based and conventional manual CT image interpretation.The pathological examination results of pulmonary lesions served as a gold standard for comparison of artificial intelligence-based and conventional manual CT image interpretation in detection rate of positive pulmonary nodules and differential dia-gnosis of benign and malignant pulmonary nodules.Results A total of 327 positive pulmonary nodules were identified in 207 patients.The detection rate of positive pulmonary nodules was significantly higher with artificial intelligence-based CT image interpretation than with conventional manual CT image interpretation(95.72%vs.86.85%;χ^(2)=16.16,P<0.01).Moreover,artificial intelligence-based CT image interpretation showed significantly higher detection rates for solid(χ^(2)=7.71,P<0.01)and ground-glass pulmonary nodules(χ^(2)=5.80,P<0.05)than conventional manual CT image interpretation.The detection rates for pulmonary nodules with<1 cm(χ^(2)=4.97,P<0.05),1 to<2 cm(χ^(2)=7.04,P<0.01),and 2 to<3 cm(χ^(2)=4.91,P<0.05)diameters were significantly higher with artificial intelligence-based CT image interpretation than with conventional manual CT image interpretation.The sensitivity,specificity,positive predictive value,negative predictive value,and accuracy for differential diagnosis of benign and malignant pulmonary nodules were 98.08%,91.53%,95.33%,96.43%,and 95.71%with artificial intelligence-based CT image interpretation and 91.34%,77.97%,87.96%,32.62%,and 86.50%with conventional CT image interpretation.Th

关 键 词：低剂量CT扫描 人工智能 肺结节 鉴别诊断 诊断价值 

分 类 号：R816[医药卫生—放射医学]