肺结节CT筛查中人工智能诊断系统应用价值评估

Application Value of the Artificial Intelligence Assisted Diagnostic System for Pulmonary Nodules CT Screening

目的评价在肺结节电子计算机断层扫描(computed tomography,CT)筛查中人工智能(artificial intelligence,AI)诊断系统的优劣势。方法收集作者医院1000例患者肺结节筛查CT影像资料,采用4种阅片方式(场景A:AI软件独立阅片;场景B:住院医师阅片后主治医师审核;场景C:AI软件辅助下放射医师阅片;场景D:2位高年资放射医师综合以上阅片结果达成共识),以场景D诊断结果作为参照,分组比较其他不同阅片场景下肺结节检出效能。结果在场景D下1000例患者中共查出1874枚肺结节。以场景D检查结果为参照,A、B、C3种场景下肺结节的准确检出率分别为92.90%、92.80%和97.81%,组间差异具有统计学意义(χ^(2)=59.26,P<0.01),其中场景C准确检出率明显高于A、B场景(P均<0.01)。A、B、C3种场景对下肺结节假阳性率分别为7.31%、2.99%和1.92%,组间差异具有统计学意义(χ^(2)=37.28,P<0.01),其中C场景假阳性率明显低于A、B场景(P均<0.01)。结论AI诊断软件的应用可以明显提高放射科医师在CT筛查中对肺结节的检出能力,但仍可能会出现少数假阳性病例。

Objective To evaluate the benefits and drawbacks of the artificial intelligence(AI)diagnostic system for pulmonary nodules screening in chest computed tomography(CT).Methods The CT data for pulmonary nodule screening of 1000 patients in author′s hospital were collected and were evaluated through following4 modes:(Scenario A:films were read by AI system independently;scenario B:films were read by a resident and then reviewed by attending doctor;scenario C:a radiologist read films assisted by the AI software;scenario D:two senior radiologists reached a consensus based on above results).With the diagnostic results in scenario D as reference,the efficiency for pulmonary nodule detection in other scenarios was compared among each other.Results In scenario D,a total of 1,874 lung nodules were found in 1,000cases.With the results from scenario D as a reference,the detection rates of lung nodules in scenario A(92.90%),scenario B(92.80%)and scenario C(97.81%)were significantly different(χ^(2)=59.26,P<0.01);and the detection rate of scenario C was higher than that of scenario A and scenario B(all P<0.01).The false positive rates of lung nodules in scenario A(7.31%),B(2.99%)and C(1.92%)were significantly different(χ^(2)=37.28,P<0.01);and the false positive of scenario C was lower than that of scenario A and scenario B(all P<0.01).Conclusion The application of AI diagnostic system can significantly improve radiologists′ability to detect pulmonary nodules in CT screening.However,the possibility of misdiagnosis such as false positive cases still can happen.