AI辅助CT在肺结节良、恶性诊断中的系统评价

Comparative performance of AI and radiologists in diagnosing pulmonary nodules on CT

摘要

目的 评价人工智能（AI）辅助计算机断层扫描（CT）诊断肺结节良、恶性的价值。方法 通过检索PubMed、The Cochrane Library、Embase、Web of Science、中国知网、中国生物医学文献服务系统，搜集AI结合CT评估肺结节良、恶性的相关研究。对符合纳入、排除标准的文献使用QUADAS-2量表进行方法学质量评估。运用RevMan5.3和Stata18.0进行数据整合与分析。分组比较AI与临床医师在肺结节性质判别中的诊断效能，用 Z检验比较2组的差异，并构建综合受试者操作特征曲线；用异质性检验分析研究间差异的影响因素，以AI系统类型分组进行亚组分析，比较不同系统间敏感度、特异度等指标的差异。结果 AI辅助CT诊断与医生阅片的合并诊断效能指标敏感度、特异度、阳性似然比、阴性似然比、诊断比值比，及综合受试者操作特征曲线下面积，差异均无统计学意义（P > 0.05）。结论 AI辅助CT在肺结节性质判别中展现出与传统医生阅片相当的诊断效能，敏感度略高，提示AI辅助CT在降低漏诊率方面或有潜力，但需大规模临床验证。

Abstract

Objective To evaluate artificial intelligence (AI) assisted computed tomography (CT) in diagnosing benign and malignant pulmonary nodules. Methods Systematic searches were conducted through PubMed, The Cochrane Library, Embase, Web of Science, the China National Knowledge Infrastructure and China Biology Medicine disc for studies on AI, integrated CT evaluation of pulmonary nodules. By searching PubMed, The Cochrane Library, Embase, Web of Science, the China National Knowledge Infrastructure and China Biology Medicine disc to collect relevant studies on AI combined with CT evaluation of benign and malignant pulmonary nodules. According to the established inclusion and exclusion criteria, the QUADAS-2 scale was used to evaluate the methodological quality of literature that met the requirements. RevMan5.3 and Stata18.0 were used for data integration and analysis, comparing pooled diagnostic metrics between AI systems and radiologists, specificity, sensitivity, negative likelihood ratio, positive likelihood ratio, and diagnostic odds ratio. Inter-group differences were statistically tested using Z-tests. Summary receiver operating characteristic curves were constructed to visualize diagnostic accuracy, with heterogeneity explored through statistics. Subgroup analyses stratified by AI systems were conducted to compare the differences in sensitivity, specificity, and other indicators between different systems. Results Pooled metrics demonstrated the following comparisons (AI-assisted CT vs. radiologist interpretation): sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio, and summary receiver operating characteristic curve area under the curve. Statistical comparisons revealed no significant inter-group differences (all P > 0.05). Conclusion AI-assisted CT demonstrates diagnostic efficacy comparable to conventional radiologist interpretation in pulmonary nodule characterization. The marginally higher sensitivity suggests potential utility for reducing missed diagnoses, though large-scale clinical validation remains warranted.

关键词

计算机断层扫描 / 人工智能 / 肺结节 / 人工智能辅助诊断 / 深度学习 / Meta分析

Key words

computed tomography / artificial intelligence / pulmonary nodule / artificial intelligence-assisted diagnosis / deep learning / Meta-analysis

参考文献

[1] 王培宇， 黄祺， 王少东， 等. 《全球癌症统计数据2022》要点解读[J]. 中国胸心血管外科临床杂志， 2024， 31(7)：933-954.

[2] THAI A A, SOLOMON B J, SEQUIST L V, et al. Lung cancer[J]. The lancet, 2021, 398(10299): 535-554.

[3] MAZZONE P J, LAM L. Evaluating the patient with a pulmonary nodule: a review[J]. JAMA, 2022, 327(3): 264-273.

[4] LARICI A R, FARCHIONE A, FRANCHI P, et al. Lung nodules: size still matters[J]. European respiratory review, 2017, 26(146): 170025.

[5] 陆晓骞， 宗凌燕， 沈琪. 正电子发射断层扫描/计算机断层扫描(PET/CT)成像在肿瘤筛查中的应用进展[J]. 中国辐射卫生， 2023， 32(1)：66—69， 74.

[6] LIU H, CHEN R H, TONG C, et al. MRI versus CT for the detection of pulmonary nodules: a meta—analysis[J]. Medicine, 2021, 100(42): e27270.

[7] VENKADESH K V, SETIO A A A, SCHREUDER A, et al. Deep learning for malignancy risk estimation of pulmonary nodules detected at low—dose screening CT[J]. Radiology, 2021, 300(2): 438-447.

[8] 董来东， 黄果. 基于CT影像的人工智能辅助诊断系统对4771例肺癌诊断价值的系统评价与Meta分析[J]. 中国胸心血管外科临床杂志， 2021， 28(10)：1183-1191.

[9] 程顺达， 程颖， 孙士江. 基于机器学习的肿瘤智能辅助诊断方法[J]. 电子科技， 2022， 35(5)：56-59.

[10] DU W, HE B, LUO X J, et al. Diagnostic value of artificial intelligence based on CT image in benign and malignant pulmonary nodules[J]. Journal of oncology, 2022, 2022: 5818423.

[11] 耿然. 基于深度学习的人工智能影像辅助诊断系统对肺结节的诊断效能评价[J]. 新乡医学院学报， 2022， 39(11)：1031-1035.

[12] 贺文俊， 张周兵， 胡永松， 等. 基于人工智能的CT定量分析在肺结节患者中的诊断价值及效能ROC曲线研究[J]. 现代仪器与医疗， 2023， 29(3)：89-92.

[13] 刘亚龙， 安小康， 郭明杰， 等. 基于人工智能技术的肺部CT扫描在肺结节筛查中的应用分析[J]. 医学理论与实践， 2023， 36(11)：1918-1919.

[14] 王彬， 顾文权， 蔡孙美， 等. 基于人工智能的胸部CT肺结节检出及良恶性诊断效能评估[J]. 影像研究与医学应用， 2023， 7(5)：95-97.

[15] FAN W, LIU H T, ZHANG Y, et al. Diagnostic value of artificial intelligence based on computed tomography (CT) density in benign and malignant pulmonary nodules: a retrospective investigation[J]. Peer J, 2024, 12: e16577.

[16] PAN W, FANG X T, ZANG Z Y, et al. Diagnostic efficiency of artificial intelligence for pulmonary nodules based on CT scans[J]. American journal of translational research, 2023, 15(5): 3318-3325.

[17] 袁飞， 钟临锋， 陈兴， 等. 肺结节人工智能评估危险度辅助人工阅片的应用价值[J]. 分子影像学杂志， 2024， 47(2)：132-137.

[18] 钟丽茹， 罗娜， 贺露瑶. 多个人工智能辅助诊断系统对肺结节的诊断效能比较[J]. 影像研究与医学应用， 2024， 8(17)：26-29.

[19] 沈小钦， 梁红， 朱晓琼. CT图像人工智能辅助诊断系统鉴别肺结节良恶性的价值[J]. 中国辐射卫生， 2024， 33(5)：578-583.

[20] BERNHARDT M, JONES C, GLOCKER B. Investigating underdiagnosis of AI algorithms in the presence of multiple sources of dataset bias[EB/OL]. (2022—01—19)[2024—07—30]. arXiv: 2201.07856. https://arxiv.org/abs/2201.07856.

[21] 关松林. 机器学习在疾病诊断中的应用[D]. 重庆：重庆大学， 2019.

基金资助

甘肃省自然科学基金资助项目(23JRRA1292)