摘要
目的研究肺结节人工智能辅助诊断系统在医院门诊中实际临床应用情况以及对于恶性肺结节的诊断准确率分析。方法回顾性收集本院2018年03月-2020年10月之间肺结节人工智能检测系统的使用情况,统计使用AI系统分析的病例数量和肺结节检测数量,以及医生采纳率和诊断用时。对于在本院进行病理检测或手术的病人,随机匹配相同数量的良性结节病例,收集AI系统对这些结节的恶性预测概率,通过受试者工作特征曲线(receiver operating characteristic curve,ROC curve)分析AI系统对于恶性结节的预测准确率。结果2018年03月-2020年10月本院利用AI系统共分析了14510例患者的胸部平扫CT影像,其中在13539例中检测到有肺结节。AI系统总共检测出151782个疑似肺结节,平均每个CT影像发现10.46个,其中有风险的结节(AI预测风险概率大于30%)为7718个,平均每个CT影像发现0.53个。在AI的辅助下,医生的读片时间缩短了68%。共计有69例患者在本院进行了活检病理或手术检查,ROC分析结果表明AI系统在风险概率为47.3%的阈值下的敏感性为100%,特异性为83.13%,对于恶性肺结节诊断的AUC为0.907(95%置信区间,0.845~0.950)。结论肺结节人工智能辅助诊断系统在实际临床应用中可以有效辅助肺结节的检测,且在肺结节良恶性诊断上具有较好的准确率。
Objective To study the real clinical application of AI-aided diagnose system for pulmonary nodule,and analyzed its accuracy to diagnose malignant pulmonary nodule.Methods Retrospectively collect the usage of AI-aided diagnose system from Mar2018 and Oct 2020,analyzed the patient number,detected pulmonary nodule count,doctor’s accept rate and diagnose time.Randomly match same number cases with benign pulmonary nodules for the patients who had undertaken pathological examination or operation,collect probabilities of malignant predictions from AI-aided diagnose system,analyzed the system’s accuracy of malignant predictions by receiver operating characteristic curve(ROC curve).Results By using AI-aided system,analyzed chest routine scan CT images of14,510 patients from Mar 2018 and Oct 2020,of which 13,539 cases has been detected pulmonary nodules.The system had detected151,782 suspected pulmonary nodules(10.46 nodules per CT image),7,718 nodules with risk(predicted probability>30%,0.53 nodules per CT image).In the assistance of AI,doctor’s’reading time has been reduced 68%.From biopsy pathology and operation result of 69patients in total,the ROC result indicated that AI-aided system has 100%sensitivity and 83.13%specificity at risk threshold 47.3%,and0.907 AUC for malignant pulmonary nodule detection(95%confidence interval:0.845~0.950).Conclusion AI-aided diagnose system have effectively supported to detect pulmonary nodule,and have a good accuracy on diagnosing malignant pulmonary nodule.
作者
邹振宇
杨建丽
姚娟
查永将
印宏坤
ZOU Zhenyu;YANG Jianli;YAO Juan;ZHA Yongjiang;YIN Hongkun(Changji Branch,The First Affiliated Hospital of Xinjiang Medical University,Changji,831100,Xinjiang,China;Infer Vision,Beijing,100000,China)
出处
《新疆医学》
2022年第5期524-526,537,共4页
Xinjiang Medical Journal
基金
昌吉州科学研究与技术开发计划项目(项目编号:2019S02-16)
关键词
肺结节
人工智能辅助诊断系统
预测恶性概率
pulmonary nodule
AI-aided diagnose system
predict malignant probability
作者简介
邹振宇,男,主任医师,研究方向:肺部疾病及神经系统影像研究。