摘要
目的:探讨基于深度学习人工智能在肺结节定性诊断中的临床应用价值.方法:回顾性分析 2020 年 1 月—2022 年 12 月东莞市松山湖中心医院符合纳入标准的 230 例患者的 325 个肺结节的资料,根据阅片方式分为A组(双人阅片),B组(AI 单独阅片),C组(双人+AI综合阅片),以病理为金标准,统计和比较各组对 325 个肺结节检出的阅片时间、灵敏度、特异度、阳性预测值、阴性预测值、准确率.结果:A组阅片时间(14.37±2.12)min,B组阅片时间(1.34±0.12)min,C组阅片时间(8.34±1.26)min,B组阅片时间短于A组和C组,且C组阅片时间短于A组,差异具有统计学意义(P＜0.05).A组诊断的灵敏度、特异度、阳性预测值、阴性预测值、准确率分别为 85.97%、36.17%、88.85%、30.36%、78.76%,B组分别为 91.01%、42.55%、90.36%、44.44%、84.00%,C组分别为90.29%、55.32%、92.27%、49.27%、85.23%,各组差异有统计学意义(P＜0.05).结论:AI辅助医师阅片可以缩短阅片时间,有效提高肺结节诊断的灵敏度、特异度和准确率.
Abstract
Objective To explore the clinical application of deep learning artificial intelligence in the qualitative diagnosis of pulmonary nodules.Methods A retrospective analysis was conducted on the data of 325 pulmonary nodules from 230 patients who met the inclusion criteria at Dongguan Songshan Lake Central Hospital from January 2020 to December 2022.They were divided into group A(double reading),group B(AI reading alone),and group C(double reading+AI comprehensive reading)according to the reading method.Using pathology as the gold standard,the reading time,sensitivity,specificity,positive predictive value,negative predictive value,and accuracy of 325 lung nodules detected in each group were counted and compared.Results The film reading time in Group A was(14.37±2.12)minutes,the film reading time in Group B was(1.34±0.12)minutes,and the film reading time in Group C was(8.34±1.26)minutes.The film reading time of Group B was statistically shorter than that of Group A and Group C,and the film reading time of Group C was statistically shorter than that of Group A,with a statistically significant difference(P＜0.05).The sensitivity,specificity,positive predictive value,negative predictive value,and accuracy of Group A diagnosis were 85.97%,36.17%,88.85%,30.36%,and 78.76%,respectively,and the indicators of Group B diagnosis were 91.01%,42.55%,90.36%,44.44%,and 84.00%,respectively,and indicators of group C diagnosis were 90.29%,55.32%,92.27%,49.27%,and 85.23%,respectively,with statistical significance(P＜0.05).Conclusion AI assisted radiologists can shorten the reading time and effectively improve the sensitivity,specificity,and accuracy of pulmonary nodule diagnosis.
基金项目
广东省东莞市社会发展科技项目(20231800904182)
维普
万方数据