期刊文献+

深度学习辅助诊断系统在胸片的应用研究:气胸及肺结节检测 被引量:4

A Study Using Deep Learning-Based Computer Aided Diagnostic System with Chest Radiographs-Pneumothorax and Pulmonary Nodules Detection
原文传递
导出
摘要 目的评估基于深度学习的计算机辅助诊断系统(DL-CAD)的X线胸片气胸及肺结节检出效能,以及DL-CAD辅助对医师诊断效能的影响。方法搜集经CT扫描证实的80例气胸及100例肺结节患者的X线胸片。将所有胸片上传至DL-CAD以及PACS系统。气胸及肺结节的胸片各由两名低年资放射医师(小于5年诊断经验)及两名高年资医师(大于10年诊断经验)分别采取独立阅片、DL-CAD辅助阅片方式对所有胸片进行判读,共有八名医师参与。记录每例患者的判读所需时间,计算两种阅片方式的敏感度、假阳性数,利用单因素方差分析评估不同级别医师利用两种阅片方式判读所需时间的差别,使用χ^(2)检验评估不同阅片方式的肺结节诊断的敏感度、假阳性数。结果在肺结节、气胸的诊断效率方面,DL-CAD均在≤40 s的时间内完了检测,检测用时低于低年资医师,与高年资医师相当。相比于独立阅片组,利用DL-CAD辅助阅片的高、低年资医师的平均诊断气胸所用时间分别减少约52.28%、43.02%,平均诊断肺结节所用时间分别减少42.67%、34.92%。在诊断质量方面,DL-CAD对气胸检出的敏感度为97.5%(78/80),与高、低年资放射科医师的敏感度均无显著性差异。在DL-CAD辅助阅片下,两组医师的诊断敏感度均为100%。在诊断肺结节方面,DL-CAD检出154个真阳性结节,显著高于独立阅片的低年资医师(127个,P=0.014),略低于高年资医师(165个,P=0.299)。利用DL-CAD辅助阅片的低、高年资医师的敏感度分别为65.45%、76.02%,高出独立阅片的低、高年资医师13.82%、8.95%,假阳性数也更低,分别为28个、14个,提示DL-CAD对高、低年资医师在肺结节检测上均有帮助,且降低了不同经验医师的诊断差异。结论DL-CAD系统具有良好的胸片气胸及肺结节检测能力,利用DL-CAD辅助诊断可以显著提高放射医师的诊断质量及诊断效率,降低不同经验医师的诊断差异。 Objective To evaluate the ability of deep learning-based computer aided diagnostic system(DL-CAD)to detect pneumothorax and pulmonary nodules on chest radiographs,and the impact of DL-CAD assistance on the diagnostic effectiveness of radiologists.Methods Chest radiographs of 80 cases of pneumothorax and 100 cases of pulmonary nodules confirmed by CT examination were collected and uploaded to DL-CAD and PACS system.Both pneumothorax cases and pulmonary nodule cases were interpreted by two junior radiologists(less than 5 years of diagnostic experience)and two senior radiologists(more than 10 years of diagnostic experience)by means of independent reading and DL-CAD assisted reading with a total of 8 radiologists participating.The time taken for each patient’s interpretation was recorded,and the sensitivity and false positive number of the two-interpretation means were calculated.The difference of the time taken by all 8 radiologists was evaluated by one-way ANOVA.The sensitivity and false positive number of different interpretation means were evaluated byχ^(2) test.Results In terms of the diagnostic efficiency,DL-CAD completed pulmonary nodules or pneumothorax in less than 40 seconds.This time taken was lower than that of junior doctors and equivalent to that of senior doctors.Compared with the independent group,the average time consumption of senior and junior radiologists for completing a diagnosis was reduced by approximately 52.28%and 43.02%for pneumothorax,and 42.67%and 34.92%for pulmonary nodules.In terms of diagnosis efficacy,the pneumothorax detection sensitivity of DL-CAD was 97.5%(78/80)and there was no significant difference among DL-CAD,senior radiologists and junior radiologists.With the aid of DL-CAD,the diagnostic sensitivity of both radiologist groups was 100%.For the diagnosis of pulmonary nodules,154 true positive nodules were detected by DL-CAD,which were significantly higher than that of junior radiologists(127,P=0.014)and slightly lower than that of senior radiologists(165,P=0.299).The sensitivity of junior radiologists and senior radiologists in assisted reading was 65.45%and 76.02%respectively,which were 13.82%and 8.95%higher than those of independent reading radiologists.The false positive number of junior radiologists and senior radiologists was also lower,with a value of 28 and 14 respectively.It is suggested that DL-CAD assistance was helpful for both senior and junior radiologists in the detection of pulmonary nodules,as well as in reducing the diagnostic discrepancy between radiologists with different diagnostic experience.Conclusion The DL-CAD system has a good ability to detect pneumothorax and pulmonary nodules.DL-CAD assisted diagnosis can significantly improve the diagnostic efficacy and efficiency of radiologists,and reduce the diagnostic discrepancy among radiologists with different diagnostic experience.
作者 魏一娟 潘宁 陈岩 吕培杰 高剑波 WEI Yijuan;PAN Ning;CHEN Yan(Department of Radiology,The First Affiliated Hospital of Zhengzhou University,Zhengzhou,Henan Province 450052,P.R.China)
出处 《临床放射学杂志》 北大核心 2021年第2期252-257,共6页 Journal of Clinical Radiology
关键词 X线摄影 气胸 肺结节 计算机辅助诊断 Radiography Pneumothorax Pulmonary nodule Computer-aided diagnosis
作者简介 通讯作者:高剑波
  • 相关文献

参考文献13

二级参考文献65

共引文献314

同被引文献49

引证文献4

二级引证文献8

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部