【摘要】 目的评估基于人工智能肺结核筛查技术在基层医院肺结核影像诊断中的应用价值,提高基层医院肺结核筛查水平。方法收集2019年3月至2020年7月在喀什地区第一人民医院就诊的11 616例患者的胸部X线图像及临床资料,其中10 399例作为训练集,1217例作为测试集。训练集中诊断为肺结核的病例为535例,正常病例为6972例,非肺结核异常病例为2892例。测试集中诊断为肺结核的病例为305例,正常病例为840例,非肺结核异常病例为72例。对比人工智能诊断模型与放射科医生诊断效能;对比有AI辅助的放射科医生和无AI辅助的放射科医生诊断效能。结果①人工智能诊断灵敏度为86.8%,高于当地初级放射医生诊断灵敏度63.3%,两种诊断方法灵敏度差异有统计学意义(P<0.05)。②在AI的辅助诊断下,当地初级放射科医生的平均诊断时间由(37.4±1.2)s降低为(14.0±4.3)s,两者差异有统计学意义(P<0.05)。结论基于人工智能肺结核筛查技术可提高肺结核影像诊断效率和效能,有利于偏远地区或基层医院进行肺结核筛查。更多还原

【Abstract】 Objective To evaluate the performance of artificial intelligence system in detecting tuberculosis (TB) in primary hospital,and improve the level of tuberculosis screening in primary hospitals.Methods The chest X-ray images and clinical data of 11616 patients in the first people’s Hospital of Kashgar region from March 2019 to July 2020 were collected,among which 10 399 cases were selected as the training set and 1217 cases were selected as the test set.There were 535 TB cases,6972 normal cases and 2892 non-tuberculosis abnormal cases in the training set,and there were 305 TB cases,840 normal cases and 72 cases of non-tuberculosis abnormal cases in the test set.① Comparing TB detection performances of AI system and local radiologists.② Comparing the diagnostic efficiency of radiologists with AI assistance and radiologists without AI assistance.Results ① The sensitivity of intelligent diagnosis was 86.8%,which was higher than that of local junior radiologists (63.3%).There was significant difference between the two diagnostic methods (P<0.05).② With AI-assisted diagnosis,the average diagnosis time of local primary radiologists decreased from (37.4±1.2) s to (14.0±4.3) s,The difference was statistically significant (P<0.05).Conclusion Based on artificial intelligence tuberculosis screening technology,the efficiency and effectiveness of tuberculosis imaging diagnosis can be improved,and it is helpful to the tuberculosis screening in remote areas or grass-roots hospitals.更多还原