摘要
目的采用经过机器学习的肺结节检测分析系统(0.4.2.1版本)和放射诊断医师两种方式,分别对有病理结果的肺结核瘤病灶进行判读,探讨当前人工智能在肺结核瘤病灶中的应用价值。方法2014年4月至2018年10月期间就诊于北京大学深圳医院,行胸部高分辨CT平扫且经病理证实,共11例作为研究对象。(1)共纳入的11处病灶,经过人工智能及传统放射医师两种方法对CT图像进行判读,分为A、B两组。A组为AI组,B组采用2名胸部影像诊断方向的放射诊断医师。统计分析A、B两组对肺结核瘤病灶的检出情况。(2)对CT图像对应的病灶进行良恶性判断,其中σ-Discover/Lung肺结节检测分析系统(0.4.2.1版本)以大于50%作为高恶性判断标准。结果本研究共纳入11例肺结核患者,共11处病灶。A组一共检出11处病灶,其中4处结核病灶被诊断为高恶性,漏检0处;B组一共检出10处病灶,其中1处误判,3处病灶未定性,漏检1处病灶。肺结节检测分析系统(0.4.2.1版本)判断一共4处病灶恶性概率大于50%,36.36%的肺结核病灶分析为高恶性概率。结论肺结节检测分析系统对肺结核瘤病灶的检出具有较好的能力,对肺结核瘤病灶的误判率有所减低但是仍高于放射诊断医师组。
Objective To explore the application value of machine learning in diagnosising pathologically confirmed pulmonary tuberculosis lesions.Methods A retrospective analysis was conducted to analysis the computed tomography images of patients with tuberculosis patients from April 2014 to October 2018 in Peking University Shenzhen hospital.CT images were retrospectively analyzed by two methods and divided into two groups,A and B.The group A was analyzed byσ-Discover/Lung pulmonary nodule detection and analysis system(0.4.2.1)(more than 50%as a high malignant criterion).The group B was analyzed by two experienced chest radiologists.Results A total of 11 patients with pulmonary tuberculosis were enrolled in the study,with a total of 11 lesions.All lesions were detected in the group A.Ten lesions were detected in the group B and one lesion was missed.The detection rate of lesions in the group A was 100%,and the rate of false positive diagnosis was 36.36%.The detection rate of lesions in the group B was 90.91%,the rate of false positive diagnosis was 10.00%,and the rate of missed diagnosis was 9.09%.The pulmonary nodule detection and analysis system judged that the total malignant probability of 4 lesions was greater than 50%,and 36.36%of benign lung lesions were analyzed for high malignant probability.Conclusion The pulmonary nodule detection and analysis system has a good ability to detect the lesions of pulmonary tuberculosis patients.The current accuracy of diagnosis of benign lesions is reduced but still higher than radiologists.
作者
闫明艳
李娇
戚玉龙
刘维湘
成官迅
YAN Ming-yan;LI Jiao;QI Yu-long;LIU Wei-xiang;CHENG Guan-xun(Department of Medical Image,Peking University Shenzhen Hospital,Shenzhen 518036,Guangdong Province,China;School of Biomedical Engineering,Health Science Center,Shenzhen University,Shenzhen518060,Guangdong Province,China)
出处
《中国CT和MRI杂志》
2021年第8期69-71,共3页
Chinese Journal of CT and MRI
基金
深圳市科创委基金项目(JCYJ20160422113119640)。
关键词
人工智能
机器学习
肺结节检测系统
肺结核瘤
检出率
良恶性
Machine Learning
Pulmonary Nodule Detection and Analysis System
Tuberculosis
Detection Rate
the Rate of Missed Diagnosis
Benign and Malignant