摘要
目的通过对人工智能的染色体图像扫描处理系统、Metafer 4染色体扫描分析系统和人工分析双着丝粒染色体(dic)之间的结果比较,探究人工智能新技术检测dic估算生物剂量的可行性。方法通过采集健康人肘静脉血,经60Co离体照射,常规方法制备染色体样本,使用AI染色体图像扫描处理系统和Metafer 4染色体扫描分析系统对玻片进行扫描及自动分析,并对结果进行人工分析及确认。结果AI染色体图像扫描处理系统和Metafer 4染色体扫描分析系统的分析细胞数较为接近,但AI系统扫描速度为4.5 s/张,显著快于Metafer 4系统的7.3 s/张(t=-6.19,P<0.05)。AI系统在置信率0.7时真阳性率为96.7%,假阳性率为6.5%,显著优于Metafer 4系统的真阳性率(45.4%~54.5%)和假阳性率(22.2%~29.2%)(P均<0.05)。在生物剂量估算中,Metafer 4系统使用自动分析剂量-效应曲线的偏差≤±10%,AI系统因使用人工剂量-效应曲线偏差≤±15%,但两者估算剂量与人工分析差异均无统计学意义(P>0.05)。结论AI染色体图像扫描处理系统和Metafer 4染色体扫描分析系统均能极大提升染色体畸变分析速度。但AI染色体图像扫描处理系统的扫描速度、真阳性率、假阳性率均优于Metafer 4染色体扫描分析系统,更适用于大规模辐射事故下快速高通量生物剂量估算。
Objective To compare the results obtained from an artificial intelligence(AI)-based chromosome image scanning and processing system,the Metafer 4 chromosome scanning and analysis system,and manual analysis of dicentric chromosomes,and to explore the feasibility of applying AI technology for dicentric chromosome detection and biological dose estimation.Methods Healthy human elbow vein blood was collected and subjected to 60Co in vitro irradiation.Chromosome samples were prepared using conventional methods.The slides were scanned and automatically analyzed using the AIbased system and the Metafer 4 system.The results were manually analyzed and confirmed.Results The number of cells was comparable between the AI-based system and the Metafer 4 system.However,the scanning speed of the AI-based system was 4.5 seconds per image,which was significantly faster than the 7.3 seconds per image of the Metafer 4 system(t=−6.19,P<0.05).At a confidence level of 0.7,the AI-based system demonstrated a true positive rate of 96.7%and a false positive rate of 6.5%,which were significantly better than the true positive rate(45.4%-54.5%)and false positive rate(22.2%-29.2%)of the Metafer 4 system(all P<0.05).In the biological dose estimation,the deviation of the dose-response curve was≤±10%in the automatic analysis using the Metafer 4 system.Due to the use of the manual dose-response curve,the deviation of the AI-based System was≤±15%.However,there were no significant differences in the estimated doses when the two systems were compared with the manual analysis(P>0.05).Conclusion Both the AI-based chromosome image scanning and processing system and the Metafer 4 chromosome scanning and analysis system greatly improved the analysis speed of chromosome aberrations.However,the scanning speed,true positive rate,and false positive rate of the AIbased system were superior to those of the Metafer 4 system.Therefore,the AI-based system is more suitable for rapid and high-throughput biological dose estimation in large-scale radiation accidents.
作者
冯骏超
刘畅
刘玉龙
李杰
高宇
FENG Junchao;LIU Chang;LIU Yulong;LI Jie;GAO Yu(The Second Affiliated Hospital of Soochow University,Suzhou 215004,China;The Third People’s Hospital of Henan Province,Henan Occupational Disease Hospital,Henan Key Laboratory of Medicine on Radiobiology and Epidemiology,Zhengzhou 450052,China)
出处
《中国辐射卫生》
2025年第4期571-577,共7页
Chinese Journal of Radiological Health
基金
国家自然科学基金项目(U2267220)
国家自然科学基金项目(12405392)
江苏省自然科学基金青年项目(BK20240358)
中核集团“青年英才”科研项目(启明星)
省部共建放射医学与辐射防护国家重点实验室开放课题(GZK12020042)
河南省辐射生物与流行病学医学重点实验室课题(HNRBEKF202401)
省部共建放射医学与辐射防护国家重点实验室内部协作课题(GZN-1202201)
苏大附二院科研预研基金项目(SDFEYBS2304)。
关键词
AI人工智能
染色体畸变自动分析
高通量
生物剂量估算
Artificial intelligence
Automatic chromosome aberration analysis
High throughput
Biological dose estimation
