多光谱眼底疾病人工智能识别系统对视网膜静脉阻塞的识别研究  

Research on Recognition of Retinal Vein Occlusion by Artificial Intelligence Recognition System for Multispectral Fundus Diseases

作　　者：马菲妍 张彩霞 冬雪川 MA Feiyan;ZHANG Caixia;DONG Xuechuan(Department of Ophthalmology,Second Hospital of Hebei Medical University,Shijiazhuang City,Hebei Province 050005;不详)

机构地区：[1]河北医科大学第二医院眼科,河北省石家庄市050005 [2]深圳市新产业眼科新技术有限公司

出　　处：《医学理论与实践》2022年第2期202-205,共4页The Journal of Medical Theory and Practice

基　　金：深圳市科技研发资金—深科技创新[2019]33号(JSGG20180507182010237);河北省2020年度医学科学研究课题计划(20200069)。

摘　　要：目的:探索基于多光谱眼底图像开发的人工智能在视网膜静脉阻塞疾病的早期诊断效能。方法:采用诊断试验的研究方法,以150张经过专家标定的可能患有不同类型视网膜静脉阻塞的多光谱眼底图像作为阅片标注的参考标准,比较AI组、高年资眼科医师组、低年资眼科医师组及心血管内科医师组的诊断一致性和阅片速度。不同阅片者的阅片结果和参考标准的比较采用加权kappa系数进行评价,AI组和医师组的比较以Kendall系数进行评价。AI系统和各医师组的单张平均阅片时间比较采用重复测量方差Bonferroni法分析。结果:对于各组的阅片一致性,AI组和高年资眼科医师组、低年资眼科医师组、心血管内科医师组相比诊断结果基本一致(P<0.01),与AI组比较,心血管内科医师组和低年资眼科医师组协调系数依次显著低于高年资眼科医师组(P<0.01),AI组的阅片时间显著少于各医师组。结论:通过人工智能和多光谱眼底成像技术的结合能够提升对眼底静脉阻塞类疾病的诊断效能,避免临床工作中的误诊和漏诊。Objective:To explore the early diagnosis efficacy of artificial intelligence developed based on multispectral fundus images in the early diagnosis of retinal vein occlusion diseases.Methods:Using diagnostic test research methods,150 multispectral fundus images that may have different types of retinal vein occlusions calibrated by experts were used as the reference standard for reading and annotation.To compared the consistency of diagnosis and speed of reading pictures between the AI group,senior physician group,low-seniority physician group,cardiovascular physician group.The comparison between the reading results of different readers and the reference standard was evaluated by the weighted kappa coefficient,and the comparison between the AI group and the physician group was evaluated by the Kendall coefficient.The comparison of the average reading time of a single sheet between the AI system and each physician group was analyzed by the repeated measures variance Bonferroni method.Results:For the consistency of the reading of each group,the diagnosis results of the AI group and the senior physician group,the low-seniority physician group,and the cardiovascular physician group were basically the same(P<0.01).Compared with the AI group,the coordination coefficients of the cardiovascular physician group and the low-seniority physician group were significantly lower than those of the senior physician group(P<0.01),and the reading time of the AI group was significantly less than that of each physician group.Conclusion:The combination of artificial intelligence and multispectral fundus imaging technology can improve the diagnostic efficiency of fundus vein occlusion diseases,and avoid misdiagnosis and missed diagnosis in clinical work.

关 键 词：多光谱眼底成像 人工智能 视网膜静脉阻塞 早期诊断 

分 类 号：R445[医药卫生—影像医学与核医学]