摘要
目的 探讨人工智能(AI)读片对评估DR及预测慢性并发症的价值。方法 选取2017年1月至2018年12月于我院内分泌科治疗的DR筛查患者2260例,采用无散瞳眼底照相结合AI读片进行筛查,评估AI读片的诊断效能。根据眼底分析结果将所有患者分为单纯T2DM组(T2DM,n=1803)、非增生型糖尿病视网膜病变组(NPDR,n=291)、增生型糖尿病视网膜病变组(PDR,n=166),比较各组血糖谱和DM慢性并发症情况,评估PDR、NPDR与DM慢性并发症的关系。结果 AI系统在T2DM人群中筛查DR的灵敏度为93.87%,特异度为96.95%,人工系统诊断DR的灵敏度为95.84%,特异度为96.01%,κ值为0.825,二者诊断效能相似(P>0.05)。NPDR、PDR组FPG、HbA1c高于T2DM组(P<0.05)。NPDR、PDR组DKD、糖尿病足病(DF)、糖尿病周围神经病变(DPN)发生率高于T2DM组(P<0.05),PDR组DKD、DF发生率高于NPDR组(P<0.05)。Spearman相关分析显示,NPDR、PDR与DKD、DF、DPN呈正相关(P<0.05)。结论AI读片可用于T2DM人群DR筛查,有效预测其他慢性并发症发生。
Objective retinopathy(DR)and predicting chronic complications.in Endocrinology Department in our hospital were enrolled in this study from January 2017 to December2018. DR was screened by non-mydriatic fundus photography combined with artificial intelligence reading.The diagnostic efficiency of AI film reading was evaluated. All the participants were divided into three groups:patients with simple T2DM group(T2DM,n=1803),patients with non-proliferative DR group(NPDR,n=291)and patients with proliferative DR group(PDR,n=166). The blood glucose profiles and the prevalence of diabetes chronic complications including diabetic kidney disease(DKD),peripheral neuropathy and diabetic foot were compared among different groups. The associations between PDR,NPDR and diabetic chronic complications were evaluated.DR were 93. 87% and 96. 95% in T2DM population. The sensitivity and specificity of artificial system in diagnosing DR were 95. 84% and 96. 01%. The κ value was 0. 825,indicating similar diagnostic efficacy between the two systems(P>0. 05). The levels of FPG and Hb A1c were higher in both NPDR and PDR group than in T2DM group(P<0. 05). The incidences of DKD,diabetic foot and peripheral neuropathy were higher in NPDR and PDR group than in T2DM group(P<0. 05). The incidences of DKD and diabetic foot were higher in PDR group than in NPDR group(P<0. 05). Spearman correlation analysis showed that NPDR and PDR were positively correlated with the occurrence of DKD,diabetic foot and peripheral neuropathy(P<0. 05).ConclusionAI film reading can be used in screening DR in patients with T2DM,and can effectively predict other diabetic chronic complications.
作者
李政章
顾建芬
梁伟
张咪
莫小活
LI Zhengzhang;GU Jianfen;LIANG Wei(Department of Endocrinology,The University of HongKong-Shenzhen Hospital,Shenzhen 518000,China)
出处
《中国糖尿病杂志》
CAS
CSCD
北大核心
2022年第11期817-820,共4页
Chinese Journal of Diabetes
基金
中国微循环学会糖尿病基金(TW-2018H006)。
关键词
糖尿病
2型
视网膜病变
人工智能读片
并发症
应用价值
Diabetes mellitus
type 2
Retinopathy
Artificial intelligence film reading
Complications
Application value