目的：探讨基于多模态眼底影像定量特征构建的预测模型对评估非增殖性糖尿病视网膜病变(NPDR)进展风险的效能。

方法： 本研究为单中心回顾性队列研究。选取2021年1月至2023年12月武汉市第九医院确诊的中重度NPDR患者。采集基线期彩色眼底照相(CFP)、光学相干断层扫描(OCT)及眼底自发荧光(FAF)影像，提取并量化微动脉瘤计数、硬性渗出面积、黄斑中心凹视网膜厚度、视网膜神经纤维层厚度及强荧光点总面积。定期随访，主要终点为随访期内进展至需治疗的增殖性DR(PDR)或临床有意义的糖尿病性黄斑水肿(DME)，根据在随访期内是否发生终点事件分为进展组与非进展组。结合临床意义及组间比较结果，将候选影像定量参数纳入多因素Logistic回归构建预测模型，以Nomogram可视化，通过ROC曲线及校准曲线评估模型区分度与校准度。

结果：患者共300例300眼完成随访，进展组93例，其中男47例，女46例，平均年龄59.02±8.21岁； 非进展组207例，其中男105例，女102例，平均年龄58.01±8.10岁； 多因素Logistic回归显示，FAF强荧光点总面积(OR=1.580，95%CI：1.061-1.718，P=0.002)、OCT-视网膜神经纤维层厚度(OR=1.520，95%CI：1.307-6.212，P=0.013)及CFP微动脉瘤计数(OR=1.459，95%CI：1.089-2.095，P=0.015)为DR进展危险因素。Nomogram预测随访期内进展风险的AUC为0.889(95%CI：0.826-0.952)，校准曲线显示预测概率与实际风险一致性良好。

结论：整合CFP、OCT及FAF定量影像特征的预测模型能有效识别NPDR高进展风险人群，效能良好，可为个体化风险管理提供量化工具。

METHODS:A single-center retrospective cohort study was conducted. Patients diagnosed with moderate to severe NPDR between January 2021 and December 2023 in Wuhan Ninth Hospital were selected. Baseline color fundus photography(CFP), optical coherence tomography(OCT), and fundus autofluorescence(FAF)images were collected. Microaneurysm count, hard exudate area, central retinal thickness, retinal nerve fiber layer thickness, and total area of hyperautofluorescent spots were extracted and quantified. All patients were followed up regularly. The primary endpoint was the progression to treatment-requiring proliferative diabetic retinopathy(PDR)or clinically significant diabetic macular edema(DME)during the follow-up period. According to the occurrence of endpoint events during follow-up, the patients were divided into the progression group and the non-progression group. Combined with clinical significance and inter-group comparison results, candidate quantitative imaging parameters were included in the multivariate logistic regression analysis to construct a prediction model. The model was visualized using a Nomogram. The discriminative ability and calibration degree of the model were evaluated via receiver operating characteristic(ROC)curves and calibration curves.

RESULTS:A total of 300 patients(300 eyes)completed follow-up, with 93 in the progression group, including 47 males and 46 females, with an average age of 59.02±8.21 y. There were 207 cases in the non-progression group, including 105 males and 102 females, with an average age of 58.01±8.10 y. Multivariate logistic regression showed that the total area of FAF hyperautofluorescent spots(OR=1.580, 95%CI: 1.061-1.718, P=0.002),OCT retinal nerve fiber layer thickness(OR=1.520, 95%CI: 1.307-6.212, P=0.013), and CFP microaneurysm count(OR=1.459, 95%CI: 1.089-2.095, P=0.015)were risk factors for DR progression. The AUC of the Nomogram for predicting progression risk during the follow-up period was 0.889(95%CI: 0.826-0.952), and the calibration curve showed good consistency between predicted probability and actual risk.

CONCLUSION: The prediction model integrating quantitative imaging features of CFP, OCT and FAF can effectively identify NPDR patients with high progression risk and exhibits good predictive efficacy, which can serve as a quantitative tool for individualized clinical risk management.