Background/aims To investigate the comprehensive prediction ability for cognitive impairment in a general elder population using the combination of the multimodal ophthalmic imaging and artificial neural networks. Methods Patients with cognitive impairment and cognitively healthy individuals were recruited. All subjects underwent medical history, blood pressure measurement, the Montreal Cognitive Assessment, medical optometry, intraocular pressure and custom-built multimodal ophthalmic imaging, which integrated pupillary light reaction, multispectral imaging, laser speckle contrast imaging and retinal oximetry. Multidimensional parameters were analysed by Student’s t-test. Logistic regression analysis and back-propagation neural network (BPNN) were used to identify the predictive capability for cognitive impairment. Results This study included 104 cognitive impairment patients (61.5% female; mean (SD) age, 68.3 (9.4) years), and 94 cognitively healthy age-matched and sex-matched subjects (56.4% female; mean (SD) age, 65.9 (7.6) years). The variation of most parameters including decreased pupil constriction amplitude (CA), relative CA, average constriction velocity, venous diameter, venous blood flow and increased centred retinal reflectance in 548 nm (RC548) in cognitive impairment was consistent with previous studies while the reduced flow acceleration index and oxygen metabolism were reported for the first time. Compared with the logistic regression model, BPNN had better predictive performance (accuracy: 0.91 vs 0.69; sensitivity: 93.3% vs 61.70%; specificity: 90.0% vs 68.66%). Conclusions This study demonstrates retinal spectral signature alteration, neurodegeneration and angiopathy occur concurrently in cognitive impairment. The combination of multimodal ophthalmic imaging and BPNN can be a useful tool for predicting cognitive impairment with high performance for community screening. Data are available on reasonable request. The datasets used and/or analysed in this study are available from the corresponding author on reasonable request.

中文翻译：

---

使用多模态眼成像和人工神经网络预测认知障碍以进行社区筛查


背景/目的 利用多模态眼科成像和人工神经网络相结合，探讨一般老年人群认知障碍的综合预测能力。方法 招募认知障碍患者和认知健康个体。所有受试者均接受了病史、血压测量、蒙特利尔认知评估、医学验光、眼压和定制的多模态眼科成像，其中集成了瞳孔光反应、多光谱成像、激光散斑对比成像和视网膜血氧饱和度。通过 Student 的 t 检验分析多维参数。采用 Logistic 回归分析和反向传播神经网络 （BPNN） 确定认知障碍的预测能力。结果 本研究包括 104 名认知障碍患者 （61.5% 为女性;平均 （SD） 年龄为 68.3 （9.4） 岁），以及 94 名认知健康的年龄匹配和性别匹配的受试者 （56.4% 为女性;平均 （SD） 年龄，65.9 （7.6） 岁）。认知障碍中大多数参数的变化包括瞳孔收缩幅度 （CA） 降低、相对 CA、平均收缩速度、静脉直径、静脉血流和 548 nm 中心视网膜反射率增加 （RC548） 与以前的研究一致，而血流加速指数和氧代谢降低是首次报道。与 logistic 回归模型相比，BPNN 具有更好的预测性能 （准确性： 0.91 vs 0.69;敏感性： 93.3% vs 61.70%;特异性： 90.0% vs 68.66%）。结论 本研究证明视网膜光谱特征改变、神经变性和血管病同时发生在认知障碍中。 多模态眼成像和 BPNN 的结合可以成为预测认知障碍的有用工具，并且具有高性能的社区筛查。数据可应合理要求提供。本研究中使用和/或分析的数据集可应合理要求从通讯作者处获得。