AbstractAlzheimer’s disease (AD) is a global healthcare challenge lacking a simple and affordable detection method. We propose a novel deep learning framework, Eye-AD, to detect Early-onset Alzheimer’s Disease (EOAD) and Mild Cognitive Impairment (MCI) using OCTA images of retinal microvasculature and choriocapillaris.

摘要阿尔茨海默病（AD）是一种全球性的医疗保健挑战，缺乏简单且负担得起的检测方法。我们提出了一种新的深度学习框架Eye-AD，使用视网膜微血管和脉络膜毛细血管的OCTA图像检测早发性阿尔茨海默氏病（EOAD）和轻度认知障碍（MCI）。

Eye-AD employs a multilevel graph representation to analyze intra- and inter-instance relationships in retinal layers. Using 5751 OCTA images from 1671 participants in a multi-center study, our model demonstrated superior performance in EOAD (internal data: AUC = 0.9355, external data: AUC = 0.9007) and MCI detection (internal data: AUC = 0.8630, external data: AUC = 0.8037).

Eye AD采用多级图表示来分析视网膜层中的实例内和实例间关系。使用来自1671名多中心研究参与者的5751张OCTA图像，我们的模型在EOAD（内部数据：AUC=0.9355，外部数据：AUC=0.9007）和MCI检测（内部数据：AUC=0.8630，外部数据：AUC=0.8037）中表现出优异的性能。

Furthermore, we explored the associations between retinal structural biomarkers in OCTA images and EOAD/MCI, and the results align well with the conclusions drawn from our deep learning interpretability analysis. Our findings provide further evidence that retinal OCTA imaging, coupled with artificial intelligence, will serve as a rapid, noninvasive, and affordable dementia detection..

此外，我们探讨了OCTA图像中视网膜结构生物标志物与EOAD/MCI之间的关联，结果与我们的深度学习可解释性分析得出的结论非常吻合。我们的研究结果提供了进一步的证据，表明视网膜OCTA成像与人工智能相结合，将成为一种快速，无创且负担得起的痴呆症检测方法。。

IntroductionDementia affects approximately 50 million individuals worldwide, and the number of patients is increasing due to the aging population1. The development of new methods for studying dementia’s early pathophysiology is imperative. Current tests, such as magnetic resonance imaging (MRI), and biochemical quantification of proteins in cerebrospinal fluid are expensive, either time-consuming or invasive, and cannot be used on a large population.The retina has been postulated to be a window to the brain and hence may provide an opportunity for the study of neurodegeneration and microvascular changes in the early pathophysiology of dementia2,3.

引言痴呆症影响全球约5000万人，由于人口老龄化，患者人数正在增加1。开发研究痴呆早期病理生理学的新方法势在必行。目前的测试，如磁共振成像（MRI）和脑脊液中蛋白质的生化定量，无论是耗时的还是侵入性的，都是昂贵的，并且不能用于大量人群。视网膜被认为是大脑的窗口，因此可能为研究痴呆早期病理生理学中的神经变性和微血管变化提供机会2,3。

Histopathological reports from post-mortem cases have confirmed retinal microvascular changes in patients with Alzheimer’s disease (AD)4. Several clinical studies have supported this theory, confirming changes in retinal vasculature, vessels of different calibers, and structural changes around the optic nerve head and retinal structure in AD patients5,6,7,8.

尸检病例的组织病理学报告证实了阿尔茨海默病（AD）患者的视网膜微血管改变4。一些临床研究支持了这一理论，证实了AD患者视网膜脉管系统，不同口径血管以及视神经头周围和视网膜结构的变化5,6,7,8。

These changes can be detected with ophthalmic imaging modalities such as color fundus photography (CFP), optical coherence tomography (OCT), and OCT angiography (OCTA). Currently, accumulating evidence suggests that retinal imaging tools may provide useful biomarkers for the study and management of the early pathobiology of dementia.CFP imaging has advantages in accessibility and cost.

这些变化可以通过眼科成像方式检测，例如彩色眼底照相（CFP），光学相干断层扫描（OCT）和OCT血管造影（OCTA）。目前，越来越多的证据表明，视网膜成像工具可能为痴呆早期病理生物学的研究和管理提供有用的生物标志物。CFP成像在可访问性和成本方面具有优势。

However, its analysis detects retinal microvasculature changes mainly confined to arterioles and venules (60–300 μm in diameter) due to limited image resolutions9 and finds it challenging to detect more subtle vascular changes at the very early stages of the disease, such as mild cognitive impairment8,9.

。

OCTA is a novel imaging modality that noninvasively and quickly images the retinal micro.

OCTA是一种新型成像方式，可无创快速成像视网膜微。

(1)

(1)

where eij ∈ E indicates the attention value of node j to node i, and \(E\in {{\mathbb{R}}}^{{n}^{2}\times {n}^{2}}\) is the attention coefficient matrix. ∥ is the concatenation operation: \({\vec{a}}^{T}\in {{\mathbb{R}}}^{2{F}^{{\prime} }}\) is a learnable weight vector implemented by a fully connected layer, followed by the LeakyReLU activation (with a negative input slope α = 0.2).Considering the importance of each region, i.e., the node of the instance-level graph, during the disease diagnosis, we re-weight the attention coefficients using the importance value obtained by the ICM.

其中eij∈E表示节点j对节点i的注意值，而\（E \ in{mathbb{R}}}^{n}^{2}\次{n}^{2}}\）是注意系数矩阵。是级联运算：\（{vec{a}}^{T}\ in{mathbb{R}}}}^{2{F}^{{\ prime}}}是一个可学习的权重向量，由一个完全连接的层实现，然后是LeakyReLU激活（输入斜率为负α=0.2）。考虑到疾病诊断过程中每个区域（即实例级图的节点）的重要性，我们使用ICM获得的重要性值重新加权注意系数。

We denote the importance matrix of \({{\mathbb{G}}}^{k}\) as Mk, whose size is n2 × 1. The re-weighting operation is defined as:$${E}^{{\prime} }=E(Diag({M}^{k})),$$.

我们将\（{\ mathbb{G}}}^{k}\）的重要性矩阵表示为Mk，其大小为n2××1。重加权运算定义为：$${E}^{\ prime}}=E（Diag（{M}^{k}）），$$。

(2)

(2)

where Diag(⋅) is the diagonalization operation. Then we normalize the coefficient \({e}_{ij}^{{\prime} }\) using the softmax function to make it comparable across different nodes:$${\alpha }_{ij}={{Softmax}}_{j}({e}_{ij}^{{\prime} })=\frac{exp({e}_{ij}^{{\prime} })}{{\sum }_{l\in {N}_{i}}exp({e}_{il}^{{\prime} })},$$.

其中Diag（⋅）是对角化操作。\({e}_{ij}^{{\ prime}}\）使用softmax函数使其在不同节点之间具有可比性：$${\ alpha}{ij}={{softmax}}{j}({e}_{ij}^{\素数}）=\ frac{exp({e}_{ij}^{{\素数}}{{\求和}{l \ in{N}_{i} }经验({e}_{il}^{\素数}}，$$。

(3)

(3)

where Ni denotes the neighborhood of node i in the graph, showing that only \({e}_{ij}^{{\prime} }\) for neighboring node j ∈ Ni is considered during the update of the node feature to avoid involving irrelevant nodes. Finally, we use the normalized attention coefficients αij to calculate a weighted sum of the involved node features to obtain the updated features for each node:$${\vec{h}}_{i}^{{\prime} }=ELU\left(\sum _{j\in {N}_{i}}{\alpha }_{ij}W\,{\vec{h}}_{j}\right),$$.

其中Ni表示图中节点i的邻域，仅显示\({e}_在更新节点特征期间，考虑相邻节点j的{ij}^{\ prime}}Ni，以避免涉及不相关的节点。最后，我们使用归一化的注意系数αij来计算所涉及的节点特征的加权和，以获得每个节点的更新特征：$${\ vec{h}}{i}^{\ prime}}=ELU \ left（\ sum{j \ in{N}_{i} }{\alpha}_{ij}W\，{\vec{h}}\uj}\右），$$。

(4)

(4)

where ELU represents the exponential linear unit (ELU) nonlinearity. In this way, the proposed IAGAT layer can extract instance embedding by cascading the updated node features of \({{\mathbb{G}}}^{k}\), which are used to construct the subject-level graph \({\mathcal{G}}\). Then, we employ the GAT layer19, a popular GNN method with an attention mechanism, as the extractor to obtain the subject features.

其中ELU表示指数线性单位（ELU）非线性。通过这种方式，所提出的IAGAT层可以通过级联更新的节点特征来提取实例嵌入，这些特征用于构建主题级图（主题级图）。然后，我们使用GAT layer19（一种带有注意机制的流行GNN方法）作为提取器来获取主题特征。

Finally, the subject feature is fed into a fully connected layer followed by a softmax activation layer for the final subject classification.To improve generalization and capture the subtle differences between different subgraphs of the same subject, we design a sub-graph consistency regularization (SCR)-based loss function.

最后，主题特征被送入完全连接的层，然后是softmax激活层，用于最终的主题分类。为了改进泛化并捕获同一主题的不同子图之间的细微差异，我们设计了基于子图一致性正则化（SCR）的损失函数。

The SCR randomly samples two subgraphs from the same subject-level graph and minimizes their squared L2 distance after passing through a two-layer MLP network. Specifically, for each input, a graph consisting of different projection layers is generated with n nodes. During training, two graphs consisting of different n-1 nodes, i.e., subgraphs, are sampled from n nodes through random sampling.

SCR从同一主题级图中随机采样两个子图，并在通过两层MLP网络后最小化它们的平方L2距离。具体而言，对于每个输入，生成由n个节点的不同投影层组成的图。在训练期间，通过随机采样从n个节点中采样由不同n-1个节点组成的两个图，即子图。

We aim for the features related to disease prediction represented by these two subgraphs to be consistent, achieved through the consistency constraints of the SCR loss function. This SCR encourages the model to learn more robust and discriminative representations that can capture the subtle differences between different subgraphs of the same subject, thus improving the generalization ability of the model.

我们的目标是通过SCR损失函数的一致性约束，使这两个子图所代表的与疾病预测相关的特征保持一致。该SCR鼓励模型学习更健壮和有区别的表示，这些表示可以捕获同一主题的不同子图之间的细微差异，从而提高模型的泛化能力。

This regularization term is particularly effective for small datasets, where overfitting is a major concern, and has been shown to improve classification performance in our experiments.Training strategyEye-AD is an end-to-end framework. We have.

这个正则化项对于小数据集特别有效，其中过度拟合是一个主要问题，并且在我们的实验中已经证明可以提高分类性能。培训策略眼广告是一个端到端的框架。

VAD is defined as the total length in millimeters of perfused retinal microvasculature per unit area of the analyzed image.

VAD定义为分析图像每单位面积灌注的视网膜微血管的总长度（毫米）。

VLD is defined as the ratio of the total number of pixels on microvascular centerlines to the measurement area.

VLD定义为微血管中心线上的像素总数与测量区域的比率。

VFD is a measure of the global branching complexity of the vasculature.

VFD是衡量脉管系统整体分支复杂性的指标。

VB is the total number of bifurcations in the analyzed area.

VB是分析区域中分叉的总数。

FA is defined as the total number of pixels in the FAZ region.

FA被定义为FAZ区域中的像素总数。

FC measures the degree of roundness of the FAZ, calculated as: FC = 4π*FA/FP2. FP is the perimeter of FAZ. A larger FC indicates a more circular shape. A value of 1.0 denotes a perfect circle.

FC测量FAZ的圆度，计算公式为：FC=4π*FA/FP2。FP是FAZ的周长。FC越大表示形状越圆。值1.0表示一个完美的圆。

FR is similar to FC but is less sensitive to irregular borders along the perimeter of the FAZ, and is calculated as: \(FR=4\pi * FA/{L}_{major}^{2}\). /Lmajor is the major axial length of FAZ.

FR类似于FC，但对沿FAZ周长的不规则边界不太敏感，计算公式为：\（FR=4 \ pi*FA/{L}_{专业}^{2}\）/Lmajor是FAZ的主要轴向长度。

FS describes the extent to which the FAZ is convex or concave: it is defined as the ratio between the area of the FAZ, and the area of the convex hull covering the FAZ. The farther the solidity deviates from 1, the greater the concavity of the structure.

FS描述了FAZ凸凹的程度：它定义为FAZ面积与覆盖FAZ的凸包面积之比。稠度越偏离1，结构的凹度越大。

We explore the differences of these parameters between AD and healthy control groups in the whole image as well as in sub-sectors. The sub-sectors include the superior inner, temporal inner, inferior inner, nasal inner, superior outer, temporal outer, inferior outer, and nasal outer8. We use a multiple linear regression model with generalized estimating equations to correlate AD and retinal microvasculature and FAZ measurements, adjusting for age, gender, hypertension, diabetes, and education level.

我们探讨了AD和健康对照组在整个图像以及子部门中这些参数的差异。子部门包括上内，颞内，下内，鼻内，上外，颞外，下外和鼻外8。我们使用具有广义估计方程的多元线性回归模型来关联AD和视网膜微血管以及FAZ测量值，并根据年龄，性别，高血压，糖尿病和教育水平进行调整。

We perform the analyses using standard statistical software (SPSS, v.24.0, IBM, US). The results are presented in Table 8.Table 8 The results of the global statistical analysis of the parameters, adjusted for factors such as age, gender, hypertension, diabetes, and education level, were obtained utilizing Generalized Estimation EquationFull size table.

我们使用标准统计软件（SPSS，v.24.0，IBM，US）进行分析。结果如表8所示。表8使用广义估计方程全尺寸表获得了参数的全球统计分析结果，并根据年龄，性别，高血压，糖尿病和教育水平等因素进行了调整。

Data availability

数据可用性

Restrictions apply to the availability of the developmental and validation datasets, which were used with the permission of the participants for the current study. De-identified data may be available for research purposes from the corresponding authors upon reasonable request.

限制适用于开发和验证数据集的可用性，这些数据集是在参与者的许可下用于当前研究的。经合理要求，通讯作者可能会出于研究目的提供未识别的数据。

Code availability

代码可用性

All algorithms used in this study were developed using libraries and scripts in PyTorch. The source code is publicly available at https://github.com/iMED-Lab/Eye-AD.

本研究中使用的所有算法都是使用PyTorch中的库和脚本开发的。源代码可在https://github.com/iMED-Lab/Eye-AD.

ReferencesNichols, E. et al. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the global burden of disease study 2019. Lancet Public Health 7, e105–e125 (2022).Article

参考文献Nichols，E。等人，《2019年全球痴呆患病率的估计和2050年的预测患病率：2019年全球疾病负担研究的分析》。柳叶刀公共卫生7，e105-e125（2022）。文章

Google Scholar

谷歌学者

Abajobir, A. et al. Neurological disorders collaborator group, global, regional, and national burden of neurological disorders during 1990–2015: a systematic analysis for the global burden of disease study 2015. Lancet Neurol. 16, 877–897 (2017).Article

Abajobir，A.等人，《1990-2015年全球、地区和国家神经疾病负担：2015年全球疾病负担研究的系统分析》，神经疾病合作者小组。柳叶刀神经学。16877-897（2017）。文章

Google Scholar

谷歌学者

Wu, Y.-T. et al. The changing prevalence and incidence of dementia over time-current evidence. Nat. Rev. Neurol. 13, 327–339 (2017).Article

Wu，Y.-T.等人。随着时间的推移，痴呆症的患病率和发病率不断变化。神经病学杂志。。文章

PubMed

PubMed

Google Scholar

谷歌学者

Koronyo, Y. et al. Retinal pathological features and proteome signatures of Alzheimer’s disease. Acta Neuropathol. 145, 409–438 (2023).Article

Koronyo，Y.等人。阿尔茨海默病的视网膜病理特征和蛋白质组特征。神经病学报。145409-438（2023）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Zabel, P. et al. Comparison of retinal microvasculature in patients with Alzheimer’s disease and primary open-angle glaucoma by optical coherence tomography angiography. Investig. Ophthalmol. Vis. Sci. 60, 3447–3455 (2019).Article

Zabel，P。等人。通过光学相干断层扫描血管造影比较阿尔茨海默病和原发性开角型青光眼患者的视网膜微血管。调查。眼科。。科学。603447-3455（2019）。文章

Google Scholar

谷歌学者

Wu, J. et al. Retinal microvascular attenuation in mental cognitive impairment and Alzheimer’s disease by optical coherence tomography angiography. Acta ophthalmol. 98, e781–e787 (2020).Article

Wu，J。等人。通过光学相干断层扫描血管造影在精神认知障碍和阿尔茨海默病中的视网膜微血管衰减。眼科学报。98，e781–e787（2020）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Cheung, C. Y. et al. A deep learning model for detection of Alzheimer’s disease based on retinal photographs: a retrospective, multicentre case-control study. Lancet Digit. Health 4, e806–e815 (2022).Article

Cheung，C.Y.等人。基于视网膜照片检测阿尔茨海默病的深度学习模型：一项回顾性多中心病例对照研究。柳叶刀数字。健康4，e806–e815（2022）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Xie, J. et al. Deep segmentation of octa for evaluation and association of changes of retinal microvasculature with alzheimer’s disease and mild cognitive impairment. Br. J. Ophthalmol. 108, 432–439 (2024).Article

Xie，J.等。octa的深度分割，用于评估视网膜微血管变化与阿尔茨海默病和轻度认知障碍的关系。Br.J.眼科。108432-439（2024）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Chua, J. et al. Retinal microvasculature dysfunction is associated with alzheimer’s disease and mild cognitive impairment. Alzheimer’s Res. Ther. 12, 1–13 (2020).

Chua，J。等人。视网膜微血管功能障碍与阿尔茨海默病和轻度认知障碍有关。阿尔茨海默病研究。12，1-13（2020）。

Google Scholar

谷歌学者

Cheung, C. Y.-l et al. Microvascular network alterations in the retina of patients with alzheimer’s disease. Alzheimer’s Dement. 10, 135–142 (2014).Article

Cheung，C.Y.-l等人。阿尔茨海默病患者视网膜微血管网络的改变。阿尔茨海默氏症。10135-142（2014）。文章

Google Scholar

谷歌学者

Tian, J. et al. Modular machine learning for alzheimer’s disease classification from retinal vasculature. Sci. Rep. 11, 238 (2021).Article

Tian，J。等人。用于从视网膜脉管系统分类阿尔茨海默病的模块化机器学习。科学。第11238页（2021年）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Lim, Y. J. et al. Efficient deep learning algorithm for alzheimer’s disease diagnosis using retinal images. In 2022 IEEE 4th International Conference on Artificial Intelligence Circuits and Systems (AICAS) 254–257 (IEEE, 2022).Hermessi, H., Mourali, O. & Zagrouba, E. Multimodal medical image fusion review: Theoretical background and recent advances.

Lim，Y.J.等人。使用视网膜图像诊断阿尔茨海默病的有效深度学习算法。。Hermessi，H.，Mourali，O。＆Zagrouba，E。多模式医学图像融合综述：理论背景和最新进展。

Signal Process. 183, 108036 (2021).Article .

信号处理。183108036（2021）。文章。

Google Scholar

谷歌学者

Zhou, T. et al. Deep multi-modal latent representation learning for automated dementia diagnosis. In International Conference on Medical Image Computing and Computer-Assisted Intervention, 629–638 (Springer, 2019).Heisler, M. et al. Ensemble deep learning for diabetic retinopathy detection using optical coherence tomography angiography.

Zhou，T。等人。用于自动痴呆诊断的深度多模式潜在表征学习。在国际医学图像计算和计算机辅助干预会议上，629-638（Springer，2019）。Heisler，M.等人。使用光学相干断层扫描血管造影进行糖尿病视网膜病变检测的集成深度学习。

Transl. Vis. Sci. Technol. 9, 20 (2020).Article .

翻译。。科学。技术。9，20（2020）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Zhou, T., Canu, S., Vera, P. & Ruan, S. 3D medical multi-modal segmentation network guided by multi-source correlation constraint. In 2020 25th International Conference on Pattern Recognition (ICPR) 10243–10250 (IEEE, 2021).Wang, X. et al. Screening of dementia on octa images via multi-projection consistency and complementarity.

Zhou，T.，Canu，S.，Vera，P。＆Ruan，S。由多源相关约束引导的3D医学多模态分割网络。2020年，第25届国际模式识别会议（ICPR）10243–10250（IEEE，2021）。Wang，X。等人。通过多投影一致性和互补性在octa图像上筛选痴呆症。

In International Conference on Medical Image Computing and Computer-Assisted Intervention 688–698 (Springer, 2022).Kipf, T. N. & Welling, M. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations (2022).Velickovic, P. et al. Graph attention networks.

在国际医学图像计算和计算机辅助干预会议688-698（Springer，2022）。Kipf，T。N。＆Welling，M。使用图卷积网络的半监督分类。在国际学习表征会议（2022年）上。Velickovic，P。等人。图形注意网络。

Statistics 1050, 10–48550 (2017)..

统计1050，10–48550（2017）。。

Google Scholar

谷歌学者

Hao, J. et al. Uncertainty-guided graph attention network for parapneumonic effusion diagnosis. Med. Image Anal. 75, 102217 (2022).Article

郝，J。等。不确定性引导图形注意网络用于肺炎旁积液诊断。医学图像肛门。75102217（2022）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Dosovitskiy, A. et al. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations (2020).Liu, Z. et al. Swin transformer v2: Scaling up capacity and resolution. In Proc. IEEE/CVF conference on computer vision and pattern recognition, 12009–12019 (IEEE, 2022).Katsimpris, A.

Dosovitskiy，A。等人。一幅图像相当于16x16个单词：用于大规模图像识别的变压器。在国际学习表征会议（2020年）上。Liu，Z.等人。Swin transformer v2：放大容量和分辨率。在过程中。IEEE/CVF计算机视觉和模式识别会议，12009-12019（IEEE，2022）。Katsimpris，A。

et al. Optical coherence tomography angiography in alzheimer’s disease: a systematic review and meta-analysis. Eye 36, 1419–1426 (2022).Article .

阿尔茨海默病的光学相干断层扫描血管造影：系统综述和荟萃分析。眼睛361419-1426（2022）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Campbell, J. et al. Detailed vascular anatomy of the human retina by projection-resolved optical coherence tomography angiography. Sci. Rep. 7, 42201 (2017).Article

Campbell，J.等人通过投影分辨光学相干断层扫描血管造影对人类视网膜进行了详细的血管解剖。科学。。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Lahme, L. et al. Evaluation of ocular perfusion in alzheimer’s disease using optical coherence tomography angiography. J. Alzheimer’s Dis. 66, 1745–1752 (2018).Article

Lahme，L。等人。使用光学相干断层扫描血管造影评估阿尔茨海默病的眼部灌注。J、 阿尔茨海默病。661745-1752（2018）。文章

Google Scholar

谷歌学者

Zhang, Y. S. et al. Parafoveal vessel loss and correlation between peripapillary vessel density and cognitive performance in amnestic mild cognitive impairment and early alzheimer’s disease on optical coherence tomography angiography. PLoS ONE 14, e0214685 (2019).Article

Zhang，Y.S.等。光学相干断层扫描血管造影中遗忘型轻度认知障碍和早期阿尔茨海默病的中央凹旁血管丢失以及乳头周围血管密度与认知表现之间的相关性。PLoS ONE 14，e0214685（2019）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Jiang, H. et al. Altered macular microvasculature in mild cognitive impairment and alzheimer disease. J. Neuro-Ophthalmol. Off. J. North Am. Neuro-Ophthalmol. Soc. 38, 292 (2018).Article

Jiang，H.等人在轻度认知障碍和阿尔茨海默病中改变了黄斑微血管。J、 神经眼科。Off。J。North Am。神经眼科。Soc.38292（2018）。文章

Google Scholar

谷歌学者

Liang, W., Olwen, M., Gonzalez, C. N., Calabresi, P. A. & Shiv, S. Emerging applications of optical coherence tomography angiography (octa) in neurological research. Eye Vis. 5, 11– (2018).Article

Liang，W.，Olwen，M.，Gonzalez，C.N.，Calabresi，P.A。＆Shiv，S。光学相干断层扫描血管造影（octa）在神经学研究中的新兴应用。眼睛可见。5,11–（2018）。文章

Google Scholar

谷歌学者

Hormel, T. T., Jia, Y., Jian, Y., Hwang, T. S. & Huang, D. Plexus-specific retinal vascular anatomy and pathologies as seen by projection-resolved optical coherence tomographic angiography. Prog. Retin. Eye Res. 80, 100878 (2020).Article

Hormel，T.T.，Jia，Y.，Jian，Y.，Hwang，T.S.＆Huang，D。通过投影分辨光学相干断层扫描血管造影观察到的丛特异性视网膜血管解剖和病理。程序。雷丁。Eye Res.80100878（2020）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Lai, A. Y. et al. Venular degeneration leads to vascular dysfunction in a transgenic model of alzheimer’s disease. Brain 138, 1046–1058 (2015).Article

Lai，A.Y.等人。在阿尔茨海默病的转基因模型中，静脉变性导致血管功能障碍。大脑1381046-1058（2015）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Fisher, R. A., Miners, J. S. & Love, S. Pathological changes within the cerebral vasculature in alzheimer’s disease: New perspectives. Brain Pathol. 32, e13061 (2022).Article

Fisher，R.A.，Miners，J.S。＆Love，S。阿尔茨海默病脑血管系统内的病理变化：新观点。大脑病理学。32，e13061（2022）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

He, L. et al. Effect of dynamic cerebral autoregulation on the association between deep medullary vein changes and cerebral small vessel disease. Front. Physiol. 14, 1037871 (2023).Article

He，L.等人。动态脑自动调节对髓质深静脉改变与脑小血管疾病之间关系的影响。正面。生理学。141037871（2023）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Asanad, S. et al. The retinal choroid as an oculovascular biomarker for alzheimer’s dementia: a histopathological study in severe disease. Alzheimer’s Dement. Diagn. Assess. Dis. Monit. 11, 775–783 (2019).

视网膜脉络膜作为阿尔茨海默病痴呆的眼血管生物标志物：严重疾病的组织病理学研究。阿尔茨海默氏症。诊断。评估。Dis。莫尼特。11775-783（2019）。

Google Scholar

谷歌学者

McKhann, G. M. et al. The diagnosis of dementia due to alzheimer’s disease: Recommendations from the national institute on aging-alzheimer’s association workgroups on diagnostic guidelines for alzheimer’s disease. Alzheimer’s Dement. 7, 263–269 (2011).Article

McKhann，G.M.等人，《阿尔茨海默病痴呆症的诊断：国家老年痴呆症研究所协会阿尔茨海默病诊断指南工作组的建议》。阿尔茨海默氏症。7263-269（2011）。文章

Google Scholar

谷歌学者

Albert, M. S. et al. The diagnosis of mild cognitive impairment due to alzheimer’s disease: recommendations from the national institute on aging-alzheimer’s association workgroups on diagnostic guidelines for alzheimer’s disease. Alzheimer’s Dement. 7, 270–279 (2011).Article

阿尔伯特，M.S。等人，《阿尔茨海默病引起的轻度认知障碍的诊断：国家老年痴呆症研究所协会阿尔茨海默病诊断指南工作组的建议》。阿尔茨海默氏症。7270-279（2011）。文章

Google Scholar

谷歌学者

He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. In Proc. IEEE conference on computer vision and pattern recognition 770–778 (IEEE, 2016).Chen, Z., Liu, J., Zhu, M., Woo, P. Y. & Yuan, Y. Instance importance-aware graph convolutional network for 3d medical diagnosis.

He，K.，Zhang，X.，Ren，S。＆Sun，J。用于图像识别的深度残差学习。在过程中。IEEE计算机视觉和模式识别会议770-778（IEEE，2016）。Chen，Z.，Liu，J.，Zhu，M.，Woo，P.Y。＆Yuan，Y。用于3d医学诊断的实例重要性感知图卷积网络。

Med. Image Anal. 78, 102421 (2022).Article .

医学图像肛门。78102421（2022）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Selvaraju, R. R. et al. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proc. IEEE international conference on computer vision 618–626 (IEEE, 2017).Hao, J. et al. Retinal structure detection in octa image via voting-based multitask learning. IEEE Trans.

Selvaraju，R.R.等人，Grad cam：通过基于梯度的定位从深度网络进行视觉解释。在过程中。IEEE国际计算机视觉会议618-626（IEEE，2017）。Hao，J。等人。通过基于投票的多任务学习在octa图像中检测视网膜结构。IEEE Trans。

Med. Imaging 41, 3969–3980 (2022).Article .

医学成像413969-3980（2022）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Download referencesAcknowledgementsThis work was supported in part by the National Science Foundation Program of China (62422122, 62272444, 62371442, 62302488), in part by the Youth Innovation Promotion Association CAS (2021298), in part by the Zhejiang Provincial Natural Science Foundation of China (LR22F020008, LQ23F010007, LR24F010002, LZ23F010002), in part by Key research and development program of Zhejiang Province (2024C03101, 2024C03204) and Key Project of Ningbo Public Welfare Science and Technology (2023S012).

下载参考文献致谢这项工作部分得到了国家科学基金项目（62422122、62272444、62371442、62302488）的支持，部分得到了中国科学院青年创新促进会（2021298）的支持，部分得到了浙江省自然科学基金（LR22F020008、LQ23F010007、LR24F010002、LZ23F010002）的支持，部分得到了浙江省重点研究发展计划（2024C03101、2024C03204）和宁波市公益科技重点项目（2023S012）的支持。

AFF acknowledges the support of the Royal Academy of Engineering Chair INSILEX (CiET1819\9), the UKRI Frontier Research Guarantee INSILICO (EP\Y030494\1). The research of AFF was carried out at the National Institute for Health and Care Research (NIHR) Manchester Biomedical Research Centre (BRC) (NIHR203308).Author informationAuthor notesThese authors contributed equally: Jinkui Hao, William R.

AFF感谢皇家工程学院主席INSILEX（CiET1819）和UKRI前沿研究保障INSILICO（EP Y030494 1）的支持。AFF的研究是在国家卫生与保健研究所（NIHR）曼彻斯特生物医学研究中心（BRC）（NIHR203308）进行的。。

Kwapong, Ting Shen.Authors and AffiliationsLaboratory of Advanced Theranostic Materials and Technology, Ningbo Institute of Materials Technology and Engineering, Chinese Academy of Sciences, Ningbo, ChinaJinkui Hao, Shouyue Liu, Jiong Zhang & Yitian ZhaoDepartment of Neurology, West China Hospital, Sichuan University, Chengdu, ChinaWilliam R.

Kwapong，Ting Shen。作者和附属机构中国科学院宁波材料技术与工程研究所先进治疗诊断材料与技术实验室，宁波，中国金魁浩，刘寿月，张炯和赵益田四川大学华西医院神经内科，成都，中国威廉R。

Kwapong & Shuting ZhangDepartment of Ophthalmology, the Second Affiliated Hospital of Zhejiang University, Hangzhou, ChinaTing ShenInstitute of High-Performance Computing, Agency for Science, Technology and Research, Singapore, SingaporeHuazhu FuSchool of Future Technology, South China University of Technology, Guangzhou, ChinaYanwu XuDepartment of Ophthalmology, the Affiliated People’s Hospital of Ningbo University, Ningbo, ChinaQinkang Lu & Yitian ZhaoDepartment of Computer Science, Edge Hill University, Orms.

Kwapong＆Shuting Zhang浙江大学第二附属医院眼科，杭州，中国科学院高性能计算研究所，新加坡科学技术与研究机构，新加坡华珠未来技术学院，华南理工大学，广州，中国宁波大学附属人民医院眼科，宁波，中国秦康路和赵益田，边缘山大学计算机科学系，Orms。

PubMed Google ScholarWilliam R. KwapongView author publicationsYou can also search for this author in

PubMed谷歌学者William R.KwapongView作者出版物您也可以在

PubMed Google ScholarTing ShenView author publicationsYou can also search for this author in

PubMed Google ScholarTing ShenView作者出版物您也可以在

PubMed Google ScholarHuazhu FuView author publicationsYou can also search for this author in

PubMed Google ScholarHuazhu FuView作者出版物您也可以在

PubMed Google ScholarYanwu XuView author publicationsYou can also search for this author in

PubMed Google ScholarYanwu XuView作者出版物您也可以在

PubMed Google ScholarQinkang LuView author publicationsYou can also search for this author in

PubMed谷歌学者Qinkang LuView作者出版物您也可以在

PubMed Google ScholarShouyue LiuView author publicationsYou can also search for this author in

PubMed Google ScholarShouyue LiuView作者出版物您也可以在

PubMed Google ScholarJiong ZhangView author publicationsYou can also search for this author in

PubMed Google ScholarJiong ZhangView作者出版物您也可以在

PubMed Google ScholarYonghuai LiuView author publicationsYou can also search for this author in

PubMed谷歌学术刘永怀查看作者出版物您也可以在

PubMed Google ScholarYifan ZhaoView author publicationsYou can also search for this author in

PubMed Google ScholarYifan ZhaoView作者出版物您也可以在

PubMed Google ScholarYalin ZhengView author publicationsYou can also search for this author in

PubMed Google ScholarYalin ZhengView作者出版物您也可以在

PubMed Google ScholarAlejandro F. FrangiView author publicationsYou can also search for this author in

PubMed Google ScholarAlejandro F.FrangiView作者出版物您也可以在

PubMed Google ScholarShuting ZhangView author publicationsYou can also search for this author in

PubMed Google ScholarShuting ZhangView作者出版物您也可以在

PubMed Google ScholarHong QiView author publicationsYou can also search for this author in

PubMed Google ScholarHong QiView作者出版物您也可以在

PubMed Google ScholarYitian ZhaoView author publicationsYou can also search for this author in

PubMed谷歌学者ZhaoView作者出版物您也可以在

PubMed Google ScholarContributionsJ.H. designed and executed all the experiments conducted all the subsequent statistical analyses, and drafted the manuscript. W.R.K. and A.F. helped design the experiments, the writing, and the data interpretation and made substantial revisions and edits of the draft manuscript.

PubMed谷歌学术贡献。H、 设计并执行了所有实验，进行了所有后续的统计分析，并起草了手稿。W、 R.K.和A.F.帮助设计了实验，写作和数据解释，并对稿件草案进行了实质性修改和编辑。

S.Z., Q.L., and T.S. contributed to the data analysis and data cleaning. S.L. and J.L. helped design the experiments and contributed to the data analysis. Y.L., Y.Z., and Y.Z. contributed to the external validation of the proposed method. H.Q., Y.Z. conceived the methodology, designed the experiments, and contributed to the writing.

S、 Z.，Q.L。和T.S.为数据分析和数据清理做出了贡献。S、 。Y、 L.，Y.Z.和Y.Z.为所提出方法的外部验证做出了贡献。H、 Q.，Y.Z.构思了方法论，设计了实验，并为写作做出了贡献。

All authors contributed to the manuscript.Corresponding authorsCorrespondence to.

所有作者都为手稿做出了贡献。通讯作者通讯。

Shuting Zhang, Hong Qi or Yitian Zhao.Ethics declarations

张舒婷、红旗或赵益田。道德宣言

Competing interests

相互竞争的利益

The authors declare no competing interests.

。

Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary informationSupplemental MaterialRights and permissions

Additional informationPublisher的注释Springer Nature在已发布的地图和机构隶属关系中的管辖权主张方面保持中立。补充信息补充材料权利和权限

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.

。

You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

根据本许可证，您无权共享源自本文或其部分的改编材料。本文中的图像或其他第三方材料包含在文章的知识共享许可证中，除非该材料的信用额度中另有说明。如果材料未包含在文章的知识共享许可中，并且您的预期用途不受法律法规的许可或超出许可用途，则您需要直接获得版权所有者的许可。

To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/..

要查看此许可证的副本，请访问http://creativecommons.org/licenses/by-nc-nd/4.0/..

Reprints and permissionsAbout this articleCite this articleHao, J., Kwapong, W.R., Shen, T. et al. Early detection of dementia through retinal imaging and trustworthy AI.

转载和许可本文引用本文Hao，J.，Kwapong，W.R.，Shen，T。等人。通过视网膜成像和可靠的AI早期发现痴呆症。

npj Digit. Med. 7, 294 (2024). https://doi.org/10.1038/s41746-024-01292-5Download citationReceived: 22 May 2024Accepted: 04 October 2024Published: 20 October 2024DOI: https://doi.org/10.1038/s41746-024-01292-5Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard.

npj数字。医学7294（2024）。https://doi.org/10.1038/s41746-024-01292-5Download引文接收日期：2024年5月22日接受日期：2024年10月4日发布日期：2024年10月20日OI：https://doi.org/10.1038/s41746-024-01292-5Share本文与您共享以下链接的任何人都可以阅读此内容：获取可共享链接对不起，本文目前没有可共享的链接。复制到剪贴板。

Provided by the Springer Nature SharedIt content-sharing initiative

由Springer Nature SharedIt内容共享计划提供