Intracranial atherosclerosis stenosis (ICAS) is a leading cause of ischemic stroke worldwide and increase the global burden of stroke, especially in the Asian population. Compared with the other stroke subtypes, patients with ICAS had more severe stroke, stayed longer in the hospital and higher risk of recurrent ischemic events. Thus, early screening and effective intervention for intracranial atherosclerotic stenosis can improve the level of early warning and prevention of stroke, decrease the incidence and mortality of stroke, which is of vital significance. Although digital subtraction angiography (DSA), computed tomography angiography (CTA) and magnetic resonance angiography (MRA) have a high diagnostic value for ICAS, it is invasive and not available for mass population screening due to expertise, expensive cost, and poor economic performance. Potential screening tools such as transcranial doppler sonography (TCD) are promising but limited by temporal bone window quality and highly depends on operators' experience. Therefore, it is imperative to explore a novel, non-invasive, economic and complementary screening method for identifying the subjects with ICAS in mass populations such as primary health-care institutions and physical examination centers. The retina develops from the diencephalon, shares the same embryological origin, anatomic features and physiological properties with brain, including blood supply via the internal carotid artery. A prospective cohort study has confirmed that retinal vascular signs (enhanced arteriolar light reflex) are related to intracranial large artery disease8. Rhee et al' study have also showed that retinal diameter variation is associated with ICAS9. Hence, these findings hint retinal vascular signs may be a biomarker for ICAS. Besides, the traditional vascular risk factors, such as older age, hypertension, diabetes, dyslipidemia, smoking and others, are also tightly associated with ICAS. However, few studies have reported the discrimination performance of retinal vascular signs itself or combining with the traditional vascular risk factors in identifying ICAS. Fundus photography has great advantages including non-invasive, easy to popularize, inexpensive and possess good economic benefits, particularly in the age of artificial intelligence. Artificial intelligence (AI), especially deep learning algorithm has widely applied to accurate varieties of retinal diseases detection and classification such as diabetic retinopathy and glaucoma. Besides, deep learning algorithm was also used to automatic segmentation of retinal arteries or veins, which provides a basis for the subsequent automatic calculation of retinal vascular parameters. In this study, we aim to study on screening for intracranial atherosclerosis and predicting stroke risk based on fundus imaging features.