Cerebrovascular reserve (CVR) is an important hemodynamic parameter for moyamoya disease. Acetazolamide (ACZ) test is often used to measure CVR clinically. However, ACZ is contraindicated in patients with sulfa allergies, severe kidney and liver disease and potentially has severe adverse side effect. Thus, there is a need to assess CVR without pharmacological vasodilation. We utilized a simultaneous [15O]-water PET/MRI dataset to train a convolutional neural network (CNN) to predict CVR. The CNN combined multi-contrast information from baseline perfusion and structural images to predict whole-brain PET-level CVR, with high image quality, quantiﬁcation accuracy, and diagnostic accuracy for identifying impaired CVR.