Intracranial atherosclerosis (ICAS) is a major cause of ischemic stroke and is challenging to localize manually on Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) images. The difficulty arises from tortuous, overlapping vessels, low vessel–tissue contrast, and subtle stenotic lesions appearing as small gaps easily missed by radiologists. To address these challenges, we present a deep learning–based framework for fully automated, radiation-free ICAS localization on TOF-MRA scans. By accurately identifying stenotic regions, this tool enhances diagnostic sensitivity, spatial precision, and clinical workflow efficiency, supporting more effective patient care.