面向脑卒中早期预防和病因探寻的重大需求，本项目拟研究用于缺血性脑卒中血管斑块的快速磁共振成像及智能诊断新方法。针对现有技术采集时间过长图像质量不稳定等问题，研究新型三维血管壁成像序列和基于压缩感知的快速重建算法，以及与人工智能诊断相关的血管中心线跟踪算法、主动标记、深度学习Dense-UNet网络等。重点解决卷积特征图像重建框架及其对图像信噪比的影响、适应多种复杂组织信号演变的可变翻转角算法、针对斑块数据特征的高效网络结构设计方法与训练策略等关键科学问题。预期在国产3T磁共振实现头颈一体和主动脉血管壁成像各在5分钟以内完成，空间分辨率分别达到0.5mm和1mm各向同性，并实现对斑块的智能识别和分类，相似度和重叠度分别达到0.95和0.90。本项目的实施不仅为我国脑卒中防治提供一种更加先进有效的手段，而且有助于提高我国在高端磁共振成像和人工智能领域的自主知识产权和国际竞争力。

The project aims to develop fast MR imaging techniques and intelligent diagnostic methods for early detection of culprit plaques for stroke. To address the several challenges including long acquisition time and unstable image quality which limiting the ability to identify plaques automatically, we plan to develop a new 3D vessel wall imaging technique, compressed sensing based fast imaging reconstruction algorithms, curved planar reconstruction algorithms based on vessel centerline tracking, active label, and plaque identification and classification using DenseNet and UNet (DUNet). Several key issues will be investigated including (1) reconstruction framework based on convolution features and its effect on signal-to-noise ratio of MR images; (2) variable flip angle calculated algorithm apply to signal evolution of multiple complex tissue; (3) framework parameters selection and training mode optimization of DUNet. We expect to complete intra- and extra-cranial arterial wall joint imaging and aortic arterial wall imaging separately in 5 minutes with isotropic spatial resolution of 0.5mm and 1mm, respectively, and to achieve accurate identification and classification of stroke plaques based on artificial intelligence with intraclass correlation coefficient 0.95 and dice coefficient 0.9.