题  目：Physics-Informed Deep Learning Methods for Accelerated MRI

内容简介：Accelerated MRI can be mathematically modeled as an inverse problem, and regularization is a crucial tool to achieve stable and accurate solutions. However, the acceleration rates of conventional MRI methods using traditional regularization have approached their limits. In recent years, deep learning methods have garnered increasing attention and are widely regarded as a breakthrough for further accelerating imaging. Nevertheless, existing deep learning imaging methods mostly lack the necessary interpretability, exposing imaging to the risk of instability. Fortunately, in contrast to general natural image processing problems, MRI is driven by MR physical principles. Therefore, we propose a physics-driven learnable regularization approach, wherein the design of inference algorithms, network structures, and loss functions is guided by physical priors. This results in a series of interpretable deep learning methods for MRI.

报告人：崔卓须

报告人简介：中国科学院深圳先进技术研究院，PI，副研究员，中科院特别研究助理项目资助，深圳市“鹏城孔雀计划”特聘岗位。2020年毕业于武汉大学数学与统计学院应用数学专业，获得理学博士学位。主要研究领域为计算磁共振成像，近五年在IEEE SPM (IF 14.9)、IEEE TMI (IF 10.6)、MEDIA (IF 10.9)和Inverse Problems (IF 2.1)等学术期刊发表论文20余篇，在ISMRM、AAAI等会议上发表会议论文/摘要10余篇。主持国家自然科学基金、中国博士后科学基金等，作为科研骨干参与科技部“十四五”数学和应用研究重点专项、国家自然科学基金重点项目、数学天元基金项目等。

时  间：2024年1月14日（周日）下午14：00开始

地  点：腾讯会议号：63129598867

热烈欢迎广大师生参加!

信息科学技术学院

2024年1月12日