Title: A Deep Learning Approach For Compressive Sensing MRI
Date/Time: 11/05/2025 , 9:30am
Speaker: Bo Zhang, Academy of Mathematics and Systems Science(AMSS), Chinese Academy of Sciences
Location: Mathematics and Statistics Building, Room 213
Abstract: Magnetic resonance imaging (MRI) is an important but slow imaging technique. Undersampling in k-space can accelerate the sampling process but it brings challenges to high quality image reconstruction. Compressive sensing (CS) is an effective technique for high quality MRI image reconstruction from undersampled k-space measurements. In this talk, we propose a deep learning approach to the CS-based MRI image reconstruction from undersampled k-space measurements, named as NPDHG-CSnets. In NPDHG-CSnets, a general nonlinear regularization functional can be learned by a novel unroll deep neural network architecture, and the learned CS-MRI model is solved by a Nonlinear Primal-Dual Hybrid Gradient algorithm. Extensive experiments on real MRI image reconstruction show that the proposed NPDHG-CSnets achieves state-of-the-art reconstruction performance and good generalization ability. This talk is based on joint work with Ruihang Xu and Haiwen Zhang.
