Keynotes
Markus Haltmaier, University of Innsbruck
Bio: Markus Haltmeier received his Ph.D. degree in mathematics from the University of Innsbruck, Tyrol, Austria, in 2007, for research on computed tomography. He was then involved in various aspects of inverse problems as a research scientist with the University of Innsbruck, the University of Vienna, Austria, and the Max Planck Institute for Biophysical Chemistry, Göttingen, Germany. Since 2012, he is full professor with the Department of Mathematics, University of Innsbruck. His current research interests include inverse problems, signal and image processing, computerized tomography, and machine learning.
Regularizing Inverse Problems with Deep Neural Networks
Recently, deep learning and neural network based algorithms appeared as new paradigm for solving inverse problems. We propose and analyze NETT, naming Tikhonov regularization using a neural network as regularizer. We present a convergence analysis, derive convergence rates, and propose a possible training strategy. Numerical results are presented where NETT demonstrates good performance even for unknowns very different from the training data. Additionally, we discuss regularizing two-step networks.
Dong Liang, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
Dr. Dong Liang is a full professor in Biomedical Engineering, He is the Director of Research center for Medical AI and Deputy director of Research center for Biomedical Imaging, the Institute of Biomedical and Health Engineering, Shenzhen Institutes of Advanced Technology (SIAT), Chinese Academy of Sciences. His research interests include biomedical imaging, signal processing and machine learning. Dr. Dong Liang has published over 100 scientific papers and abstracts in international journals and conference proceedings. He is the Associate Editor of IEEE Transactions on Medical Imaging, Editorial Board Member of Magnetic Resonance in Medicine and the member of IEEE Computational Imaging Technical Committee.
Deep Fast MR Imaging: when Compressed Sensing meets Deep Neural Network
Accelerating magnetic resonance imaging (MRI) has been an ongoing research topic since its invention in the 1970s. Among a variety of acceleration techniques, compressed sensing (CS) has become an important strategy during the past decades. Although CS-based methods can achieve high performance with many theoretical guarantees, it is challenging to determine the numerical uncertainties in the reconstruction model such as the optimal sparse transformations, sparse regularizer in the transform do-main, regularization parameters and the parameters of the optimization algorithm. Recently, deep learning has been introduced in MR reconstruction to address these issues and shown potential to significantly improve image quality. In this presentation, we survey model-driven deep learning methods for MR reconstruction. Using some algorithms as examples, we explain how to unroll the iterations of a reconstruction process to a learnable deep network architecture. A general framework on combining the CS-MR model with deep learning is proposed to maximize the potential of deep learning and model-based reconstruction for fast MR imaging.
Yong Long, Shanghai Jiao Tong University
Yong Long received the Ph.D. degree in electrical engineering: systems from the University of Michigan (U-M) in 2011, the M.S. degree in electrical engineering from Fudan University, Shanghai, China in 2006, and the B.S. degree in electrical engineering from East China Normal University, Shanghai, China, in 2003. She is currently an assistant professor in the University of Michigan-Shanghai Jiao Tong University (SJTU) Joint Institute (JI) at SJTU, Shanghai, China. Prior to joining SJTU in 2014, she was a research scientist in the X-Ray Computed Tomography (CT) Systems and Applications Laboratory at General Electric (GE) Global Research Center from 2012 to 2014, and was a postdoctoral research fellow in the Department of Electrical Engineering and Computer Science (EECS) at U-M from 2011 to 2012. Her research interests include image reconstruction, machine learning, compreseed sensing, and image processing.
Learning Model-Based Image Reconstruction for X-Ray CT
Model-based image reconstruction (MBIR) methods that iteratively find the image that best fits the measurement, according to the system physical model, the measurement statistical model and prior information about the object, improve the ability to produce high-quality and accurate images, while greatly reducing patient exposure to potentially harmful levels of radiation. The challenge with further dose reduction to an ultra low-dose level is that ultra low-dose CT (ULDCT) data itself is not adequate enough for producing high quality images. MBIR combined with learned object models or networks from big data of CT images using machine learning techniques is a promising reconstruction method for achieving high quality ULDCT imaging.