Classical & AI Methods for Image Recon: From Fundamentals to Translation
Weekend Course
ORGANIZERS: Berkin Bilgic, Tolga Cukur, Yogesh Rathi
Sunday, 05 May 2024
Nicoll 2
13:15 -
17:00
Moderators: Yohan Jun & Lipeng Ning
Skill Level: Basic to Advanced
Session Number: WE-26
CME Credit
Session Number: WE-26
Overview
This weekend session focuses on classical and emerging AI-based image reconstruction strategies, including fundamentals of image encoding, compressed sensing and low-rank regularization, supervised and unsupervised deep learning models. The session includes applications in quantitative MRI as well as the translation of fast imaging techniques, and concludes with a panel discussion on the expectations from new DL methods.
Target Audience
Researchers and clinicians interested in the fundamentals and applications of classical and AI-based image reconstruction strategies, and opportunities/challenges in their clinical translation.
Educational Objectives
As a result of attending this course, participants should be able to:
- Explain the basic principles of MR image reconstruction using classical and emerging AI-based algorithms;
- Describe the mathematical principles and practical significance of advanced image reconstruction methods; and
- Discuss the potential opportunities and challenges in clinical translation of MRI recon techniques.
13:15 |  | Image Encoding in MRI & Bayesian Image Reconstruction Suyash Awate Keywords: Image acquisition: Image processing This talk presents the fundamentals underlying magnetic resonance imaging, including various parallel-imaging schemes, and the fundamentals of (Bayesian) image reconstruction methods from subsampled multicoil k-space data. |
13:40 | | Low-Rank Methods for MR Image Reconstruction Fan Lam Keywords: Image acquisition: Reconstruction Low-rank models that exploit the intrinsic redundancy in multidimensional MR signals for image reconstruction from sparse, noisy, and/or corrupted data have been widely used. These models serve as effective constraints for high-dimensional imaging problems that arise in many applications, e.g., dynamic MRI, quantitative MRI, and spectroscopic imaging. This talk will review what low-rank models are, how low-rank structures emerge or can be purposely induced from multidimensional MR data, and how they may be used in image reconstruction. Potential synergy with recent deep learning based reconstruction approaches will also be discussed. |
14:05 | | Frontiers in Image Recon: Rapid & Reproducible Quantitative MRI Nan Wang |
14:30 | | Supervised Deep Learning for MRI Recon Thomas Kuestner Keywords: Image acquisition: Reconstruction, Image acquisition: Machine learning Motivation: See synopsis Goal(s): See synopsis Approach: See synopsis Results: See synopsis Impact: See summary of main findings |
14:55 | | Break & Meet the Teachers |
15:25 | | Unsupervised Methods for Deep MRI Recon Dong Liang Keywords: Image acquisition: Reconstruction In recent years, deep learning has made significant advancements in MRI reconstruction. However, conventional methods often require full-sampled MRI data, presenting challenges in data acquisition. Consequently, unsupervised learning methods have garnered attention. This discussion delves into various unsupervised deep learning approaches for MRI reconstruction, including unpaired, self-supervised, and zero-shot (untrained) learning. Moreover, we foresee a promising future for unsupervised learning in MRI reconstruction, particularly in collaboration with large-scale foundation models, thereby facilitating further progress in MRI technology. |
15:50 |  | Opportunities & Challenges in Clinical Translation Susie Huang Keywords: Image acquisition: Reconstruction, Image acquisition: Fast imaging This lecture will provide a brief overview of the considerations involved in the clinical translation of classical and artificial intelligence approaches to image reconstruction. The systematic evaluation and validation of new reconstruction methods will be discussed in the context of fast imaging methods for acquisition and reconstruction, with an emphasis on prioritizing image quality, minimizing artifacts, and maximizing diagnostic impact. A multidisciplinary, team-based approach with close collaboration between MRI physicists, engineers, data scientists, radiologists, and radiologic technologists is of paramount importance to ensure that new reconstruction methods are integrated seamlessly into the clinical workflow. |
16:15 | | Expectations from New DL Methods: How Can Academia Contribute? Mariya Doneva Keywords: Image acquisition: Machine learning The collaboration between academia and industry is vital for advancing the research, translation, and practical implementation of new deep learning techniques for MR reconstruction. There are multiple ways in which academia can contribute, which will be discussed in this lecture. |
16:40 | | Panel Discussion |