Speakers: Ulugbek S. Kamilov - Speaker Website
Washington University in St. Louis
Singanallur V. Venkatakrishnan - Speaker Website
Oak Ridge National Laboratory (ORNL)
Image reconstruction refers to a diverse family of computational methods used to form or recover an unknown image given a set of measurements collected through some imaging system. Image reconstruction in a critical part of modern biomedical imaging systems including magnetic resonance imaging (MRI), computerized tomography (CT), optical diffraction tomography (ODT), and positron emission microscopy (PET). The design of image reconstruction methods is a fascinating engineering problem, due to measurements collected by instruments being noisy, very high-dimensional, and often incomplete. Image reconstruction can also be a significant computational challenge, especially in the context of large-scale problems typical in modern 3D/4D biomedical imaging applications.
There has been a significant recent interest in machine learning within the biomedical imaging community. The goal of our tutorial “Image Reconstruction Methods in Biomedical Imaging: Reconciling Models and Learning” is to trace the evolution of image reconstruction from the classical model-based techniques, such as Tikhonov and total variation regularization, to the modern algorithms based on deep learning. We will discuss Plug-and-Play Priors (PnP) and Regularization by Denoising (RED) frameworks that enable combination of the physics-based forward models with the priors specified through deep neural net denoisers. We will discuss the theoretical foundation of the corresponding algorithms by going beyond traditional optimization and adopting monotone operator theory. We believe that this topic is very timely for ISBI 2020, and many of the advanced results that we will cover in this tutorial will benefit the wider biomedical imaging community. Additionally, the tutorial format will enable us to cover this rich area in a way that will be accessible to many researchers in the field.