Programme Overview Session: ES 02

Learning Objective

To obtain a structured understanding of the field of biomedical imaging from the perspective of inverse problems. Specifically, to form an understanding of

• the basic building blocks of any imaging problem,
• the taxonomy of image reconstruction algorithms in 1st (classical), 2nd (sparsity-based) and 3rd (deep learning) generation techniques, together with their respective foundations, advantages, and disadvantages,
• how to prototype reconstruction techniques using GlobalBioIm, a convenient MATLAB library to accelerate development in the field.

Content

The impact of biomedical imaging has increased steadily over the past four decades. Part of this is due to the improvement of reconstruction methods, which have provided increasing image quality and resolution. In this tutorial, we present a well-structured view of the field of biomedical image reconstruction based on the few basic building blocks of any imaging system and the three generations of available methods, i.e., classical, sparsity-based, and deep neural networks techniques. For any given imaging problem, attendants will learn to quickly identify its fundamental building blocks from an inverse-problems perspective and propose adequate methodology. Furthermore, we will introduce how to obtain fast prototypes of reconstruction techniques using GlobalBioIm, an efficient MATLAB library tailored to image reconstruction problems.

Relevant Publications

M.T. McCann, M. Unser, "Biomedical Image Reconstruction: From the Foundations to Deep Neural Networks," Foundations and Trends^® in Signal Processing, vol. 13, no. 3, pp. 280-359, December 3, 2019.

E. Soubies, F. Soulez, M.T. McCann, T.-a. Pham, L. Donati, T. Debarre, D. Sage, M. Unser, "Pocket Guide to Solve Inverse Problems with GlobalBioIm," Inverse Problems, vol. 35, no. 10, paper no. 104006, pp. 1-20, October 2019.

Learning Objective

The talk will provide an introduction to image reconstruction algorithms applied in Nuclear Medicine (PET and SPECT).

Content

The topics that will be covered are e.g. filtered backprojection, iterative image reconstruction, maximum likelihood expectation maximization (MLEM), ordered subset expectation maximization (OSEM), resolution modeling (PSF), iterative reconstruction algorithms that aim for absolute SPECT quantification and recent advances in image reconstruction that apply Artificial Intelligence (AI) techniques, such as deep learning.

Relevant Publications

Learning Objective

• To understand magnetic image reconstruction as an inverse problem
• To recognise conventional MRI reconstruction methods such as parallel imaging and compressed sensing and latest advances in the field.

Content

Long acquisition times are still a limitation for many applications of Magnetic Resonance Imaging (MRI). Several undersampling reconstruction techniques have been proposed over the last decades to overcome this problem. These techniques are based on acquiring less samples than the specified by the Nyquist rate, and estimating the non-acquired data by using some sort of assumption or prior information.

Parallel imaging allows to decrease the scan time by reducing the number of phase increment steps (undersampling) and exploiting the sensitivity encoding of the multiple receiver coils to recover the non-acquired data. Parallel imaging has been widely integrated into commercial MR systems and is routinely used in clinical practice. Undersampled reconstruction techniques such as compressed sensing (CS) have been also employed to accelerate MRI. CS works under the assumption that the k-space data is randomly undersampled, the image has a sparse representation in some pre-defined basis or dictionary, and a non-linear reconstruction is performed to enforce the sparsity of the image and consistency with the acquired MR data. In practice, CS-based reconstruction techniques employ pseudo-random trajectories along with one or several (e.g., spatial and temporal dimensions) sparse transforms such as finite differences (e.g., total variation) or wavelets operators. Recent efforts have been made to further improve CS- based reconstruction quality by learning dictionary-based representations of the sparse domain from the acquired data itself (or jointly during reconstruction) instead of exploiting known analytical transform domains. However, CS-based reconstruction techniques usually suffer from long computational times and their performance depends on the choice of the sparsity representation and the tuning of the corresponding reconstruction parameters. More recently covolutionals neural networks based techniques have been proposed to overcome these challenges by learning optimal reconstruction parameters. This talk will include an introduction to undersampled MRI reconstruction as well as an overview of the latest advances in the field, discussing their strengths and limitations.

Relevant Publications

1. K.P. Pruessmann, M. Weiger, M.B. Sheidegger and P. Boesiger. SENSE: Sensitivity En- coding for Fast MRI. Mag. Res. Med., 42(5):952-962, 1999.
2. M. Lustig, D. Donoho and J.M. Pauly. Sparse MRI: The application of Compressed Sensing for Rapid MR Imaging. Mag. Res. Med., 58(6):1182-1195, 2007.
3. ​​​​​​​Bustin A, Fuin N, Botnar RM and Prieto C (2020) From Compressed-Sensing to Artificial Intelligence-Based Cardiac MRI Reconstruction. Front. Cardiovasc. Med. 7:17. doi: 10.3389/fcvm.2020.00017.

Learning Objective

1. To appreciate hybrid imagign as a mean towards image fusion
2. To aprpeciate the pitfalls in hybrid imaging that impact co-registration, and, thus image fusion
3. To understand key correction schemes for better image fusion
4. To conceive of future concepts of image fusion beyond hybrid imaging

Content

This presentation will address challenges in fusing complementary clinical image information by using hybrid imaging technologies, that is physically combined dual-modality tomographs. In theory, these systems provide “one-stop-shop” imaging of patients without the need to reposition subjects between examinations. However, even in this context, image fusion can be challenged by involuntary patient motion, for example. Likewise, there is a need to co-register and fuse longitudinal images that are acquired as part of therapy follow-up schemes. Hence, hybrid imaging as such does not resolve the task of accurate image fusion a priori.

Ideally, the audience will become familiar with the state-of-the-art image fusion concept through hybrid imaging, with pitfalls in accurate image fusion from intrinsic mis-registration, with solutions to this problem. Finally, the presentation will highlight areas of ongoing research activity towards multi-parametric image fusion and data analysis by means of combined software and hardware fusion concepts.

Relevant Publications

pubmed - pls