Guanxiong Luo¹, Martin Heide¹, and Martin Uecker^1,2,3,4
1University Medical Center Göttingen, Göttingen, Germany, ²Institute of Medical Engineering, Graz, Austria, ³German Centre for Cardiovascular Research (DZHK), Partner Site Göttingen, Göttingen, Germany, ⁴Cluster of Excellence "Multiscale Bioimaging: from Molecular Machines to Networks of Excitable Cells'' (MBExC), University of Göttingen, Göttingen, Germany

This works explores the use of data-driven Markov chains that are constructed from generative models for Bayesian MRI reconstruction, where the generative models utilize prior knowledge learned from an existing image database. Given the measured k-space, samples are then drawn from the posterior using Markov chain Monte Carlo (MCMC) method. In addition to the maximum a posteriori (MAP) estimate for the image which is obtained with conventional methods, also a minimum mean square error (MMSE) estimate and uncertainty maps can be computed from these samples.

Elena Martín-González¹, Ebraham Alskaf², Amedeo Chiribiri ², Pablo Casaseca-de-la-Higuera¹, Carlos Alberola-López¹, Rita G. Nunes^3,4, and Teresa Correia^2,5
1Laboratorio de Procesado de Imagen, Universidad de Valladolid, Valladolid, Spain, ²School of Biomedical Engineering and Imaging Sciences, King’s College London, London, United Kingdom, ³Institute for Systems and Robotics, Lisbon, Portugal, ⁴Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa, Lisbon, Portugal, ⁵Centre for Marine Sciences - CCMAR, Faro, Portugal

First-pass perfusion cardiac magnetic resonance (FPP-CMR) is becoming essential to detect blow flow anomalies. However, the need for real-time acquisitions limits the achievable spatial resolution and coverage of the heart. To keep both within a reasonable range, FPP-CMR needs to be accelerated. A SElf-Supervised aCcelerated REconsTruction (SECRET) DL framework is presented to speed-up reconstruction of FPP-CMR images from undersampled (k,t)-space data. The physical reconstruction models are used to train deep neural networks without requiring fully sampled images. SECRET achieves good quality reconstructions at a variety of acceleration rates, with significant speed-ups compared to the state-of-the-art.

Hamza Kebiri^1,2, Erick J. Canales-Rodríguez³, Priscille de Dumast^1,2, Athena Taymourtash⁴, Hélène Lajous^1,2, Yasser Alemán-Gómez², Georg Langs⁴, and Meritxell Bach Cuadra^1,2,3
1CIBM Center for Biomedical Imaging, Lausanne, Switzerland, ²Department of Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ³Signal Processing Laboratory 5 (LTS5), Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ⁴Computational Imaging Research Lab, Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna, Austria

Although super-resolution diffusion MRI for isotropic volumes has been explored, to date no unsupervised SR techniques have been investigated for anisotropic dMRI. We propose an autoencoder based framework to enhance the through-plane spatial resolution and to replace slice outliers by leveraging existing high-quality datasets. Quantitative evaluation on 31 pre-term subjects show that the proposed framework significantly outperforms conventionally used interpolation methods at the raw data and estimated diffusion tensor maps. This can hence contribute to the depiction of more accurate white matter properties of the developing brain.

Harris Beg^1,2, Beliz Gunel^2,3, Batu M Ozturkler^2,3, Christopher M Sandino³, John M Pauly³, Shreyas Vasanawala⁴, Akshay S Chaudhari^4,5, and Arjun D Desai^3
1Computing and Mathematical Sciences, California Institute of Technology, Pasadena, CA, United States, ²Equal Contribution, Stanford University, Stanford, CA, United States, ³Electrical Engineering, Stanford University, Stanford, CA, United States, ⁴Radiology, Stanford University, Stanford, CA, United States, ⁵Biomedical Data Science, Stanford University, Stanford, CA, United States

In this study, we propose Motion2Recon, a semi-supervised consistency-based approach for robust accelerated MR reconstruction of motion-corrupted images. Motion2Recon reduced dependence on fully-sampled (supervised) training data and improves reconstruction performance among motion-corrupted scans. It also maintained superior performance among non-motion, in-distribution scans, which may help eliminate the need for manual motion detection. All code and experimental configurations are openly available in Meddlr (https://github.com/ad12/meddlr).

Zihao Chen^1,2, Yibin Xie¹, Debiao Li^1,2, Yijen Wu³, and Anthony Christodoulou^1,2
1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ²Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States, ³Department of Developmental Biology, University of Pittsburgh, Pittsburgh, PA, United States

Deep learning MR super-resolution (SR) is a promising approach to reduce scan time without requiring custom sequences and iterative reconstruction. However, previous SR approaches are incompatible with turbo spin echo (TSE) sequence due to differences in T2 blurring effects between high-resolution and low-resolution TSE images. Here we propose a T2-deblurred deep learning SR model for 3x3 in-plane super-resolution of 3D TSE images. Our method accelerated scanning by 9x in both retrospective and prospective testing and provided better image quality than previous SR methods.  

Katja Lauer^1,2, Jonas Kleineisel¹, Alfio Borzì², Thorsten Alexander Bley¹, Herbert Köstler¹, and Tobias Wech^1
1Department of Diagnostic and Interventional Radiology, University Hospital Würzburg, Würzburg, Germany, ²Institute of Mathematics, University of Würzburg, Würzburg, Germany

Data-driven reconstruction of undersampled raw data has gained more and more importance in recent years. Due to the non-linear and non-stationary transform characteristics of these imaging methods, objective image quality assessment is difficult. We propose a heuristic approach based on local point spread functions and multiple replica reconstructions, to enable the derivation of resolution- and g-factor-maps for individual images. The method is exemplarily applied in T1- and T2-weighted images of the brain, using a UNet and a Variational Network trained with data from the fastMRI project.

Thomas Yu^1,2, Tom Hilbert^1,3,4, Gian Franco Piredda^1,3,4, Erick Canales-Rodrıguez¹, Tobias Kober^1,3,4, and Jean-Philippe Thiran^1,4
1Signal Processing Lab 5 (LTS5), Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ²Medical Image Analysis Laboratory, Center for Biomedical Imaging (CIBM), University of Lausanne, Lausanne, Switzerland, ³Advanced Clinical Imaging Technology, Siemens Healthcare, Lausanne, Switzerland, ⁴Department of Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland

Self-supervised reconstruction methods for undersampled acquisitions are becoming increasingly used. We compare different self-supervised reconstruction methods using fully sampled and prospectively/retrospectively accelerated data; we find that prospective and retrospective reconstructions can differ significantly in quantitative metrics and perceptual quality. To test the methods’ generalizability, prospectively accelerated data from multiple field strengths is reconstructed without retraining/retuning. We find that no-reference image quality metrics can distinguish state of the art methods from the baseline, albeit with ambiguity between the state of the art methods. 

Maarten Terpstra^1,2, Matteo Maspero^1,2, Joost JC Verhoeff¹, and Cornelis A.T. van den Berg^1,2
1Deparment of Radiotherapy, University Medical Center Utrecht, Utrecht, Netherlands, ²Computational Imaging Group for MR Diagnostics & Therapy, University Medical Center Utrecht, Utrecht, Netherlands

Four-dimensional (4D) respiratory-resolved imaging is crucial for managing respiratory motion in radiotherapy, enabling irradiation of highly mobile tumours. However, acquiring high-quality 4D-MRI requires long acquisition (typically ≥5 minutes) and long iterative reconstructions, limiting treatment efficiency. Recently, deep learning has been proposed to accelerate undersampled MRI reconstruction. However, it has not been established whether deep learning may reconstruct high-quality 4D-MRI from accelerated acquisitions. This work proposes a small deep learning model that exploits the spatio-temporal information present in 4D-MRI, allowing to split the reconstruction into two separated branches, enabling high-quality retrospectively-accelerated 4D-MRI acquired in ~60 seconds and reconstructed in 16 seconds.

Sudhanya Chatterjee¹, Naoyuki Takei², Rohan Patil¹, Sugmin Gho³, Suchandrima Banerjee⁴, Florian Wiesinger⁵, and Dattesh Dayanand Shanbhag^1
1GE Healthcare, Bengaluru, India, ²GE Healthcare, Tokyo, Japan, ³GE Healthcare, Seoul, Korea, Republic of, ⁴GE Healthcare, Menlo Park, CA, United States, ⁵GE Healthcare, Munich, Germany

Magnetic resonance image compilation (MAGiC) is a single click scan that provides multiple contrast-weighted images in around 5 mins. This makes it an effective diagnostic option in clinical settings. However, synthetic T₂-FLAIR is known to have partial volume artifacts which impacts its diagnostic performance. In this work, we propose a method to mitigate partial-volume artifacts in synthetic T₂-FLAIR using a separately acquired fast T₂-FLAIR contrast information combined with keyhole and deep learning-based image reconstruction.