Andrzej Liebert¹, Lorenz Kapsner¹, Lukas Folle², Hannes Schreiter^1,2, Badhan Kumar Das^1,2, Sabine Ohlmeyer¹, Andreas Maier², Evelyn Wenkel¹, Frederik B. Laun¹, Michael Uder¹, and Sebastian Bickelhaupt^1
1Institute of Radiology, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany, ²Pattern Recognition Lab, Department of Computer Science, Friedrich-Alexander University Erlangen Nuremberg, Erlangen, Germany

DWI acquisitions in MRI are prone to artifacts caused e.g. by insufficient fat saturation and can significantly impede the diagnostic assessment. We investigated the ability to predict the occurrence of these artifacts in maximum intensity projections (MIPs) of high b-value DWI using a neuronal network that analyses T2w images, which were acquired prior to the DWI sequence. An AuRoC of 0.83 with sensitivity of 0.82 and specificity of 0.61 was achieved. Investigation of the regions of the T2w images most important for the decision of the neural network showed a good correlation with artifact-affected areas in the DWI MIP images.

Moss Zhao¹, Ramy Hussein¹, Michael Moseley¹, and Greg Zaharchuk^1
1Department of Radiology, Stanford University, Stanford, CA, United States

We present a deep learning technique to predict cerebral perfusion after vasodilation challenges. A 3D convolutional neural network (CNN)-based encoder-decoder architecture was constructed to transform ASL perfusion images acquired pre-vasodilation into post-vasodilation images using an improved attention-gated 3D U-Net. Results showed that the prediction and ground truth were not significantly different. This technique will enable a drug-free MR procedure to study the hemodynamic of patients with high risk cerebrovascular diseases.

Andreas Voskrebenzev^1,2, Cristian Crisosto^1,2, Maximilian Zubke^1,2, Frank Wacker^1,2, and Jens Vogel-Claussen^1,2
1Institute of Diagnostic and Interventional Radiology, Hannover Medical School, Hannover, Germany, ²German Center for Lung Research (DZL), Biomedical Research in Endstage and Obstructive Lung Disease Hannover (BREATH), Hannover, Germany

The partial volume effect (PVE) is an often-observed artifact in MR imaging. Especially images with a low spatial resolution, will show an averaged voxel signal of multiple tissue components. These artifacts can be so substantial that a further image analysis can be omitted.  This is e.g. the case for phase-resolved functional lung imaging (PREFUL), which is based on the 2D acquisition of coronal image-time-series to assess ventilation and perfusion dynamics. In this study the pretrained convolutional neural network Inception-v3 was trained via transfer-learning to detect images, which show substantial PVE with a classification accuracy of 91%.

Benjamin Lambert^1,2, Florence Forbes³, Alan Tucholka², Senan Doyle², and Michel Dojat^1
1Univ. Grenoble Alpes, Inserm, U1216, Grenoble Institut Neurosciences, GIN, 38000, Grenoble, France, ²Pixyl, Research and Development Laboratory, 38000, Grenoble, France, ³Univ. Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK, 38000, Grenoble, France

Efforts are required to design Deep Learning models that are not only powerful, but also capable of expressing the certainty of their predictions.
We evaluate 3 state-of-the-art techniques for uncertainty quantification : Monte-Carlo Dropout, Deep Ensemble, and Heteroscedastic models. Evaluation is illustrated on a task of automatic segmentation of White-Matter Hyperintensities in T2-weighted FLAIR MRI sequences of Multiple-Sclerosis patients. Analysis is performed at 3 different scales : the voxel, the lesion, and the whole image. Results indicate the superiority of the Heteroscedastic approach, which ranked first in both the uncertainty and segmentation tasks.

 

Berardino Barile¹, Claudio Stamile², Sabine Van Huffel³, and Dominique Sappey-Marinier^4
1CREATIS (UMR5220 CNRS & U1294 INSERM, Université Claude Bernard-Lyon1 & INSA-Lyon, France, Villeurbanne, France, ²R&D Department CGnal Milan, Italy, Villeurbanne, France, ³KU Leuven, Leuven, Belgium, ⁴CREATIS (UMR5220 CNRS & U1294 INSERM), Université Claude Bernard-Lyon1 & INSA-Lyon, France, Villeurbanne, France

Multiple Sclerosis (MS) is an autoimmune inflammatory disease characterized by white matter (WM) lesions and gray matter (GM) degeneration leading to cognitive and physical impairments. Recently, brain connectivity techniques were developed to better characterize such brain damages in association with the clinical profiles of MS patients. In this study, we proposed a complete automated pipeline to first extract global graph metrics from the morphological brain connectivity based on GM tissue thickness and second to train a higher-level ensemble of four machine learning models for the classification of MS clinical profiles using only the T1-weighted image modality.

Masoumeh Javanbakhat¹, Ludger Starke², Sonia Waiczies², and Christoph Lippert^1
1Digital Health-Machine Learning Group, Hasso Plattner Institute, Potsdam, Germany, ²Berlin Ultrahigh Field Facility, Max Delbrück Centre for Moleculare Medicine in the Helmholtz Association, Berlin, Germany

Deep learning (DL) has achieved state of the art results in semantic segmentation of numerous medical imaging applications. Despite promising results deep learning models tend to produce point estimates as outputs which leads to overconfident, miscalibrated predictions. These overconfident predictions are specifically problematic in medical applications. Hence, providing a measure of a system’s confidence to identify untrustworthy predictions is essential to guide clinical decisions. Here we propose a 3D Bayesian segmentation model to provide uncertainty estimation for the Fluorine-19 MRI dataset based on Stochastic Gradient Markov Chain Monte Carlo methods.

Balint Kincses^1,2, Katarina Forkmann¹, Katharina Schmidt¹, Ulrike Bingel¹, and Tamas Spisak^2
1Bingel-laboratory, Department of Neurology, University Hospital Essen, Essen, Germany, ²Laboratory of Predictive NeuroImaging, Institute for Diagnostic and Interventional Radiology and Neuroradiology, University Hospital Essen, Essen, Germany

Fear conditioning has a role in anxiety-disorders and the neurobiological correlates of it are not yet well understood. Therefore, we trained a machine learning predictive model on individual functional resting state connectivity data to predict the emotional aspects of fear conditioning. The model was found to predict individual pain-related threat learning measured by the change of valence with an explained variance of 24%-41%. These results highlight the potential of machine learning to enhance our understanding of fear conditioning.

Elisa Colato¹, Claudia AM Wheeler-Kingshott^1,2,3, Douglas L Arnold⁴, Frederik Barkhof^1,5,6,7, Olga Ciccarelli^1,5, Declan Chard^1,5, and Arman Eshaghi^1,8
1Queen Square Multiple Sclerosis Centre, Department of Neuroinflammation, Queen Square Institute of Neurology, Faculty of Brain Sciences, University College London (UCL), London, United Kingdom, ²Brain MRI 3T Research Centre, C. Mondino National Neurological Institute, Pavia, Italy, ³Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy, ⁴McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University, Montreal, QC, Canada, ⁵Institute for Health Research (NIHR), University College London Hospitals (UCLH) Biomedical Research Centre (BRC), London, United Kingdom, ⁶Department of Radiology and Nuclear Medici, VU medical centre, Amsterdam, Netherlands, ⁷Centre for Medical Image Computing (CMIC), Department of Medical Physics and Biomedical Engineering, University College London (UCL), London, United Kingdom, ⁸Centre for Medical Image Computing (CMIC), Department of Computer Science, University College London (UCL), London, United Kingdom

Network-based measures can outperform regional and whole-brain grey matter (GM) measures in explaining clinical disability in several neurodegenerative disorders. However, network measures are mostly estimated at the group level and require a re-estimation of model parameters when applied to new participants. Here, we introduce a new longitudinal network analysis paradigm to extract longitudinal ICA-like components at an individual level from a discovery cohort, applied machine learning to obtain individual-level network-based measure for a validation cohort, and used them to explain clinical disability in multiple sclerosis.  

Chuyu Liu¹, Shuo Chen¹, Yibing Chen², Rui Li¹, and Xiaolei Song^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, Tsinghua University, Beijing, China, ²Xi’an Key Lab of Radiomics and Intelligent Perception, School of Information Sciences and Technology, Northwest University, Xi'an, China

Serial image acquisition is always required for quantitative and dynamic MRI, in which the spatial and temporal resolution as well as the contrast is restricted by acquisition speed. Herein we employed a novel transformer based deep-learning method for reconstruction of highly under-sampled MR serial images. The network architecture and input were designed to take advantages of Transformer’s capacity on global information learning along the redundant series dimension. SMR-transformer was compared with 3D Res-U-Net, on a self-acquired CEST dataset, a cardiac dataset from OCMR and a diffusion dataset from HCP. Both images and quantitative comparison indicated SMR-Transformer allowed fast serial MRI.