View Presentation Video

Carole Helene Sudre^1,2, Haroon R Chughtai², David L Thomas^3,4, David Cash⁴, Miguel Rosa-Grilo⁴, Millie Beament⁴, Frederik Barkhof^2,4,5, Daniel Alexander², Cath Mummery⁴, Nick Fox⁴, and Geoff JM Parker^2
1MRC Unit for Lifelong Health and Ageing at UCL, University College London, London, United Kingdom, ²Centre for Medical Image Computing, University College London, London, United Kingdom, ³Department of Brain Repair and Rehabilitation, UCL Queen Square Institute of Neurology, University College London, London, United Kingdom, ⁴Dementia Research Centre, UCL Queen Square Institute of Neurology, University College London, London, United Kingdom, ⁵Department of Radiology & Nuclear Medicine, Amsterdam UMC, Amsterdam, Netherlands

Keywords: Segmentation, Dementia

Acquisition time is a crucial factor in the tolerability and availability of MRI examinations for patients attending memory clinics. The suitability of using prototype ultra-rapid sequences for the quantification of white matter hyperintensities was assessed by comparing segmentation outputs from standard clinical and ultra-rapid sequences. Despite a slightly lower sensitivity to smaller lesions, quantification of white matter hyperintensities when using ultra-rapid sequences was very highly correlated with lesion volumes obtained from standard sequences. 

View Presentation Video

Maximilian Sackl¹, Christian Tinauer¹, Christian Enzinger¹, Reinhold Schmidt¹, and Stefan Ropele^1
1Department of Neurology, Medical University Graz, Graz, Austria

Keywords: Segmentation, Alzheimer's Disease, Multi-Contrast, AI

Segmentation of the hippocampus on T1-weighted structural MR images is required to quantify the neurodegenerative effects in Alzheimer’s disease studies. In this work, we propose an automated artificial intelligence-based pipeline for hippocampus segmentation combined with manual ground truth (GT) data that originates from high-resolution T2-weighted MR images. Results are evaluated against the manual GT-labels and compared to the segmentation results from FreeSurfer v732. Our deep learning-based segmentation outperforms FreeSurfer in terms of accuracy and speed, while reference experiments using the T2-based GT-labels yield the best results. Thus, using T2-weighted images for ground truth generation can improve automated HC segmentation.

View Presentation Video

Huixiang Zhuang¹, Yue Guan¹, Yi Ding¹, Yuhao Ma¹, Yunpeng Zhang¹, Ziyu Meng¹, Ruihao Liu^1,2, Zhi-Pei Liang^2,3, and Yao Li^1
1School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ²Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States

Keywords: Segmentation, Machine Learning/Artificial Intelligence, Unsupervised learning; Lesion segmentation

Unsupervised segmentation of brain lesions is desirable in many applications and has been investigated extensively. In this work, we proposed a new method for brain lesion segmentation, which effectively learns the spatial-intensity distribution of normal brain tissues and then treats lesion segmentation as an anomaly detection problem. We overcame the high-dimensional distribution learning problem using a subspace-assisted generative network. With the learned distribution, the anomaly detection problem was solved using Bayesian hypothesis testing. Our method has been validated using simulated and real brain MR images with stroke and tumor lesions, and produced significantly improved results than several state-of-the-art methods.

View Presentation Video

Sebastian Hübner¹, Stefano Tambalo¹, Tobias Kober^2,3,4, and Jorge Jovicich^1
1Center for Mind/Brain Sciences, University of Trento, Rovereto (Trento), Italy, ²Advanced Clinical Imaging Technology, Siemens Healthineers International AG, Lausanne, Switzerland, ³Department of Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ⁴LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

Keywords: Segmentation, Brain, Thalamus

Accurate segmentation of thalamic nuclei can significantly support clinical decisions, especially in diseases such as Parkinson’s. At the same time, highly accelerated T1w acquisition techniques find wider clinical acceptance as they not only increase radiological efficiency and patient comfort, but also reduce motion artefacts. Here, we study how compressed sensing affects thalamic nuclei volumetry based on MPRAGE or MP2RAGE in healthy subjects with respect to clinical baseline. Segmentation results are promising and agree with expectations: acceleration to about 1 min is possible maintaining overall good image quality, contrast and thalamic segmentations, with acceleration-related biases.
 

View Presentation Video

Yuan-Chiao Lu^1,2, Blake E Dewey³, Yi-Yu Chou^1,2, Danielle Greenman^1,2, Russell T Shinohara⁴, Daniel S Reich⁵, Jerry L Prince⁶, John A Butman², and Dzung L Pham^2,7
1Center for Neuroscience and Regenerative Medicine, Henry M Jackson Foundation for the Advancement of Military Medicine, Bethesda, MD, United States, ²Radiology and Imaging Sciences, National Institutes of Health, Bethesda, MD, United States, ³Neurology, Johns Hopkins University, Baltimore, MD, United States, ⁴Department of Biostatistics, Epidemiology and Informatics, University of Pennsylvania, Philadelphia, PA, United States, ⁵Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, Bethesda, MD, United States, ⁶Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, United States, ⁷Radiology and Radiological Sciences, Uniformed Services University, Bethesda, MD, United States

Keywords: Segmentation, Brain, Harmonization

Image harmonization approaches have been proposed for reducing the variation of brain image measurements in studies involving acquisitions from multiple scanner protocols and hardware. This study compares two harmonization methods, DeepHarmony, a deep learning-based image synthesis approach, and ComBat, a statistical batch correction tool, based on their ability to yield consistent brain volume measurements from two different T1-weighted acquisitions. Our study showed that DeepHarmony outperformed the ComBat approach, although both approaches significantly improved consistency when compared to unharmonized images.

View Presentation Video

Kilian Hett¹, Colin D. McKnight², Jennifer S. Lindsey², Melanie Leguizamon¹, Jarrod Eisma¹, Alexander K. Song¹, Jason Elenberger¹, Ciaran M. Considine¹, Daniel O. Claassen¹, and Manus J. Donahue^1
1Department of Neurology, Vanderbilt University Medical Center, Nashville, TN, United States, ²Department of Radiology, Vanderbilt University Medical Center, Nashville, TN, United States

Keywords: Segmentation, Neurofluids

The overarching goal of this work is to develop and validate novel deep learning algorithms for segmenting the peri-sinus space, including parasagittal dural (PSD) space, which has been hypothesized to harbor cerebral lymphatic channels, and intra-veinous arachnoid granulations, which has been long-hypothesized as a site a CSF egress, from standard non-contrast anatomical imaging. The new segmentation method is based on cascaded neural networks using non-contrasted 3D T2-weighted MRI; the method is method in a mixed cohort of adults with and without neurodegeneration.

View Presentation Video

Julie P. Vidal^1,2, Lola Danet^2,3, Patrice Péran², Jérémie Pariente^2,3, Dolf Pfefferbaum⁴, Edith Sullivan⁴, Emmanuel J. Barbeau¹, and Manojkumar Saranathan^5
1CNRS, CerCo (Brain and Cognition Research Center) - Université Paul Sabatier, Toulouse, France, ²INSERM, ToNiC (Toulouse NeuroImaging Center) - Université Paul Sabatier, Toulouse, France, ³Hôpital Purpan, Centre Hospitalier Universitaire de Toulouse, Département de Neurologie, Toulouse, France, ⁴Stanford University School of Medicine, Department of Psychiatry & Behavioral Sciences, Stanford, CA, United States, ⁵UMass Chan Medical School, Department of Radiology, Worcester, MA, United States

Keywords: Segmentation, Data Processing, WMn synthesis, HIPS

This work presents a methodology for thalamic nuclei segmentation from T1w MRI. Thalamus Optimized Multi-Atlas Segmentation (THOMAS), which was originally developed for white-matter nulled (WMn) MPRAGE, is adapted for standard T1 MRI by employing synthesis techniques to make the T1 images closer to WMn contrast. Robustness is tested across image contrast, MRI manufacturer, and field strength.  

View Presentation Video

Lidia Luque^1,2,3, Karoline Skogen⁴, Bradley J MacIntosh^3,5,6, Kyrre Eeg Emblem¹, Christopher Larsson^3,7, Einar O Vik-Mo⁷, and Atle Bjørnerud^1,3
1Department of Physics and Computational Radiology, Oslo University Hospital, Oslo, Norway, ²Department of Physics, University of Oslo, Oslo, Norway, ³Computational Radiology and Artificial Intelligence (CRAI), Oslo University Hospital, Oslo, Norway, ⁴Department of Radiology and Nuclear Medicine, Oslo University Hospital, Oslo, Norway, ⁵Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada, ⁶Sandra E Black Centre for Brain Resilience and Recovery, Sunnybrook Research Institute, Toronto, ON, Canada, ⁷Department of Neurosurgery, Oslo University Hospital, Oslo, Norway

Keywords: Segmentation, Machine Learning/Artificial Intelligence, Semi-supervision

Improving automatic segmentation of glioblastoma on early post-operative MRI is key to study the effect of resection volumes on patient outcomes. We curate a dataset of over 700 MRI examinations, of which 87 include annotations, and train a supervised and a semi-supervised deep-learning model. Semi-supervision improves the segmentation of the high-intensity FLAIR signal with 3% to a Dice score of 0.83 (p=0.031), while the segmentation of the enhancing tumor increases with 9% to 0.55 (p=0.056). However, enhancing tumor segmentations show high variability, possibly due to imperfect annotations. Segmentation of enhancing tumor on early post-operative MRI remains a challenging task.

View Presentation Video

ross shaw¹, Penny Gowland¹, and Richard Bowtell^2
1University of Nottingham, Nottingham, United Kingdom, ²Physics, University of Nottingham, Nottingham, United Kingdom

Keywords: Segmentation, Contrast Mechanisms, Thalamus

The Thalamic Reticular Nucleus is a micro-structure surrounding the dorsal edge of the thalamus that is of fundamental significance to understanding many neurological disorders in the brain. Here we present methods to visualise the TRN using various imaging modalities to explore different sources of contrast at the edge of the thalamus.

View Presentation Video

Marc-Antoine Fortin¹, Yannik Völzke², Rüdiger Stirnberg², Siya Sherif³, Laurent Lamalle³, Tony Stöcker^2,4, and Pål Erik Goa^1
1Department of Physics, Norwegian University of Science and Technology (NTNU), Trondheim, Norway, ²German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, ³GIGA Cyclotron Research Centre (GIGA-CRC) in vivo imaging, University of Liège, Liège, Belgium, ⁴Department of Physics and Astronomy, University of Bonn, Bonn, Germany

Keywords: Segmentation, High-Field MRI, Multi-Parameter Mapping, Quantitative MRI, 7T

Very few segmentation techniques are optimized for 7T images. Due to the lack of approaches and the high level of inhomogeneity observed with Ultra-High Field images, neuroscientists typically limit the analysis to the upper half of the brain. A novel resolution- and contrast-agnostic technique, called SynthSeg, may remedy this limitation. A multi-parameter mapping protocol and MPRAGE images were acquired to test SynthSeg against FreeSurfer on 7T sub-millimeter brain images. Our results showed that SynthSeg can surpass FreeSurfer, especially for the cerebellum and temporal lobes. Moreover, we showed that quantitative maps can be great surrogates to high-resolution T1w images like MPRAGE. 

View Presentation Video

Rozemarijn M. Mattiesing¹, Serena Stel¹, Alysha S. Mangroe¹, Iman Brouwer¹, Adriaan Versteeg¹, Ronald A. van Schijndel¹, Bernard M.J. Uitdehaag², Frederik Barkhof^1,3, Hugo Vrenken¹, and Joost P.A. Kuijer^1
1MS Center Amsterdam, Radiology and Nuclear Medicine, Amsterdam Neuroscience, Amsterdam UMC location VUmc, Amsterdam, Netherlands, ²MS Center Amsterdam, Neurology, Amsterdam Neuroscience, Amsterdam UMC location VUmc, Amsterdam, Netherlands, ³UCL London, Institutes of Neurology and Healthcare Engineering, London, United Kingdom

Keywords: Segmentation, Multiple Sclerosis

Monitoring changes in white matter lesions with MRI is important to evaluate the effects of treatment in multiple sclerosis. In this study a validation of a semi-automated method to quantify lesion volume changes based on 2D proton-density-weighted images and image subtraction was performed. With this method new and enlarging but also disappearing and shrinking lesion activity can be quantified. As assessed with the intraclass correlation coefficient for absolute agreement, we found that the reproducibility was excellent and the accuracy was good overall. This semi-automated subtraction method can reliably quantify lesion volume changes in patients with (early) multiple sclerosis.

View Presentation Video

Gaurav Verma¹, Ki-Sueng Choi¹, Helen Mayberg², and Priti Balchandani^1
1Radiology, Icahn School of Medicine at Mount Sinai, New York, NY, United States, ²Neurology, Icahn School of Medicine at Mount Sinai, New York, NY, United States

Keywords: Segmentation, High-Field MRI, FreeSurfer, MDD

Automatic segmentation was performed on T1-MPRAGE structural MRI data acquired at 3T and 7T from 37 and 69 distinct healthy controls, respectively. Additionally, segmentation was performed on imaging acquired from 215 major depressive disorder (MDD) patients at 3T and 40 MDD patients at 7T. Of 259 segmentation-derived imaging features evaluated, 120 showed significant 3T vs. 7T differences among controls, and 153 among patients. 7T imaging metrics showed consistently lower cortical thickness and cortical gray/white matter ratios. Subcortical and cortical volumes measured at 7T were more mixed, with 7T images showing greater frontal lobe volume, but lower cortical volumes elsewhere.

View Presentation Video

David Rudolf van Nederpelt¹, Samantha Noteboom², Eva M.M. Strijbis³, Iman Brouwer¹, Bastiaan Moraal¹, Bas M.S. Jasperse¹, Henk-Jan M.M. Mutsaerts¹, Menno M. Schoonheim², Frederik Barkhof^1,4, Hugo Vrenken¹, and Joost P.A. Kuijer^1
1Radiology and nuclear medicine, MS Center Amsterdam, Radiology and Nuclear Medicine, Vrije Universiteit Amsterdam, Amsterdam Neuroscience, Amsterdam UMC location VUmc, Amsterdam, The Netherlands, Amsterdam, Netherlands, ²Anatomy and Neurosciences, MS Center Amsterdam, Anatomy and Neurosciences, Vrije Universiteit Amsterdam, Amsterdam Neuroscience, Amsterdam UMC location VUmc, Amsterdam, The Netherlands, Amsterdam, Netherlands, ³MS Center Amsterdam, Neurology, Vrije Universiteit Amsterdam, Amsterdam Neuroscience, Amsterdam UMC location VUmc, Amsterdam, The Netherlands, Amsterdam, Netherlands, ⁴Queen Square Institute of Neurology and Centre for Medical Image Computing, University College London, UK, London, United Kingdom

Keywords: Segmentation, Multiple Sclerosis

 New methods have enabled segmentation on sequences other than T1 for brain volume measurements in multiple sclerosis (MS), such as those more readily available in clinic as FLAIR and PD. However, reliability studies have not been performed yet in MS, limiting their use. Here, we assessed within-scanner and between-scanner reliability of FLAIR segmentations in 30 people with MS on three different MR scanners. We found similar reliability compared to T1w scans, with high intra-scanner reliability but systematic differences between scanners. This suggests that within-scanner volume measurements of FLAIR are possible, but standardization is needed for between-scanner single patient measurements.

View Presentation Video

Amritha Nayak^1,2, An N Dang Do³, Audrey E Thurm⁴, Ariane Soldatos⁵, Forbes D Porter³, and Carlo Pierpaoli^1
1Laboratory on Quantitative Medical Imaging, NIBIB, NIH, Bethesda, MD, United States, ²Henry Jackson Foundation for Advancement of Military Medicine, Bethesda, MD, United States, ³Division of Translational Medicine, NICHD,NIH, Bethesda, MD, United States, ⁴Office of Clinical Director, NIMH,NIH, Bethesda, MD, United States, ⁵Office of Clinical Director, NINDS,NIH, Bethesda, MD, United States

Keywords: Segmentation, Quantitative Imaging, Neurodegeneration, Neurodegeneration diseases, aging, T1-weighted segmentation, T1W, DTI, Diffusion Tensor Imaging, brain parenchyma volume, brain atrophy

In this work, we have compared if whole brain parenchyma fraction (BPF) measured from conventionally used T1-weighted (T1W) based brain segmentation method is comparable to signal fraction attributable to parenchymal water (Par-SF) measured from method using diffusion MRI, in assessing the overall disease state of participants with CLN3, a pediatric neurodegenerative disease.

View Presentation Video

Aaron Sinclair¹, Ross Callaghan², and Hui Zhang^1
1Centre for Medical Image Computing and Department of Computer Science, University College London, London, United Kingdom, ²AINOSTICS Ltd., Manchester, United Kingdom

Keywords: Segmentation, Neuroinflammation, White Matter Hyperintensity Segmentation

We propose a tract-specific white matter hyperintensity (WMH) segmentation evaluation as a new way to assess WMH segmentation techniques. Currently, WMH segmentation is assessed globally or between periventricular and deep white matter (WM) regions, which has no obvious functional relevance. WM tracts have known functional relevance and associations to neurological conditions. We demonstrate that this new approach allows for functionally-relevant assessment of techniques. Two WMH segmentation techniques are compared to highlight the performance differences across a key tract associated with processing speed. We also found significant performance differences across this tract, which would be missed by standard evaluation methods. 

View Presentation Video

Harshan Ravi¹, Samuel H. Hawkins^1,2, Olya Stringfield³, Malesa Pereira^1,4,5, Heiko Enderling^6,7, H-H Michael Yu^7,8, John A. Arrington^5,7, Solmaz Sahebjam^7,9,10, and Natarajan Raghunand^1,7
1Moffitt Cancer Center and Research Center, Tampa, FL, United States, ²Department of Computer Science, Bradley University, Peoria, IL, United States, ³Quantitative Imaging Core, Moffitt Cancer Center and Research Institute, Tampa,, FL, United States, ⁴Behavioral and Community Health Sciences,, LSU Health School of Public Health, New Orleans, LA, United States, ⁵Department of Radiology, Moffitt Cancer Center and Research Center, Tampa,, FL, United States, ⁶Department of Integrated Mathematical Oncology, Moffitt Cancer Center and Research Institute, Tampa,, FL, United States, ⁷Department of Oncologic Sciences, University of South Florida, Tampa, FL, United States, ⁸Department of Radiation Oncology, Moffitt Cancer Center and Research Center, Tampa,, FL, United States, ⁹National Cancer Institute, National Institutes of Health, Bethesda, MD, United States, ¹⁰Department of Neuro-Oncology, Moffitt Cancer Center and Research Center, Tampa,, FL, United States

Keywords: Segmentation, Machine Learning/Artificial Intelligence, Glioblastoma, immunotherapy, radiotherapy

Confounding appearance of radiographic changes in recurrent high grade glioma (HGG) patients treated with multimodality immunotherapy presents a challenge to the neuro-radiologist. A clinical need exists to improve upon conventional criteria for assessment of GBM on standard-of-care (SOC; T1w, T2w, FLAIR, and T1w-enhanced) MRI to help distinguish treatment-related effects from true disease progression. We have investigated the feasibility of intensity-based segmentation of tumor tissue types on multiparametric MRI (mpMRI) to inform response assessment in HGG patients treated with bevacizumab, hypofractionated stereotactic radiotherapy, and pembrolizumab.

View Presentation Video

Ruifeng Dong¹, Amritha Nayak^1,2, Leighton Chan³, and Carlo Pierpaoli^1
1Laboratory on Quantitative Medical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, MD, United States, ²Henry Jackson Foundation for Advancement of Military Medicine, Bethesda, MD, United States, ³Rehabilitation Medicine Department, National Institutes of Health, Bethesda, MD, United States

Keywords: Segmentation, Software Tools

We propose a framework for differentiating the contributions to the reproducibility in brain morphometry from the true inter-individual differences, experimental procedures, and data-processing methods. As an application, we build a linear mixed-effect model to evaluate and compare two segmentation software tools, Freesurfer and vol2Brain, in their reproducibility in measuring volumes of 32 regions of interest. For both software and for most structures under study, our approach successfully reveals the dominance of inter-subject variability over noise. Vol2Brain introduces less noise than Freesurfer for all subcortical nuclei while Freesurfer shows better performance for gray matter, cortex, cerebral white matter, and cerebellum cortex.

View Presentation Video

Artem Mikheev¹ and Henry Rusinek^1
1Radiology, NYU Grossman School of Medicine, New York, NY, United States

Keywords: Segmentation, Segmentation, Thresholding

In medical image processing thresholding of the image region is one of the most basic operations and preprocessing. While there are well-proven histogram-based partitioning methods including Otsu and Gaussian Mixture Model ) it is challenging to combine those methods with physiological constraints such as tissue volume ratio or average signal ratio to avoid anatomically invalid segmentation. We propose a new method GOTC when the distribution of the signal over the ROI is described as a bimodal Gaussian of two tissues. We describe the algorithm and validate it on  brain MRI segmentation  into the White Matter and Gray Matter.

View Presentation Video

Jiale Cheng^1,2, Fenqiang Zhao², Zhengwang Wu², Ya Wang², Xinrui Yuan², Weili Lin², Li Wang², Xin Zhang¹, and Gang Li^2
1School of Electronic and Information Engineering, South China University of Technology, Guangzhou, China, ²Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States

Keywords: Segmentation, Brain

Accurate parcellation of the extremely folded cerebellar cortex is of immense importance for both brain structural and functional studies. Manual parcellation is time-consuming and expertise dependent, which motivates us to propose a novel end-to-end deep learning-based method for cerebellar cortical surface parcellation. Leveraging the spherical topology of the cerebellar surface, we propose the Deformable Spherical Transformer, which combines the advantages of the Spherical Transformer to extract the long-range dependency and the deformable attention mechanism to adaptively focus on the critical regions. Its superior performance has been validated by comparing with advanced algorithms with an average Dice ratio of 86.40%.

View Presentation Video

Bing-Fong Lin¹, Dao-Chen Lin^2,3,4, and Chia-Feng Lu^1
1Department of Biomedical Imaging and Radiological Sciences, National Yang Ming Chiao Tung University, Taipei, Taiwan, ²Department of Radiology, Taipei Veterans General Hospital, Taipei, Taiwan, Taipei, Taiwan, ³Division of Endocrine and Metabolism, Department of Medicine, Taipei Veterans General Hospital, Taipei, Taiwan, ⁴School of Medicine, National Yang Ming Chiao Tung University, Taipei, Taiwan

Keywords: Radiomics, Quantitative Imaging, Pituitary adenoma

Pituitary adenoma (PA) accounts for approximately 15% in intracranial neoplasms. The classification of PA generally based on the hormone level of blood as the gold standard test, while the analysis of hormone condition using neuroimaging biomarkers was less explored. Accordingly, our study developed a model to automatically segment PA and further used the quantitative and non-invasive MRI technique as image biomarkers to classify the three types of hormone pattern, focusing on corticortroph, gonadotroph, and plurihormonal type. We aimed to provide a feasible classification model based MRI to benefit the clinical management of patients with PA.