View Presentation Video

Keywords: Segmentation, Segmentation

We compared 3D, 2.5D, and 2D approaches to brain MRI auto-segmentation and concluded that the 3D approach is more accurate, achieves better performance when training data is limited, and is faster to train and deploy. Our results hold across various deep-learning architectures, including capsule networks, UNets, and nnUNets. The only downside of 3D approach is that it requires 20 times more computational memory compared to 2.5D or 2D approaches. Because 3D capsule networks only need twice the computational memory that 2.5D or 2D UNets and nnUNets need, we suggest using 3D capsule networks in settings where computational memory is limited.

View Presentation Video

Brendan Williams¹, Dan Nguyen², and Manojkumar Saranathan^2
1University of Reading, Reading, United Kingdom, ²University of Massachusetts Chan Medical School, Worcester, MA, United States

Keywords: Segmentation, Segmentation

We present a systematic comparison of three state of the art thalamic segmentation methods for T1 MRI. Segmentation performance against Krauth-Morel atlas was quantified on 100 young healthy subjects and thalamic atrophy as a function of Alzheimer's disease status was characterized on 540 older subjects.

View Presentation Video

Daiki Tamada¹, Thekla H Oechtering^1,2, Eisuke Takai³, and Scott B Reeder^1,4,5,6,7
1Radiology, University of Wisconsin-Madison, Madison, WI, United States, ²Department of Radiology, Universität zu Lübeck, Lübeck, Germany, ³MIRAI Technology Institute, Shiseido Co., Ltd., Tokyo, Japan, ⁴Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ⁵Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States, ⁶Medicine, University of Wisconsin-Madison, Madison, WI, United States, ⁷Emergency, University of Wisconsin-Madison, Madison, WI, United States

Keywords: Segmentation, Segmentation, phase contrast MRA

We developed a segmentation algorithm for PC-MRA using a deep-learning approach, with the goal of achieving artifact-robust segmentation for PC-MRA. To simulate flow-related artifacts of MRA, Gaussian noise and phase error were added to the k-space domain of the datasets. LadderNet consists of two consecutive U-nets with skip connections, and has been adopted as a training network for vessel segmentation. Retrospective studies demonstrated superior accuracy and precision of the proposed method over a conventional level set segmentation method.

View Presentation Video

Subhranil Koley¹, Kavita Singh¹, Maria G. Garcia-Gomar^1,2, and Marta Bianciardi^1,3
1Brainstem Imaging Laboratory, Department of Radiology, Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States, ²Escuela Nacional de Estudios Superiores Unidad Juriquilla, Universidad Nacional Autónoma de México, Queretaro, Mexico, ³Division of Sleep Medicine, Harvard University, Boston, MA, United States

Keywords: Segmentation, Aging, Brainstem atlas

Brainstem nuclei are involved in several vital functions; impairment in their structure and function is manifested in several clinical conditions of elderly humans, including sleep/arousal/movement/vestibular/anxiety disorders, chronic pain and altered autonomic functions. Brainstem evaluation in health and disease is currently limited by the difficulty of localizing these regions in conventional MRI. We developed an in-vivo probabilistic atlas of 31 brainstem nuclei in elderly humans by the use of multi-contrast 7 Tesla MRI in 15 elderly subjects and of an existing brainstem nuclei atlas in younger adults. This atlas can be applied to conventional MRI and aid brainstem investigation in aging.

View Presentation Video

Anna Schroder¹, James Moggridge^2,3, Jiaming Wu^1,4, Hamza A. Salhab^2,3, Sjoerd Vos⁵, Melissa Bristow⁶, Fernando Pérez-García⁶, Javier Alvarez-Valle⁶, Tarek A. Yousry^2,3, John S. Thornton^2,3, Frederik Barkhof^1,3,4,7, Matthew Grech-Sollars^1,2, and Daniel C. Alexander^1
1Centre for Medical Image Computing, Department of Computer Science, University College London, London, United Kingdom, ²Lysholm Department of Neuroradiology, National Hospital for Neurology and Neurosurgery, University College London Hospitals NHS Foundation Trust, London, United Kingdom, ³Department of Brain Repair and Rehabilitation, UCL Institute of Neurology, University College London, London, United Kingdom, ⁴Department of Medical Physics & Biomedical Engineering, University College London, London, United Kingdom, ⁵Centre for Microscopy, Characterisation & Analysis, University of Western Australia, Perth, Australia, ⁶Health Futures, Microsoft Research Cambridge, Cambridge, United Kingdom, ⁷Department of Radiology and Nuclear Medicine, Amsterdam Neuroscience, Vrije Universiteit Amsterdam, Amsterdam UMC, Amsterdam, Netherlands

Keywords: Segmentation, Brain, Hippocampus

Accurate hippocampal segmentation tools are critical for monitoring neurodegenerative disease progression on MRI and assessing the impact of interventional treatment. Here we present the InnerEye hippocampal segmentation model and evaluate this new model against three standard segmentation tools in an Alzheimer’s disease dataset. We found InnerEye performed best for Dice score, precision and Hausdorff distance. InnerEye performs consistently well across the different cognitive diagnoses, while performance for other methods decreased with cognitive decline.

View Presentation Video

Sahar Ahmad¹, Xiaoyang Chen¹, Wenjiao Lyu¹, Jinjian Wu¹, Yifan Li¹, Yicheng Zou¹, Kim-Han Thung¹, Siyuan Liu¹, and Pew-Thian Yap^1
1Department of Radiology and Biomedical Research Imaging Center (BRIC), The University of North Carolina at Chapel Hill, Chapel Hill, NC, United States

Keywords: Segmentation, Brain, Surface reconstruction; Lifespan analysis

The availability of large-scale multi-site neuroimaging data present unprecedented opportunities for revealing the brain's macroscopic and microscopic organization across the lifespan. Existing neurodevelopmental studies lack consensus, owing to variable preprocessing methods that result in inconsistent brain features. Existing neuroimage processing tools typically cater to specific life periods, limiting their applicability to data covering the entire human lifespan. Here, we present robust automatic tools for accurate and consistent processing of neuroimaging data covering a century of life.

View Presentation Video

Soumick Chatterjee^1,2,3, Franziska Gaidzik⁴, Alessandro Sciarra^3,5, Hendrik Mattern³, Gabor Janiga⁴, Oliver Speck^3,6,7, Andreas Nürnberger^1,2,7, and Sahani Pathiraja^8
1Faculty of Computer Science, Otto von Guericke University Magdeburg, Magdeburg, Germany, ²Data and Knowledge Engineering Group, Otto von Guericke University Magdeburg, Magdeburg, Germany, ³Department of Biomedical Magnetic Resonance, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁴Laboratory of Fluid Dynamics and Technical Flows, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁵MedDigit, Department of Neurology, Medical Faculty, University Hospital Magdeburg, Magdeburg, Germany, ⁶German Center for Neurodegenerative Disease, Magdeburg, Germany, ⁷Center for Behavioral Brain Sciences, Magdeburg, Germany, ⁸School of Mathematics & Statistics, University of New South Wales, Sydney, Australia

Keywords: Segmentation, Brain

Disagreements among the experts while segmenting a certain region can be observed for complex segmentation tasks. Deep learning based solution Probabilistic UNet is one of the possible solutions that can learn from a given set of labels for each individual input image and then can produce multiple segmentations for each. But, this does not incorporate the knowledge about the segmentation distribution explicitly. This research extends the idea by incorporating the distribution of the plausible labels as a loss term. The proposed method could reduce the GED by 47% and 63% for multiple sclerosis and vessel segmentation tasks.

View Presentation Video

Jong Sung Park¹, Shreyas Fadnavis², and Eleftherios Garyfallidis^1
1Intelligent Systems Engineering / Neuroscience, Indiana University, Bloomington, Bloomington, IN, United States, ²Harvard University, Cambridge, MA, United States

Keywords: Segmentation, Brain

Brain Extraction is a complicated semantic segmentation task. While Deep Learning methods are popularly used, they are heavily biased towards the training dataset. To reduce this dependency, we present EVAC (Enhanced V-net like Architecture with Conditional Random Fields), a novel Deep Learning model for Brain Extraction. Using V-net as a skeleton, we propose three improvements: multi-scale inputs, modified CRF layer and regularizing Dice Loss. Results show that these changes not only increase accuracy but also the efficiency of the model as well. Compared to the state-of-the-art methods, our model achieves high and stable accuracy across datasets.

View Presentation Video

Valentina Visani¹, Valerio Natale², Annalisa Colombi³, Agnese Tamanti³, Alessandra Bertoldo¹, Corina Marjin³, Francesca Benedetta Pizzini², Massimiliano Calabrese³, and Marco Castellaro^1
1Department of Information Engineering, University of Padova, Padova, Italy, ²University Hospital of Verona, Verona, Italy, ³Department of Neurosciences, Biomedicine and Movement Sciences, University of Verona, Verona, Italy

Keywords: Segmentation, Machine Learning/Artificial Intelligence, Choroid Plexus

The Choroid Plexus (ChP) is a brain vascular tissue involved in regulatory processes. ChP Volume (ChPV) modifications are related to neurodegenerative disorders, consequently, it was suggested the use of ChPV as biomarker. This work proposes a method for the automatic segmentation of ChP based on Deep-Learning Neural-Networks (DNNs) hyperparameters optimization. Ninety-Six hyperparameters and architectures combinations were trained on T1-w MRI in MONAI, first selection was made on bias and variance and best DNNs were ensembled by major voting. Ensemble model outperforms single DNNs and freely available software (FreeSurfer, Gaussian Mixture Model), highlighting the ensembles DNNs exploitability to automatically estimate ChPV.  

View Presentation Video

JIAMING WU¹, Giuseppe Pontillo^1,2, Zoe Mendelsohn^1,2,3, Yipeng Hu¹, Frederik Barkhof^1,4, and Ferran Prados^1,2,5
1Centre for Medical Image Computing (CMIC), University College London, London, United Kingdom, ²Neuroradiological Academic Unit, UCL Queen Square Institute of Neurology, University College London, London, United Kingdom, ³Department of Radiology, Charité School of Medicine and University Hospital Berlin, Berlin, Germany, ⁴Radiology & Nuclear Medicine, VU University Medical Center, Amsterdam, Netherlands, ⁵e-Health Centre, Universitat Oberta de Catalunya, Barcelona, Spain

Keywords: Segmentation, Segmentation

Image segmentation and parcellation can provide quantitative assessment of the brain and can guide diagnosis and treatment decision-making. Geodesic Information Flow (GIF) is a freely available brain tissue segmentation and parcellation MRI-based tool using a classical label fusion approach. In this work, we introduce GIF_boost, a hybrid solution that takes advantages of deep learning to accelerate the bottleneck step of the template library registration. We compared GIF_boost with the original version of GIF and FreeSurfer (a state-of-the-art method). GIF_boost performed parcellation minimum 16 times faster. Parcellations had a similar Dice coefficient and Hausdorff distance and an improved volumetric quantification.   

View Presentation Video

Rui Li¹, Ziming Xu¹, Runyu Yang¹, Jiaqi Dou¹, Tianyi Yan², and Huijun Chen^1
1Center for Biomedical Imaging Research, School of Medicine, Tsinghua University, Beijing, China, ²School of Life Science, Beijing Institute of Technology, Beijing, China

Keywords: Segmentation, Brain

The precise segmentation of brain structures is of great importance in quantitatively analyzing brain medical resonance images. In recent years, more and more experts attempt to apply semi-supervised learning to medical image segmentation tasks, since it could make use of the rich unlabeled data. Based on this, we modified a segmentation model based on semi-supervised learning for automatically segmenting brain structures. We tested model on two hippocampus public datasets and results show that our model has considerable segmentation performance compared with that of model trained in a supervised manner, which illustrates the effectiveness and potential of our model.

View Presentation Video

Catarina Passarinho¹, Oscar Lally², Patrícia Figueiredo¹, and Rita G. Nunes^1
1Institute for Systems and Robotics - Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa, Lisbon, Portugal, ²King's College London, London, United Kingdom

Keywords: Segmentation, Brain

The same deep learning model was trained for automated segmentation of glioblastoma tumour regions using either four or two MRI modalities. The performance of the model trained with only two images was found to be comparable to that of the longer protocol, suggesting that the excluded images and the consequent longer training time did not contribute significantly to the accuracy of the model. These findings strongly imply that this training approach may be beneficial for clinical applications, as it would result in reduced costs due to shorter scanner times, lower computational requirements and increased patient throughput, without compromising segmentation accuracy.

Yousuf Babiker M. Osman^1,2, Cheng Li^1,3, Weijian Huang^1,2,4, Nazik Elsayed^1,2,5, Zhenzhen Xue^1,3, Hairong Zheng¹, and Shanshan Wang^1,3,4
1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, ²University of Chinese Academy of Sciences, Beijing, China, ³Guangdong Provincial Key Laboratory of Artificial Intelligence in Medical Image Analysis and Application, Guangzhou, China, ⁴Peng Cheng Laboratory, Shenzhen, China, ⁵Faculty of Mathematical and Computer Sciences, University of Gezira, Wad Madani, Sudan

Keywords: Segmentation, Prostate

Deep neural networks (DNNs) have achieved unprecedented performances in various medical image segmentation tasks. Nevertheless, DNN training requires a large amount of densely labeled data, which are labor-intensive and time-consuming to obtain. Here, we address the task of segmenting volumetric MR images using extremely sparse annotations, for which only the central slices are labeled manually. In our framework, two independent sets of pseudo labels are generated for unlabeled slices using self-supervised and semi-supervised learning methods. Boolean operation is adopted to achieve robust pseudo labels. Our approach can be very important in clinical applications to reduce manual effort on dataset construction.

Nicolas Basty¹, Ramprakash Srinivasan², Marjola Thanaj¹, Elena P Sorokin², Madeleine Cule², E Louise Thomas¹, Jimmy D Bell¹, and Brandon Whitcher^1
1University of Westminster, London, United Kingdom, ²Calico Life Sciences LLC, South San Francisco, CA, United States

Keywords: Data Analysis, Body, Deep learning, Dixon

Automated image processing and organ segmentation are critical to the quantitative analysis of population-scale imaging studies. We have implemented an end-to-end pipeline for neck-to-knee Dixon MRI data based on the UK Biobank abdominal protocol. Bias-field correction, blending across series boundaries, and fat-water swap correction are performed in the preprocessing steps. A hybrid attention-convolutional neural network model segments multiple abdominal organs, major bones, along with adipose and muscle tissue. The application of neural network models, to both swap detection and segmentation, produces a computationally-efficient pipeline that scales to accommodate tens of thousands of datasets.