Zhongzheng He^1,2, Nathalie Barrau¹, Claire Pellot Barakat¹, and Xavier Maître^1
1Université Paris-Saclay, CEA, CNRS, Inserm, BioMaps, Orsay, France, ²IADI U1254, INSERM, Université de Lorraine, Nancy, France

Current MR lung image segmentation has huge challenges compared to CT images, specifically in terms of low contrast, non-homogeneity. We developed a series of processing to accelerate the manual correction of reference standard lung masks by the auto-seeds region growing. Finally, we developed a 3D automatic MRI lung segmentation method using deep learning with a limited dataset (41 volumes for training and 4 volumes for validation). In the primary result, this lung segmentation archived a Dice score of 0.917±0.013. In the case of limited data, it provides us a new way for MR lung segmentation.

salem Alkhateeb¹, Tales Santini¹, Li jinghang¹, Robin Chu¹, Daniel Ibrahim¹, Anna M. Marsland¹, Stephen B. Manuck¹, Pete Gianaros¹, and tamer S. Ibrahim^2
1University of Pittsburgh, Pittsburgh, PA, United States, ²University of Pittsburgh, pittsburgh, PA, United States

As hippocampal volume has been extensively utilized as a diagnosing tool to confirm diagnosis of many neurological disorders, this study aims to employ the high resolution 7T TSE T2w data to segment high quality images with precision based on multi atlases and machine learning. Results have proved that this pipeline provides excellent outcomes and is validated to be used for more variables. 

Michael A Green^1,2, Peter Humberg³, Iain K Ball⁴, and Caroline D Rae^1,2
1Neuroscience Research Australia, Sydney, Australia, ²School of Medical Sciences, University of New South Wales, Sydney, Australia, ³Stats Central, Mark Wainwright Analytical Centre, University of New South Wales, Sydney, Australia, ⁴Philips Australia & New Zealand, Sydney, Australia

We employed the commonly used software package, Freesurfer to obtain volume, area and thickness measurements from structural MR imaging data acquired using the Compressed SENSE acceleration technique. We assessed measurement reliability via equivalence testing on a series of increasing acceleration factors to provide guidelines for researchers wanting to reduce scan acquisition time while acquiring high-quality data.

Fares Ouadahi¹, Anais Bernard¹, Lucile Brun¹, and Julien Rouyer^1
1Department of Research & Innovation, Olea Medical, La Ciotat, France

An accurate fibroglandular tissue (FGT) segmentation model was designed using of a deep learning strategy on T1w series without fat suppression. The proposed method combined a dedicated preprocessing and the training of a two-dimensional U-Net architecture on a multi-centric representative database to achieve an automatic FGT segmentation. The final test of the generated model exhibited overall good performances with a median Dice similarity coefficient of 0.951. More contrasted performances were obtained when correlating the gland density with the discrepancy between ground truth and prediction. Indeed, the lower the breast density, the greater the uncertainty in the segmentation.

Jose Velazco-Garcia¹, Nikolaos Tsekos¹, Andrew Webb², and Kirsten Koolstra^2
1Medical Robotics and Imaging Lab, University of Houston, Houston, TX, United States, ²Leiden University Medical Center, Leiden, Netherlands

We present the computational Framework for Interactive Immersion into Imaging Data (FI3D) for 3D visualization of MR data as they are collected and on-the-fly control of the MRI scanner. The FI3D was implemented to integrate an MRI scanner with the operator who used a customized holographic scene, generated by a HoloLens head mounted display (HMD), to (1) view images and image-generated outcomes (e.g., segmentation) and (2) control the MR scanner to update the acquisition protocol. In this study, the framework linked an MR scanner in Leiden (The Netherlands) and an operator in Houston (USA).

Rafia Ahsan¹, Iram Shahzadi^2,3, Ibtisam Aslam^1,4, and Hammad Omer^1
1Medical Image Processing Research Group (MIPRG), Dept. of Elect. & Comp. Engineering, COMSATS University Islamabad, Islamabad, Pakistan, ²OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden – Rossendorf, Dresden, Germany, ³German Cancer Research Center (DKFZ), Heidelberg, Germany, ⁴Service of Radiology, Geneva University Hospitals and Faculty of Medicine, University of Geneva, Geneva, Switzerland

Detection, classification and segmentation of brain tumor simultaneously is challenging due to the heterogeneous nature of the tumor. Limited work has been done in literature in this regard. The present study, therefore, aims to identify an object detection network that would be able to solve multi-class brain tumor classification and detection problem with high accuracy. Furthermore, the best performing detection network has been cascaded with 2D U-Net for pixel level segmentation. The proposed method not only classifies the tumor with high accuracy but also provides improved segmentation results compared to the standard U-Net. 

David Rudolf van Nederpelt¹, Houshang Amiri^1,2, Amanda Mariyampillai¹, Iman Brouwer¹, Samantha Noteboom³, Frederik Barkhof^1,4, Joost P.A. Kuijer¹, and Hugo Vrenken^1
1Radiology and nuclear medicine, Amsterdam UMC, location VUmc, Amsterdam, Netherlands, ²Neuroscience research center, Kerman University of Medical Sciences, Kerman, Iran (Islamic Republic of), ³Anatomy and Neurosciences, Amsterdam UMC, location VUmc, Amsterdam, Netherlands, ⁴Institutes of Neurology and Healthcare Engineering, UCL London, London, United Kingdom

Automated segmentation of brain MR images has paved the way for large cohort atrophy studies in multiple sclerosis (MS). A variety of automated software packages is available. Here we aimed to quantify brain volume differences measured by six freely available software packages on data from 21 MS patients all scanned thrice on three different MR scanners. While intra-class correlation coefficients were high, systematic differences between scanners were found for every software. This suggests that direct comparison of volumes acquired with different scanners is not possible and standardization is needed. 
 

Augustin C. Ogier¹, Stanislas Rapacchi², and Marc-Emmanuel Bellemare^1
1Aix Marseille Univ, Universite de Toulon, CNRS, LIS, Marseille, France, ²Aix Marseille Univ, CNRS, CRMBM, Marseille, France

Pelvic floor disorders are prevalent diseases and only 2D dynamic observations of straining exercises at excretion are available in clinics. The understanding of three-dimensional pelvic organs mechanical defects is not yet achievable. We proposed a complete methodology for the 3D representation of the bladder combined with 3D representation of the location of the highest strain areas on the organ surface. We assessed the potential of our method on eight control subjects for the reconstruction of bladder during intense straining from forced breathing exercises. The proposed methodology provided, for the first time, a proper 3D+t spatial tracking of bladder non-reversible deformations.

Rachel Chae^1,2, Ahsan Javed¹, Rajiv Ramasawmy¹, Hui Xue¹, Marcus Carlsson¹, Adrienne E Campbell-Washburn¹, and Felicia Seemann^1
1Cardiovascular Branch, National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD, United States, ²Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States

High-performance 0.55T systems are well suited for structural lung imaging due to reduced susceptibility and prolonged T2*. Lung segmentation is required to derive metrics of pulmonary function from lung MRI. Convolutional neural networks are effective for lung segmentation at higher field strengths, but they do not generalize to images acquired at 0.55T. This study develops a neural network for automated lung segmentation of T1 and proton density weighted ultrashort-TE MRI at 0.55T. Training data was generated using segmentations by active contours and manual corrections. The proposed network was fast (1.07s) and as accurate as existing semi-automated segmentation (dice coefficient 0.93).

Jingpeng Li^1,2, Jonas Vardal^3,4, Inge Groote¹, and Atle Bjornerud^1,2
1Computational Radiology and Artificial Intelligence, Oslo University Hospital, Oslo, Norway, ²Department of Physics, University of Oslo, Oslo, Norway, ³The Intervention Centre, Oslo University Hospital, Oslo, Norway, ⁴Faculty of Medicine, University of Oslo, Oslo, Norway

To accurately detect and localize postoperative tumor after surgery is of critical importance to postoperative patient management and survial rate. We propose a fully automated end-to-end coarse-to-fine segmentation approach for the segmentation of posoperative tumor.

Kilian Hett¹, Colin D. McKnight², Jarrod J. Eisma¹, Jason Elenberger¹, Ciaran M. Considine¹, Daniel O. Claassen¹, and Manus J. Donahue^1,2,3
1Neurology, Vanderbilt University Medical Center, Nashville, TN, United States, ²Radiology and Radiological Sciences, Vanderbilt University Medical Center, Nashville, TN, United States, ³Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center, Nashville, TN, United States

The overarching goal of this work is to develop and validate novel deep learning algorithms for segmenting the parasagittal dural (PSD) space, which has been hypothesized to harbor cerebral lymphatic channels, from standard non-contrast anatomical imaging. Specifically, contrasted based MRI studies have recently suggested that the PSD may be important for CSF clearance, however existing methods for evaluating this space require administration of exogenous contrast and time-consuming manual tracing, thereby limiting generalizability. We propose a new segmentation method using non-contrasted MRI, and we validate this method in a mixed cohort of older adults with and without neurodegeneration. 

Wolfgang Erb^1
1Dipartimento di Matematica "Tullio Levi-Civita", University of Padova, Padova, Italy

Graph wedgelets are a novel tool for the fast decomposition of images in geometrically meaningful, wedge-shaped subregions. In this work, we study the usage of graph wedgelets as a promising splitting method in a split-and-merge segmentation scheme for Magnetic Resonance Imaging. We combine adaptive wedgelet splits of MRI images with a simple and classical merging strategy for subregions and obtain in this way an efficient and robust segmentation of diagnostic-relevant subdomains in MRI data.

Jingyuan Lyu¹, Yongquan Ye¹, Zhongqi Zhang¹, Jian Xu¹, Xiao Chen², Terrence Chen², Shanhui Sun², and Eric Z. Chen^2
1UIH America, Inc., Houston, TX, United States, ²United Imaging Intelligence, Cambridge, MA, United States

This study demonstrates the effects on image reconstruction by integrating a separate parallel imaging layer with the deep learning module. With such additional layer, deep learning reconstruction is flexible and reliable for highly under-sampled data.