Meeting Banner
Abstract #0948

Classical Segmentation Methods on Novel MR Imaging: A Study of Brain Tissue Segmentation of MP2RAGE Vs MPRAGE

Meritxell Bach Cuadra1, 2, Sebastien Gelin, 23, Alexis Roche1, 4, Oscar Esteban, 25, Tobias Kober4, Jos P. Marques6, Cristina Granziera7, Gunnar Krger4, 8

1Radiology Department, University Hospital Center (CHUV) and University of Lausanne (UNIL), Lausanne, Vaud, Switzerland; 2Signal Processing Laboratory (LTS5), Ecole Polytechnique Fdrale de Lausanne (EPFL), Lausanne, Vaud, Switzerland; 3Bern University, Bern, Switzerland; 4Advanced Clinical Imaging Technology, Siemens Healthcare Sector IM&WS S, Lausanne, Vaud, Switzerland; 5Biomedical Image Technologies (BIT), Universidad Politcnica de Madrid, Madrid, Spain; 6CIBM-Animal Imaging and Technology core, University of Lausanne, Lausanne, Vaud, Switzerland; 7Department of Clinical Neurosciences, University Hospital Center (CHUV) and University of Lausanne (UNIL), Lausanne, Vaud, Switzerland; 8CIBM-AIT, Ecole Polytechnique Fdrale de Lausanne (EPFL), Lausanne, Vaud, Switzerland

Existing brain tissue segmentation methods are optimized for conventional T1-weighted images such as MPRAGE. However, recent clinical research has highlighted the benefits of other image acquisition techniques such as MP2RAGE. In this work, we study the ability of three state-of-the-art algorithms to automatically segment WM, GM and CSF in MP2RAGE imaging. We quantify the differences between MPRAGE and MP2RAGE-based brain tissue probability maps through statistical voxel-based analysis. Results on a group of 19 healthy subjects show significant statistical differences between GM probability maps in the central nuclei and the cerebellum, for each of the three tested methods.

Abstract PDF Poster PDF

Keywords

account accurate acquisition advanced anatomy animal applicability applied arise array atlas atlases automatically axial background better bias biomedical bottom brain built cerebellum class classical classification clinical clusters columns community computer context contrast core corrections cortical covariate created default despite detailed detected determination develop diagnostic diseases distribution distributions enforcing excellent existing expected explained extension extract family field finally findings fluid foundations free frontal graph head healthy highlighted hospital house human implemented improved includes inspection intensity investigated limits lobe local made make maps marques masks maximization medical methodological middle mixing mixture model models modified neurological novel optimization overall overlap paired partial plus practice prior priori priors probability processing prognostic protocol proven providing pure quantify quantitative radiology rage random reader recent registration remove reported robust sector segment segmentation significantly smoothing spatial statistical structural subjects successfully suggests summary support system technologies technology tend tissue tissues towards unified unknown unlikely versus view visual volume white wise