Image Analysis: Novel Techniques
Click on to view the abstract pdf. Click on to view the recorded presentation.
Friday 11 May 2012
Room 212-213  10:30 - 12:30 Moderators: Brian Avants, William Wells

10:30 0746.   Introduction
Simon Warfield
10:42 0747.   
Fiber Orientation Mapping using Gradient Echo MRI: R2* mapping vs Frequency Difference Mapping
Samuel James Wharton1, and Richard Bowtell1
1Sir Peter Mansfield Magnetic Resonance Centre, School of Physics and Astronomy, University of Nottingham, Nottingham, United Kingdom

Recent work has shown that magnitude images acquired using gradient echo (GE) techniques at high field are sensitive to the fiber-orientation in white matter. Here, we show that frequency difference mapping (FDM) based on acquisition of multi-echo GE data allows the strong fiber-orientation contrast in phase images to be exploited in producing high resolution, 3D fiber-orientation maps. This approach was tested on post-mortem brain samples by comparing the resulting orientation maps to DTI data and fiber-orientation maps generated from simultaneously acquired R2* maps. The results suggest that FDM at high field may be used for high resolution fiber-orientation mapping.

10:54 0748.   
Multi-Channel Hybrid-Aided Registration of Multiple Spaces (mCHARM): Diffusion and Structural Image Co-Registration
Frederick William Damen1,2, Yi Sui1,3, and Xiaohong Joe Zhou1,4
1Center for MR Research, University of Illinois at Chicago, Chicago, IL, United States, 2Department of Biomedical Engineering, Georgia Institute of Technology, Atlanta, GA, United States, 3Department of Bioengineering, University of Illinois at Chicago, Chicago, IL, United States, 4Departments of Radiology, Neurosurgery, and Bioengineering, University of Illinois at Chicago, Chicago, IL, United States

Many diffusion MRI applications require co-registering a diffusion image based on EPI acquisitions to a high-resolution anatomical image; however, the contrast difference between diffusion and anatomical images poses a problem. We report a novel method, coined as multi-channel hybrid-aided registration of multiple spaces (mCHARM), to co-reregister diffusion images to T1-weighted images using a synthesized diffusion multi-channel hybrid (DMCH) image that mimics the contrast of a T1-weighted image. Our results from ten human volunteers have demonstrated that mCHARM improves the registration accuracy and produces reliable FA skeleton mask for tract-based spatial statistics analysis.

11:06 0749.   Registration of DCE-MRI using Robust Data Decomposition
Valentin Hamy1, Andrew Melbourne2, Benjamin Trémoulhéac2, Shonit Punwani1, and David Atkinson1
1Centre for Medical Imaging, UCL, London, United Kingdom, 2Centre for Medical Image Computing, UCL, London, United Kingdom

Dynamic Contrast Enhanced (DCE) MRI is a technique used in oncology to extract quantitative information about tumours. However, when it comes to image organs in the thorax or the abdomen, several types of motion (e.g. patient’s breathing, heartbeat and bowel peristalsis) can lead to errors in the estimation of that quantitative information. The study we present proposes a new technique aimed at compensating deformations/displacements due to motion in DCE images acquired using a repeat breath-hold protocol or a free breathing protocol.

Oleh Dzyubachyk1, Boudewijn Lelieveldt1,2, Jorik Blaas1, Monique Reijnierse1, Andrew Webb1, and Rob van der Geest1
1Department of Radiology, Leiden University Medical Center, Leiden, Netherlands, 2Mediamatics Department, Delft University of Technology, Delft, Netherlands

Initial feasibility studies have shown the promise of 7T MRI for studying spine disorders. However, diagnosis based on such data is compromised by the presence of high intensity inhomogeneities and the need to analyze several separately acquired image stacks. In practice, both intensity inhomogeneity correction and volume stitching must be performed manually. Here we present our novel fully automated algorithm for reconstructing a whole spine volume from a set of image stacks acquired on a 7T MR scanner.

11:30 0751.   Validation of Fully Automatic Adipose Tissue Segmentation and Volume Quantification
Bryan T Addeman1, Melanie Beaton2, Robert A Hegele3, Abraam S Soliman3,4, Curtis N Wiens5, and Charlies A McKenzie1,5
1Department of Medical Biophysics, University of Western Ontario, London, ON, Canada, 2Department of Medicine, University of Western Ontario, London, ON, Canada,3The Robarts Research Institute, London, ON, Canada, 4Biomedical Engineering, University of Western Ontario, London, ON, Canada, 5Department of Physics, University of Western Ontario, London, ON, Canada

The distribution of adipose tissue is associated with the long-term development of type 2 diabetes and cardiovascular disease. Most adipose tissue volume quantification techniques require manual input, are susceptible to human error, and are time consuming. We propose a novel automated process for the quantification and segmentation of Total Adipose Tissue, Subcutaneous Adipose Tissue, and Intra-Abdominal Adipose Tissue using quantitative fat fraction maps. Segmentation is robust and requires no prior knowledge or complex machine learning. Results show that automated fat volume measurements are similar to manual segmentation techniques and can be calculated very rapidly.

11:42 0752.   Computer-Aided Detection of Metastatic Brain Tumors: Comparison between MP-RAGE and Black-Blood Imaging
Seungwook Yang1, Eung Yeop Kim2, Min-Oh Kim1, Yoonho Nam1, Jaeseok Park3, and Dong-Hyun Kim1
1Electrical and Electronic Engineering, Yonsei University, Sinchon-dong, Seoul, Korea, Republic of, 2Radiology and Research Institute of Radiological Science, Yonsei University College of Medicine, Seoul, Korea, Republic of, 3Department of Brain and Cognitive Engineering, Korea University, Seoul, Korea, Republic of

The purpose of this study was to develop a computer-aided detection (CAD) system for brain metastases detection from magnetic resonance (MR) black blood images and to assess the applicability of MR black-blood imaging to CAD. Twenty-six patients with brain metastases of various sizes were imaged with a contrast-enhanced, three dimensional, whole brain MR black blood pulse sequence. The CAD system uses 3D template matching which measures normalized cross-correlation coefficient (NCCC) to generate possible metastases candidates from the patient data. Various image features were extracted from each candidate, then principal component analysis (PCA) was performed to determine dominant features. Artificial neural network (ANN) training and testing scheme were incorporated for classification.

11:54 0753.   
Atlas-free Brain Tissue Segmentation Using a Single T1-weighted MRI Acquisition
Tobias Kober1, Alexis Roche1,2, Oscar Esteban3,4, Subrahmanyam Gorthi4, Delphine Ribes5, Meritxell Bach-Cuadra2,4, Reto Meuli2, and Gunnar Krueger1,2
1Advanced Clinical Imaging Technology, Siemens Healthcare Sector IM&WS S, Lausanne, Switzerland, 2Centre Hospitalier Universitaire Vaudois and University of Lausanne, Switzerland, 3Biomedical Image Technology (BIT), Universidad Politécnica de Madrid, Madrid, Spain, 4Signal Processing Laboratory (LTS5), École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 5ARTORG Center for Computer Aided Surgery, University of Bern, Switzerland

Typical brain tissue segmentation algorithms use prior knowledge in the form of pre-segmented templates, i.e. atlases. This can lead to erroneous segmentation results if the actual image differs too much from the template, e.g. due to a too big age difference or a disease. In this work, the homogenous T1-contrast of the MP2RAGE sequence is combined with a Dixon water/fat imaging approach. It is shown that a single MP2RAGE-Dixon acquisition provides enough information for reproducible atlas-free brain tissue segmentation.

12:06 0754.   Automated quality control in MR-based brain morphometry
Bénédicte Maréchal1, Tobias Kober1, Tom Hilbert2, Delphine Ribes3, Nicolas Chevrey4, Alexis Roche1, Jean-Philippe Thiran5, Reto Meuli4, and Gunnar Krueger1
1Advanced Clinical Imaging Technology, Siemens Healthcare Sector - CIBM, Renens, Switzerland, 2Universität Heidelberg, Germany, 3ARTORG Center for Computer Aided Surgery, Univ. of Bern, Switzerland, 4Centre Hospitalier Universitaire Vaudois and Univ. of Lausanne, 5Signal Processing Laboratory (LTS5) Ecole Polytechnique Fédérale de Lausanne

Normal aging and a wide range of neurologic, inflammatory or psychiatric diseases lead to changes in the brain tissue over time. In the interest of diagnosis, prognosis and treatment monitoring, it is highly desirable to have robust tools that reliably measure brain morphometry. We explore the ability of an automated MR image quality assessment technique to predict the accuracy of subsequent algorithms for brain quantitative analysis. The approach proofs to be a very promising candidate to objectively assess quality prior to any post-processing in order to attribute tissue changes to a potential pathology rather than to image degradation.

12:18 0755.   Are two samples of parametric images statistically different? Novel significane tests on samples of density estimates, with application to ventilation-to-perfusion mapping of the lung in COPD using oxygen-enhanced MRI
Chris J Rose1,2, Penny L Hubbard1,2, Tim F Cootes1,2, Chris J Taylor1,2, Josephine H Naish1,2, Geoff J Parker1,2, Simon S Young3, and John C Waterton1,4
1University of Manchester Biomedical Imaging Institute, Manchester, Greater Manchester, United Kingdom, 2University of Manchester Academic Health Science Centre, Manchester, Greater Manchester, United Kingdom, 3AstraZeneca R&D, Charnwood, Loughborough, Leicestershire, United Kingdom, 4AstraZeneca, Alderley Park, Macclesfield, Cheshire, United Kingdom

MRI is used in natural history and intervention studies to spatially map parameters throughout organs or tumors. However, the problem of choosing the best method for drawing inferences about the populations being studied is unsolved. Conventionally, a significance test is applied using averaged parameter values, but this method neglects heterogeneity. We developed significance tests for application to a sample of density estimates (smooth histograms). Using ventilation-to-perfusion (V/Q) ratio data from an oxygen-enhanced MRI study of COPD patients and age-matched controls, we can identify biologically meaningful significant differences in V/Q where the conventional method fails.