Electronic poster

Tuesday 13:30-15:30 Computer 125

13:30 5091. A Fully Automatic Cerebellum Segmentation Method Using an Active Contour Model with Shape Prior

Jinyoung Hwang¹, Junmo Kim¹, HyunWook Park<sup>1

1</sup>Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea, Republic of

The segmentation of cerebellum in human brain is not widely used since the boundary between cerebrum and cerebellum is indistinguishable due to the partial volume effect. Although some literatures proposed the methodology in cerebellum segmentation, they are not the purpose of the cerebellum segmentation. In this work, we present fully automatic cerebellum segmentation method using shape priors in brain MR images, which already skull-stripped volume. We evaluated the proposed method to images from BrainWeb, 1.5T, and 3T MR scanner. The proposed method shows fine segmentation results, and it could be used for cerebellum generation in human brain.

14:00 5092. A Fully Automated White Matter / Gray Matter Segmentation of Mice Spinal Cord on DTI Images

Michaël Sdika¹, Virginie Callot¹, Mathias Hebert¹, Guillaume Duhamel¹, Patrick J. Cozzone<sup>1

1</sup>CRMBM/CNRS UMR6612, Faculté de médecine, Université de la Méditérranée, Marseille, France, France

In this work, a fully automated method is proposed to segment mice SC white matter (WM) and gray matter (GM) tissues on Diffusion Tensor Imaging (DTI) images. The proposed method is based on three main step: first a small patch containing the SC is detected using a machine learning procedure, then a mask of the SC is computed within this patch and finally WM/GM segmentation is performed. Specific attention has been paid to choose an appropriate modality for each steps. The segmentation results has been evaluated by visual assessment by two experts on the images of 13 mice.

14:30 5093. Extracellular Fluid Volume Measurements with Complex Signal Analysis

John David Dickson¹, Guy Barnett Williams², Thomas Adrian Carpenter², Richard E. Ansorge<sup>1

1</sup>Department of Physics, Cambridge University, Cambridge, Cambridgeshire, United Kingdom; ²Wolfson Brain Imaging Centre, Cambridge University, Cambridge, Cambridgeshire, United Kingdom

It has been suggested that extracellular fluid spins undergo bulk dephasing from those in intracellular fluid. This study provides direct evidence for this phenomenon and exploits it to make quantitative measurements of both the intra/extracellular fluid volume fractions and precession frequencies. This is achieved by fitting the data from a Gradient Echo Sampling of a Free Induction Decay (GESFID) sequence with a complex signal model.

15:00 5094. Automatic Segmentation of Brain Tumors on Non-Contrast-Enhanced Magnetic Resonance Images Using Fuzzy Clustering

Yi-Min Liu¹, Chun-Chih Liao^1,2, Furen Xiao^1,3, Jau-Min Wong¹, I-Jen Chiang<sup>1,4

1</sup>Institute of Biomedical Engineering, National Taiwan University, Taipei, Taiwan; ²Department of Neurosurgery, Taipei Hospital, Department of Health, Taipei, Taiwan; ³Department of Neurosurgery, National Taiwan University Hospital; ⁴Institute of Biomedical Informatics, Taipei Medical University, Taipei, Taiwan

An automated brain tumor segmentation method is desirable for helping human experts to obtain tumor location and volume estimation. This study was aimed to automatically segment brain tumor with two non-contrast-enhanced MR images, T1 and T2 images, via an unsupervised fuzzy c-means clustering method combined with region merging and knowledge-based analysis. The overall quantitative results percent match and correspondence ratio of this system are 0.842 and 0.716, respectively.

Wednesday 13:30-15:30 Computer 125

13:30 5095. Modeling of T2* Decay in Water/fat Imaging: Comparison of One-Decay and Two-Decay Models

Diego Hernando¹, Zhi-Pei Liang¹, Peter Kellman<sup>2

1</sup>Electrical and Computer Engineering, University of Illinois, Urbana, IL, United States; ²National Institutes of Health, Bethesda, MD, United States

In quantitative water/fat imaging, modeling the T2* decay of the signal is necessary in order to avoid significant bias. A two-decay model with separate decay rates for water and fat has recently been proposed as an alternative to the one-decay model where water and fat share a common decay rate. Even though the two-decay model is more realistic, it suffers from increased noise sensitivity with respect to the one-decay model. In this work, we analyze quantitatively this tradeoff between bias and standard deviation using simulation, phantom and in vivo data. Our results show that a one-decay model is preferable for a clinically relevant range of fat fractions and SNRs.

14:00 5096. A Method for De-Scalping Human Brain MRI Data Using Lipid Ratio Map

Anup Singh¹, Mohammad Haris¹, Kejia Cai¹, Ari Borthakur¹, Hari Hariharan¹, Ravinder Reddy<sup>1

1</sup>CMROI, Department of Radiology, University of Pennsylvania, Philadelphia, PA, United States

De-scalping the brain data is a very critical step in MRI data post-processing and analyzing. Many applications related to brain MRI either require, or benefits from the ability to accurately segment brain from the non-brain tissue. Here we present a simple, fast and robust technique for brain de-scalping. Current de-scalping procedure utilizes lipid ratio map obtained from MRI images without and with lipid saturation pulse, which are normally acquired at clinical scanners. MRI whole brain data from different clinical scanners was successfully de-scalped using current procedure.

14:30 5097. Chemical Shift-Based Water/fat Separation: Comparison of Fitting Models

Diego Hernando¹, Zhi-Pei Liang¹, Peter Kellman<sup>2

1</sup>Electrical and Computer Engineering, University of Illinois, Urbana, IL, United States; ²National Institutes of Health, Bethesda, MD, United States

Quantitative water/fat separation in MRI requires careful modeling of the acquired signal. Multiple signal models have been proposed in recent years, but their relative performance has not yet been established. This abstract presents a comparative study of 12 signal models for water/fat separation. The models were selected according to three criteria: magnitude or complex fitting, single-peak or multi-peak fat spectrum, and modeling of T2* decay. Our results show that a complex-fitting, multi-peak fat, one-decay model is preferable over a wide range of clinically relevant fat fractions and SNRs.

15:00 5098. Comparison of Magnitude and Complex Data Fitting for Quantitative Water/fat Imaging

Diego Hernando¹, Zhi-Pei Liang¹, Peter Kellman<sup>2

1</sup>Electrical and Computer Engineering, University of Illinois, Urbana, IL, United States; ²National Institutes of Health, Bethesda, MD, United States

Magnitude fitting has been proposed as an alternative to complex data fitting for quantitative chemical shift-encoded water/fat imaging. Potential advantages of magnitude fitting include the removal of sensitivity to phase errors in the signal and suppression of B0 field inhomogeneity effects. However, the noise performance of magnitude fitting, relative to complex fitting, has not been established. In this abstract, we present a quantitative comparison of both methods, based on the bias and standard deviation of their estimates. Our results show that complex fitting is preferable to magnitude fitting for quantitative water/fat imaging, both in theory and in practice.

Thursday 13:30-15:30 Computer 125

13:30 5099. Correction of the Kinetic Parameters of Human Tissue Considering RF-Field Inhomogeneities

Robert Merwa¹, Gernot Reishofer², Thorsten Feiweier³, Karin Kapp⁴, Franz Ebner², Rudolf Stollberger<sup>5

1</sup>Medical Engineering, FH OÖ - Upper Austria University of Applied Sciences, Linz, Austria; ²Department of Radiology, Medical University of Graz, Graz, Austria; ³Healthcare, Siemens AG, Germany; ⁴Department of Radiation Therapy, Medical University of Graz, Graz, Austria; ⁵Institute of Medical Engineering, Graz University of Technology, Graz, Austria

Dynamic contrast-enhanced MRI was performed at 3 T in combination with a flip angle mapping sequence in order to correct the kinetic parameters of human tissue. Due to the local magnitude of these inhomogenities the values for the AIF and tissue concentrations are widespread which lead to an overestimation or underestimation of K^trans and V_e. The peak of the arterial input function decreases of about 60 % and the absolute difference of K^trans and V_e obtained with the AIF in two comparable arteries can be improved by a factor up to 33 if the dynamic data are corrected accordingly.

14:00 5100. Registration of Histology and MR Images Using Local Rigid Registration and Differential Evolution

Zhengyi Yang¹, Viktor Vegh¹, Deming Wang¹, David Charles Reutens<sup>1,2

1</sup>University of Queensland, Brisbane, Queensland, Australia; ²Royal Brisbane and Women's Hospital, Brisbane, Queensland, Australia

Registration of histological sections and the corresponding MR images is a critical step in MR constrained histology volume reconstruction. Histological sections with dislocated segments are problematic. This issue is addressed by employing a local rigid registration method. The dislocated segments are identified by morphological operations and connectivity analysis. These segments are treated as rigid bodies having independent degree-of-freedom of motion. The registration was to find the transformation matrix for each segment to maximize the similarity, which was normal mutual information, between the transformed histological section and target MR image. The method of differential evolution was used to find optimal registrations.

14:30 5101. A Method for Planning Interventions in the Brain with Straight Access Paths

Nikhil Navkar¹, Zhigang Deng¹, Jason Stafford², Jeffrey Weinberg³, Nikolaos V. Tsekos<sup>1

1</sup>Computer Science, University fo Houston, Houston, TX, United States; ²Imaging Physics, MD Anderson, United States; ³Neurosurgery, MD Anderson, United States

The aim of the work is to designed visualization techniques for preoperative planning of neurosurgical interventional procedures using straight tabular tool. The visualization techniques include calculation of access maps on the surface of patient head, which help the neurosurgeon in selecting the optimum point for insertion. The preliminary results show that it is possible to plan the trajectory of the interventional tool to hit a target in such a way that it minimizes the trauma to vital structures inside the brain.

15:00 5102. Input Function Detection in MR Brain Perfusion Using a Blood Circulatory Model Based Algorithm

Azimeh Noorizadeh¹, Hassan Bagher-Ebadian^2,3, Reza Faghihi¹, Jayant Narang⁴, Rajan Jain⁴, James Russel Ewing<sup>2,3

1</sup>Department of Nuclear Engineering, Shiraz University, Shiraz, Fars, Iran; ²Department of Neurology, Henry Ford Hospital, Detroit, MI, United States; ³Department of Physics, Oakland University, Rochester, MI, United States; ⁴Department of Radiology, Henry Ford Hospital, Detroit, MI, United States

MR Quantification of the hemodynamic maps such as Cerebral Blood Volume, Mean Transit Time, and Cerebral Blood Flow in perfusion studies is highly susceptible to selection of the correct Arterial Input Function (AIF) and a correct AIF selection could substantially reduce bias in hemodynamic parameters. This study uses a blood circulatory model to construct an automatic and model-based algorithm for AIF detection in MR perfusion studies. The algorithm is used to detect the AIF from MR perfusion of four patients with 19 slices. This study introduces a new and reliable (performance: 84%) algorithm for AIF detection in MR perfusion studies.

Novel Image Analysis Techniques

Hall B Monday 14:00-16:00 Computer 126

14:00 5103. DIR Imaging Using GRAPPA for Cortical Thickness Estimation

Narae Choi¹, Yoonho Nam¹, Dong-Hyun Kim<sup>1,2

1</sup>Electrical and Electronic Engineering, Yonsei University, Sinchon dong, Seoul, Korea, Republic of; ²Radiology, Yonsei University, Sinchon dong, Seoul, Korea, Republic of

Most gray matter volumetric studies use T1-weighted imaging such as MP-RAGE because it provides good contrast between white matter and cortex. However, due to susceptibility artifact coming from the air-tissue interface, a reliable and accurate measurement is difficult in the regions near the air-bone interface for T1-wieghted schemes. One way to alleviate this problem is to perform gray matter spin-echo imaging through DIR sequence. One drawback of the DIR sequence, however, is its long scan time. We applied one of the parallel imaging reconstruction schemes, GRAPPA scheme, to DIR imaging to evaluate measurement changes as a function of reduction factor.

14:30 5104. Enhancing Subcortical Image Segmentation Based on Age Dependent Intensity Normalization

Mustafa Ulas Ciftcioglu¹, Didem Gokcay<sup>1

1</sup>Medical Informatics Department, Informatics Institute, Middle East Technical University, Ankara, Turkey

Automatic algorithms for subcortical segmentation often suffer due to the complex anatomic structure of this area and intersubject variability. To overcome this problem, a method that incorporates age dependent tissue volume statistics with atlas based intensity normalization is proposed. Age dependent regression equations for volumetric ratios of the tissues are constructed and included in a segmentation performed by Maximum Likelihood (ML) approach. For intensity normalization, the intensity distribution from a single subject atlas is utilized, after registering the given image with the atlas image. Improvement on the proposed method is documented by comparison with a widely accepted segmentation tool.

15:00 5105. Automated Evaluation of Structural Characteristics and Extension of Cerebral Gliomas Using DTI-MR 3D Texture Analysis

Giorgio De Nunzio^1,2, Antonella Castellano^3,4, Gabriella Pastore, ^1,2, Marina Donativi², Giuseppe Scotti³, Lorenzo Bello⁵, Andrea Falini<sup>6

1</sup>INFN (National Institute of Nuclear Physics), Lecce, Italy; ²Department of Materials Science, University of Salento, Lecce, Italy; ³Neuroradiology Unit and CERMAC, Scientific Institute and University Vita-Salute San Raffaele, Milan, Italy; ⁴Institute of Radiological Sciences, University of Milano, Milan, Italy; ⁵Neurosurgery, Department of Neurological Sciences, University of Milano, Milan, Italy; ⁶Neuroradiology Unit and CERMAC, , Scientific Institute and University Vita-Salute San Raffaele, Milan, Italy

This work illustrates the development and validation of a semi-automated Computer-Assisted Detection technique (CAD) for the recognition of cerebral glioma in Diffusion Tensor MR images (DTI-MR). The described system adheres to the classic scheme of a CAD software tool, with a data preprocessing step followed by feature calculation and supervised tissue classification. The chosen discriminating features come from 3D statistical Texture Analysis. Segmentation results are also correlated with histopathological findings from specimens obtained from image-guided tumor biopsies.

15:30 5106. Texture Analysis of MRI of Juvenile Myoclonic Epilepsy Patients

Márcia Silva de Oliveira^1,2, Luiz Eduardo Betting, ²³, Fernando Cendes, ²³, Gabriela Castellano<sup>1,2

1</sup>Neurophysics Group, State University of Campinas (Unicamp), Campinas, SP, Brazil; ²CInAPCe (Cooperação Interinstitucional de Apoio a Pesquisas sobre o Cérebro), São Paulo State, Brazil; ³NeuroImage Laboratory, State University of Campinas (Unicamp), Campinas, SP, Brazil

Juvenile myoclonic epilepsy (JME) is the most frequent subsyndrome of the idiopathic generalized epilepsies. Experimental investigations support that the thalamus is a key structure in the mechanisms of JME. The objective of this study was to investigate the thalamus of patients with JME using texture analysis, a quantitative neuroimaging technique. Patients and controls were submitted to MRI investigation. The T1 volumetric sequence was used for thalamic segmentation and extraction of texture parameters. Texture analysis revealed differences between the thalamus of patients and controls. The present investigation supports the participation of the thalamus in the mechanisms of JME.

Tuesday 13:30-15:30 Computer 126

13:30 5107. Impact of Motion and Symmetry Correction on Perfusion Lesion Segmentation in Acute Ischemic Stroke: Quantitative Evaluation

Dattesh D. Shanbhag¹, Rakesh Mullick¹, Sumit K. Nath¹, Catherine Oppenheim², Marie Luby³, Katherine D. Ku³, Lawrence L. Latour³, Steven Warach³, - NINDS Natural History of Stroke Investigators<sup>3

1</sup>Imaging Technologies, GE Global Research, Bangalore, Karnataka, India; ²Department of Neuroradiology, Université Paris-Descartes, Paris, France; ³NINDS, NIH, Bethesda, MD, United States

In context of acute ischemic stroke, we demonstrate that motion correction on PWI data should be used only if motion is detected, rather than as a standard pre-processing recipe in analysis pipeline. A motion detection scheme based on image moments is shown to be effective in capturing motion during phase volumes. Since the motion correction is the “rate-limiting” step in PWI data analysis pipeline, moment based motion detection and selective motion correction can result in significant time saving for processing PWI data. Symmetry correction, a step commonly used for contralateral analysis, produces lower estimates of perfusion lesion volumes if applied retrospectively on quantitative maps rather than on bolus signal volumes.

14:00 5108. Correction Algorithm for Singular Value Decomposition Artifact in Quantitative Cerebral Perfusion Images Using SCALE-PWI

Jessy J. Mouannes¹, Wanyong Shin², Saurabh Shah³, Anindya Sen⁴, Sameer Maheshwari¹, Timothy Carroll<sup>1,4

1</sup>Biomedical Engineering, Northwestern University, Chicago, IL, United States; ²National Institute on Drug Abuse, National Institute of Health, Baltimore, MD, United States; ³Siemens Medical Solutions USA, Chicago, IL, United States; ⁴Radiology, Northwestern University, Chicago, IL, United States

Self-CALibrated Epi Perfusion Weighted Imaging (SCALE-PWI) MRI pulse sequence produces quantitative cerebral perfusion images in a single MRI scan, using dynamic susceptibility contrast (DSC) and T1 changes in normal white matter in relation to changes in the blood pool after contrast injection. A singular value decomposition algorithmic artifact results in alternating signal intensity modulation in the reconstructed quantitative perfusion maps of consecutive slices. A correction method to eliminate this artifact is presented in this study at 1.5T. The results show a significant effect of this correction on the resulting quantitative maps, which become more accurate and suitable for clinical diagnosis.

14:30 5109. Pharmacokinetic Modelling of DCE-MRI at Moderate Temporal Resolution: Dealing with Tumours Which Wash Out Extremely Fast

David John Manton¹, Martin D. Pickles¹, Martin Lowry¹, Lindsay W. Turnbull<sup>1

1</sup>YCR Centre for MR Investigations, Hull-York Medical School, Hull, East Yorkshire, United Kingdom

Dynamic, contrast-enhanced MRI was carried out with a temporal resolution of approximately 30 s. When a simple, two-compartment pharmacokinetic model without a significant signal contribution from blood plasma (SSCP) was utilised, the quality of fit was poor in breast tumours demonstrating extremely rapid contrast wash-out. More sophisticated models were then investigated with the best performance being achieved by a Tofts-Kermode-Kety model with an SSCP as modelled by a bi-exponential fit to the latter part of the Parker population arterial input function (i.e. ignoring early bolus peaks and assuming instantaneous mixing). This model also yielded parameters which are more physiologically realistic.

15:00 5110. A Self Automated Normalization Algorithm of CBV Maps for Glioma Grading

Ravi Teja Seethamraju¹, Hui You^2,3, Jinrong Qu^4,5, Eric A. Macklin⁶, Geoffrey S. Young

¹MR R and D, Siemens Medical Solutions, USA Inc., Charlestown, MA, United States; ²Radiology, Peking Union Medical College Hospital, Beijing, China; ³Neuro Radiology, Brigham and Women's Hospital, Boston, MA, United States; ⁴Radiology, Tiantan Hospital, Beijing, China; ⁵Radiology, Henan Tumor Hospital, Zhengzhou, China; ⁶Biostatistics Center, Massachusetts General Hospital, Boston, MA, United States

The hot spot method is the most widely used technique for analysis of DSC PWI maps. Here, ROIs are selected on the relCBV maps in the portion of tumor that appears to have the highest relCBV. This value is divided by the relCBV of ROI selected in the contralateral normal appearing white matter (NAWM), to yield the normalized CBV (nCBV). The measured nCBV is highly operator dependent, so we present a method for automating the determination of NAWM relCBV in order to reduce the operator dependence of the hot spot and other analytic methods.