Image Analysis: Applications
Thursday 23 April 2009
Room 313BC 10:30-12:30


Joseph V. Hajnal and Min-Ying Lydia Su

10:30 649. Left Ventricular Wall Motion Abnormalities: Using Center Point Trajectory (CPT) Mapping to Quantify Focal Lesions
    Ting Song1, Jeffrey A. Stainsby2, Maureen N. Hood3, Vincent B. Ho4
Applied Science Laboratory, GE Healthcare, Bethesda, MD, USA; 2Applied Science Laboratory, GE Healthcare, Toronto, ON, Canada; 3Radiology, Uniformed Services University of the Health Sciences and National Naval Medical Center, Bethesda, MD, USA; 4Radiology, Uniformed Services University of the Health Sciences and National Naval Medical Center, Bethesda, Bethesda, MD
    Visual inspection of echocardiography or cardiac MR images remains the current clinical gold standard. We present a novel image-processing algorithm called Center Point Trajectory (CPT) mapping for quantification of focal left ventricular wall motion abnormalities. CPT analysis yields amplitude and direction values to define focal wall motion abnormalities. The technique uses standard cine SSFP images and does not require specialized MR pulse. The method can be potentially used quantification of systolic and diastolic wall motion changes and monitoring pharmacologic testing.
10:42 650. A Novel Method for Cine-CMR Automated Assessment of Left Ventricular Diastolic Dysfunction
    Keigo Kawaji1,2, Noel Christopher Codella2, Christopher W. Chu3, Richard B. Devereux3, Martin R. Prince2, Yi Wang1,2, Jonathan W. Weinsaft2,3
Biomedical Engineering, Cornell University, Ithaca, NY, USA; 2Radiology, Weill Cornell Medical College, New York, USA; 3Medicine/Division of Cardiology, Weill Cornell Medical College, New York, USA
    Cardiac magnetic resonance (CMR) is a well-established standard for assessment of LV systolic function, but assessment of diastolic function is limited and currently requires additional imaging, which can be time-consuming. We present a novel automated approach based upon an LV segmentation algorithm (LV-METRIC) that assesses diastolic function from SSFP cine-CMR by generating ventricular filling profiles. Our results demonstrate that automated segmentation using LV-METRIC can generate multiple diastolic parameters that are rapidly derivable, require no additional imaging beyond standard cine-CMR, and agree with echocardiographic measures of diastolic function.
10:54 651. Exploring Effective Connectivity During Unilateral Movement in Stroke Using Structural Equation Modeling
    Wan-wa Wong1, Kai-yu Tong1, Fei Meng1,2, Kwok-wing Tang3, Xiaorong Gao2, Shangkai Gao2, Suk-tak Chan1
The Hong Kong Polytechnic University, Hong Kong, China; 2Tsinghua University, Beijing, China; 3Queen Elizabeth Hospital, Hong Kong, China
    The interactions between brain areas after stroke is thought to be critical information for the exploration of abnormal patterns observed after stroke. Investigation of the interactions will enable us to explore the network in the damaged brain and explain how the adapted network contributes to the functional task. Structural equation modeling (SEM) is an approach to quantify interactions among brain regions based on connectivity models. An automated elaborative SEM analysis together with bootstrapping validation were applied to fMRI data of stroke subjects acquired during unilateral movement using unaffected wrist. Our findings showed that there are a total of 30 significant paths survived after validation test.
11:06 652. Automated Computational Analysis of Neonatal Hypoxic Injury and Implanted Therapeutic Neuronal Stem Cells
    Nirmalya Ghosh1, Stephen Ashwal1, Andre Obenaus2
Department of Pediatrics, Loma Linda University, Loma Linda, CA, USA; 2Department of Radiology and Radiation Medicine, Loma Linda University, Loma Linda, CA, USA
    Manual quantification methods have hindered objective and rapid MRI analysis of neonatal hypoxic brain injury and its complex interactions with implanted therapeutic neuronal stem cells (NSCs). To extract such information to understand stem cell therapeutic activity, we need an automated MRI analysis method. Hierarchical Region Splitting (HRS) detects and quantifies lesion and NSCs and their internal compositions over time and space to quantify individual characteristics and interactions more objectively.
11:18 653. Robust Myelin Water Quantification Using Spatially Regularized Nonnegative Least Square Algorithm
    Dosik Hwang1, Yiping P. Du2
Electrical and Electronic Engineering, Yonsei University, Seoul, Korea; 2Psychiatry, University of Colorado Denver, Denver, CO, USA
    A spatially regularized nonnegative least square algorithm was developed for robust myelin water quantification in the brain. The regularization of the conventional nonnegative least square (NNLS) algorithm has been expanded into the spatial domain in addition to the spectral domain. A substantial decrease in the myelin water fraction (MWF) variability was observed in both simulation results and the analysis of experimental data. In contrast to other filtering approaches that reduced the noise with a penalty of reduced spatial resolution, this new algorithm effectively preserved details of myelin distribution with substantial noise reduction. The visibility of small focal lesions was greatly improved.
11:30 654. Three-Dimensional Segmentation and Visualization of Cerebral Arteries and Veins from Simultaneously Acquired MRA and MRV Based on Graph-Cuts and Vessel Enhancement Filter
    Hackjoon Shim1,2, Sung-Hong Park1,3, Kyongtae Ty Bae1,3
Department of Radiology, University of Pittsburgh, Pittsburgh, PA, USA; 2School of Electrical Engineering, Seoul National University, Seoul, Korea; 3Department of Bioengineering, University of Pittsburgh, Pittsburgh, PA, USA
    In this study, we proposed a new 3D segmentation method to segment the arteries and veins from MRA and MRV in the brain, respectively, using a graph-cuts technique and a vessel enhancement filter and then displayed the arteries and veins together in 3D. The MRA and MRV were acquired simultaneously using newly-introduced compatible dual-echo arteriovenography (CODEA). Our proposed method is promising when there is a need to study the morphology of both arteries and veins together in the brain, for example, the characterization and quantification of arteriovenous malformation and brain tumor vascularity.
11:42 655.

Fuzzy Clustering-Based Segmentation of Manganese-Enhanced Neuronal Network Areas on MR Images

    Mark J.R.J. Bouts1, Jet P. van der Zijden1, Wim M. Otte1, Rick M. Dijkhuizen1
1Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands
    Regional contrast enhancement on manganese-enhanced MR images is commonly measured using ROI analysis. However such methods are subject to user bias. We evaluated four fuzzy clustering methods to unbiasedly depict prominent areas of manganese enhancement in a neuronal tracing study in rat brain. A conventional Fuzzy C-Means approach was tested against three spatial contiguity constrained approaches. Spatial contiguity was defined either by a Markov random field (MRF) or neighborhood homogeneity weighting. The third method combined the latter approaches. Our study demonstrated highest accuracy and best overlap with manual outlines for a combination of MRF with neighborhood homogeneity weighting.
11:54 656. Reduction of Motion Artefacts in Renal Perfusion DCE-MRI Data
    Gernot Reishofer1, Robert Merwa2, Manuela Aschauer3, Sabine Zitta4, Rudolf Stollberger2, Franz Ebner5
Department of Radiology / MR-Physics, Medical University of Graz, Graz, Austria; 2Institute of Medical Engineering, Graz University of Technology , Graz, Austria; 3Department of Radiology, Medical University of Graz, Graz, Austria; 4Department of Internal Medicine / Division of Nephrology, Medical University of Graz, Graz, Austria; 5Department of Radiology / Division of Neuroradiology, Medical University of Graz, Graz, Austria
    An exact analysis of renal perfusion parameter in different tissues requires a sufficient image registration of the dynamic scan. Most registration algorithms, based on the conservation of pixel intensity values fail due to fast intensity changes caused by contrast media uptake. We circumvent this pitfall by evaluating a second dynamic time series using a filter operation in the time domain. The images of this time series are used as templates for a non rigid registration algorithm. We demonstrate that our algorithm significantly reduces kidney movement and allows a more differentiated analysis of several kidney tissue types.
12:06 657. Selection of Diagnostic Features to Differentiate Between Malignant and Benign Lesions That Presented as Mass Lesions and Non-Mass Type Enhancement on Breast MRI
    Ke Nie1, Dustin Newell1, Jeon-Hor Chen1,2, Chieh-Chih Hsu2, Hon J. Yu1, Orhan Nalcioglu1, Min-Ying Lydia Su1
Tu & Yuen Center for Functional Onco-Imaging, University of California, Irvine, CA, USA; 2Department of Radiology, China Medical University, Taiwan
    Diagnostic features to differentiate between malignant and benign lesions presenting as mass and non-mass types were investigated using 116 lesions. The morphology of lesion (shape/margin and enhancement texture) and the enhancement kinetic parameters were obtained, and then a best feature set was selected by artificial neural network for making differential diagnosis. Morphology parameters can diagnose mass type benign and malignant lesions with a high accuracy (AUC=0.87), and adding Ktrans will further improve to 0.90. On the other hand, neither the morphology nor the kinetic parameters analyzed from outlined lesion ROI for non-mass lesions could differentiate between malignant and benign lesions.
12:18 658. Quantifying the Vascular Profile of a Tumor by 3D Euclidean Distance Maps
    Meiyappan Solaiyappan1, Deepak Dinakaran2, Yoshinori Kato1, Dmitri Artemov1
Department of Radiology, Johns Hopkins Medical Institutions, Baltimore, MD, USA; 2Department of Biological Sciences, University of Toronto, Toronto, Ontario, Canada
    A practical quantification method for the analysis of the vascular data will benefit MRA studies that focus on tumor vasculature for the purpose of understanding the effects of treatment. We show that the distribution of 3D Euclidean distance from each voxel within the vascularized regions of the tumor to their nearest vessel boundary can provide a useful quantification measure to monitor the changes in the tumor vasculature following treatment. Such an approach represents an objective way of monitoring the changes in the vasculature based on the changes in the distribution of the vascularized region with respect to the vessel branches.