Stuart Crozier

University of Queensland, Brisbane, Queensland, Australia

Are you Stuart Crozier?

Claim your profile

Publications (428)448.93 Total impact

  • Yang Yang, Feng Liu, Wenlong Xu, Stuart Crozier
    [Show abstract] [Hide abstract]
    ABSTRACT: Compressed sensing has been applied to magnetic resonance imaging (MRI) for the acceleration of data collection. However, existing compressed sensing (CS) techniques usually produce images with residual artifacts, particularly at high reduction factors. In this work, we propose a novel, two-stage reconstruction scheme, which takes advantage of the properties of k-space data and under-sampling patterns that are useful in CS. In this algorithm, the under-sampled k-space data is segmented into low-frequency and high-frequency domains. Then, in stage one, using dense measurements, the low-frequency region of k-space data is faithfully reconstructed. The fully reconstituted low-frequency k-space data from the first stage is then combined with the high-frequency k-space data to complete the second stage reconstruction of the whole of k-space. With this two-stage approach, each reconstruction inherently incorporates a lower data under-sampling rate and more homogeneous signal magnitudes than conventional approaches. Because the restricted isometric property is easier to satisfy, the reconstruction consequently produces lower residual errors at each step. Compared with a conventional CS reconstruction, for the cases of cardiac cine, sagittal brain MR and angiogram imaging, the proposed method achieves a more accurate reconstruction with an improvement of 2 ~ 4dB in peak signal-to-noise ratio respectively, using reduction factors of up to 6.
    IEEE transactions on bio-medical engineering 07/2014; · 2.15 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Magnetic resonance (MR) examinations of morphological characteristics of intervertebral discs (IVDs) have been used extensively for biomechanical studies and clinical investigations of the lumbar spine. Traditionally, the morphological measurements have been performed using time- and expertise-intensive manual segmentation techniques not well suited for analyses of large-scale studies.
    The spine journal: official journal of the North American Spine Society 06/2014; · 2.90 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: PurposeTo present and evaluate a fully automatic method for segmentation (i.e., detection and delineation) of suspicious tissue in breast MRI.Materials and Methods The method, based on mean-shift clustering and graph-cuts on a region adjacency graph, was developed and its parameters tuned using multimodal (T1, T2, DCE-MRI) clinical breast MRI data from 35 subjects (training data). It was then tested using two data sets. Test set 1 comprises data for 85 subjects (93 lesions) acquired using the same protocol and scanner system used to acquire the training data. Test set 2 comprises data for eight subjects (nine lesions) acquired using a similar protocol but a different vendor's scanner system. Each lesion was manually delineated in three-dimensions by an experienced breast radiographer to establish segmentation ground truth. The regions of interest identified by the method were compared with the ground truth and the detection and delineation accuracies quantitatively evaluated.ResultsOne hundred percent of the lesions were detected with a mean of 4.5 ± 1.2 false positives per subject. This false-positive rate is nearly 50% better than previously reported for a fully automatic breast lesion detection system. The median Dice coefficient for Test set 1 was 0.76 (interquartile range, 0.17), and 0.75 (interquartile range, 0.16) for Test set 2.Conclusion The results demonstrate the efficacy and accuracy of the proposed method as well as its potential for direct application across different MRI systems. It is (to the authors' knowledge) the first fully automatic method for breast lesion detection and delineation in breast MRI. J. Magn. Reson. Imaging 2014;39:795–804. © 2013 Wiley Periodicals, Inc.
    Journal of Magnetic Resonance Imaging 04/2014; 39(4). · 2.57 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this study, the effects of cardiac fibroblast proliferation on cardiac electric excitation conduction and mechanical contraction were investigated using a proposed integrated myocardial-fibroblastic electromechanical model. At the cellular level, models of the human ventricular myocyte and fibroblast were modified to incorporate a model of cardiac mechanical contraction and cooperativity mechanisms. Cellular electromechanical coupling was realized with a calcium buffer. At the tissue level, electrical excitation conduction was coupled to an elastic mechanics model in which the finite difference method (FDM) was used to solve electrical excitation equations, and the finite element method (FEM) was used to solve mechanics equations. The electromechanical properties of the proposed integrated model were investigated in one or two dimensions under normal and ischemic pathological conditions. Fibroblast proliferation slowed wave propagation, induced a conduction block, decreased strains in the fibroblast proliferous tissue, and increased dispersions in depolarization, repolarization, and action potential duration (APD). It also distorted the wave-front, leading to the initiation and maintenance of re-entry, and resulted in a sustained contraction in the proliferous areas. This study demonstrated the important role that fibroblast proliferation plays in modulating cardiac electromechanical behaviour and which should be considered in planning future heart-modeling studies.
    Journal of Zhejiang University SCIENCE B 03/2014; 15(3):225-42. · 1.11 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Tumours are known to be heterogeneous, yet typical treatment plans consider them as a single unit. This may influence treatment outcomes. However, treatment cannot be customised to intra-tumour variation without a method to establish outcomes at an intra-tumour scale. This work proposes a method to both assess and measure outcomes locally within tumours. Methods: Four patients were scanned at two post-surgery time points using contrast enhanced MRI and 3,4-dihydroxy-6-[18F]-fluoro-L-phenylalanine (18F-DOPA) PET. The shell of active tumour tissue is divided into a set of small subregions at both time points. Local outcome is measured from changes in subregion volume over time. The utility of the proposed approach is evaluated by measuring the correlation between PET uptake and documented growth. Correlation with overall survival time was also examined. Results: Local outcomes were heterogeneous and evidence of a positive correlation between local 18F-DOPA uptake and local progression was observed. Conclusions: Given that intra-tumour outcomes are heterogeneous the consistently positive correlation between FDOPA uptake and progression, local analysis of tumours could prove useful for treatment planning.
    02/2014; 489(1).
  • [Show abstract] [Hide abstract]
    ABSTRACT: Despite radical treatment with surgery, radiotherapy and chemotherapy, advanced gliomas recur within months. Geographic misses in radiotherapy planning may play a role in this seemingly ineluctable recurrence. Planning is typically performed on post-contrast MRIs, which are known to underreport tumour volume relative to FDOPA PET scans. FDOPA PET fused with contrast enhanced MRI has demonstrated greater sensitivity and specificity than MRI alone. One sign of potential misses would be differences between gross target volumes (GTVs) defined using MRI alone and when fused with PET. This work examined whether such a discrepancy may occur. Materials and Methods: For six patients, a 75 minute PET scan using 3,4-dihydroxy-6-18F-fluoro-L-phynel-alanine (18F-FDOPA) was taken within 2 days of gadolinium enhanced MRI scans. In addition to standard radiotherapy planning by an experienced radiotherapy oncologist, a second gross target volume (GTV) was defined by an experienced nuclear medicine specialist for fused PET and MRI, while blinded to the radiotherapy plans. The volumes from standard radiotherapy planning were compared to the PET defined GTV. Results: The comparison indicated radiotherapy planning would change in several cases if FDOPA PET data was available. PET-defined contours were external to 95% prescribed dose for several patients. However, due to the radiotherapy margins, the discrepancies were relatively small in size and all received a dose of 50 Gray or more. Conclusions: Given the limited size of the discrepancies it is uncertain that geographic misses played a major role in patient outcome. Even so, the existence of discrepancies indicates that FDOPA PET could assist in better defining margins when planning radiotherapy for advanced glioma, which could be important for highly conformal radiotherapy plans.
    02/2014; 489(1).
  • [Show abstract] [Hide abstract]
    ABSTRACT: Deformable models incorporating shape priors have proved to be a successful approach in segmenting anatomical regions and specific structures in medical images. This paper introduces weighted shape priors for deformable models in the context of 3D magnetic resonance (MR) image segmentation of the bony elements of the human hip joint. The fully automated approach allows the focusing of the shape model energy to a priori selected anatomical structures or regions of clinical interest by preferentially ordering the shape representation (or eigen-modes) within this type of model to the highly weighted areas. This focused shape model improves accuracy of the shape constraints in those regions compared to standard approaches. The proposed method achieved femoral head and acetabular bone segmentation mean absolute surface distance errors of 0.55±0.18mm and 0.75±0.20mm respectively in 35 3D unilateral MR datasets from 25 subjects acquired at 3T with different limited field of views for individual bony components of the hip joint.
    Medical image analysis 02/2014; 18(3):567-578. · 3.09 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In magnetic resonance imaging (MRI), rapidly changing gradient fields are applied to encode the magnetic resonance signal with spatial position; however eddy currents are induced in the surrounding conducting structures depending on the geometry of the conductor and the excitation waveform. These alternating fields change the spatial profile of the current density within the coil track with the applied frequencies of the input waveform and by their proximity to other conductors. In this paper, the impact of the conductor width and the excited frequency over the parameters that characterise the performance of split transverse and longitudinal gradient coils are studied. Thirty x-gradient coils were designed using a "free-surface" coil design method and the track width was varied from 1mm to 30mm with an increment value of 1mm; a frequency sweep analysis in the range of 100Hz to 10kHz was performed using the multi-layer integral method (MIM) and parameters such as power loss produced by the coil and generated in the cryostat, inductance, coil efficiency (field strength/operating current), magnetic field profile produced by the coil and the eddy currents were studied. An experimental validation of the theoretical model was performed on an example coil. Coils with filamentary conductor segments were also studied to compare the simulated parameters with those produced by coils with a finite track. There was found to be a significant difference between the parameters calculated using filamentary coils and those obtained when the coil is simulated using finite size tracks. A wider track width produces coil with superior efficiency and low resistance; however, due to the skin effect, the power loss increases faster in wider tracks than in those generated in coils with narrow tracks. It was demonstrated that rapidly changing current paths must be avoided in order to mitigate the power loss and the spatial asymmetry in the current density profile. The decision of using narrow or wider tracks in split coils should be carefully investigated using a temperature analysis which includes skin and proximity effects.
    Journal of Magnetic Resonance 02/2014; 242C:86-94. · 2.30 Impact Factor
  • Current Oncology 02/2014; 21(1):e172-8. · 1.63 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: MRI-LINAC is a new image-guided radiotherapy treatment system that combines magnetic resonance imaging (MRI) with a linear accelerator (LINAC) in a single unit. One drawback is that the pulsing of the split gradient coils of the system induces an electric field and currents in the patient which need to be predicted and evaluated for patient safety. In this novel numerical study the in situ electric fields and associated current densities were evaluated inside tissue-accurate male and female human voxel models when a number of different split-geometry gradient coils were operated. The body models were located in the MRI-LINAC system along the axial and radial directions in three different body positions. Each model had a region of interest (ROI) suitable for image-guided radiotherapy. The simulation results show that the amplitudes and distributions of the field and current density induced by different split x-gradient coils were similar with one another in the ROI of the body model, but varied outside of the region. The fields and current densities induced by a split classic coil with the surface unconnected showed the largest deviation from those given by the conventional non-split coils. Another finding indicated that the distributions of the peak current densities varied when the body position, orientation or gender changed, while the peak electric fields mainly occurred in the skin and fat tissues.
    Physics in Medicine and Biology 01/2014; 59(1):233-245. · 2.70 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: In magnetic resonance imaging (MRI), gradient coils are switched during fast current pulse sequences. These time-varying fields interact with the conducting structures of the scanner, producing deleterious effects such as image distortions and Joule heating. Using a multi-layer integral method, the spatiotemporal nature of the eddy currents induced by the gradient coils is investigated. The existence of the eigenmode is experimentally demonstrated by measuring the magnetic field and the time decay constant of a typical unshielded $z$-gradient coil and its interaction with a conductive cylinder. An effective current tailoring is achieved using the characteristic eigenvalues of the conducting domain-exciting coil system. The method can be used to understand and mitigate undesired effects of eddy currents in MRI.
    IEEE Transactions on Magnetics 01/2014; 50(2):945-948. · 1.42 Impact Factor
  • Jin Jin, Feng Liu, Stuart Crozier
    [Show abstract] [Hide abstract]
    ABSTRACT: It is generally a challenging task to reconstruct dynamic magnetic resonance (MR) images with high spatial and high temporal resolutions, especially with highly incomplete k-space sampling. In this work, a novel method that combines a non-rigid image registration technique with sparsity-constrained image reconstruction is introduced. Employing a multi-resolution free-form deformation technique with B-spline interpolations, the non-rigid image registration accurately models the complex deformations of the physiological dynamics, and provides artefact-suppressed high spatial-resolution predictions. Based on these prediction images, the sparsity-constrained data fidelity-enforced image reconstruction further improves the reconstruction accuracy. When compared with the k-t FOCUSS with motion estimation/motion compensation (MEMC) technique on volunteer scans, the proposed method consistently outperforms in both the spatial and the temporal accuracy with variously accelerated k-space sampling. High fidelity reconstructions for dynamic systolic phases with reduction factor of 10 and cardiac perfusion series with reduction factor of 3 are presented.
    Magnetic Resonance Imaging. 01/2014;
  • Source
    Adnan Trakic, Jin Jin, Ewald Weber, Stuart Crozier
    [Show abstract] [Hide abstract]
    ABSTRACT: Conventionally, magnetic resonance imaging (MRI) is performed by pulsing gradient coils, which invariably leads to strong acoustic noise, patient safety concerns due to induced currents, and costly power/space requirements. This modeling study investigates a new silent, gradient coil-free MR imaging method, in which a radiofrequency (RF) coil and its nonuniform field (B 1 (+)) are mechanically rotated about the patient. The advantage of the rotating B 1 (+) field is that, for the first time, it provides a large number of degrees of freedom to aid a successful B 1 (+) image encoding process. The mathematical modeling was performed using flip angle modulation as part of a finite-difference-based Bloch equation solver. Preliminary results suggest that representative MR images with intensity deviations of <5% from the original image can be obtained using rotating RF field approach. This method may open up new avenues towards anatomical and functional imaging in medicine.
    Computational and Mathematical Methods in Medicine 01/2014; 2014:461647. · 0.79 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Objective To validate an automatic scheme for the segmentation and quantitative analysis of the medial (MM) and lateral meniscus (LM) in magnetic resonance (MR) images of the knee. Method We analysed sagittal water-excited dual-echo steady-state MR images of the knee from a subset of the Osteoarthritis Initiative cohort. The MM and LM were automatically segmented in the MR images based on a deformable model approach. Quantitative parameters including volume, subluxation and tibial-coverage were automatically calculated for comparison (Wilcoxon tests) between knees with variable radiographic osteoarthritis (rOA), medial and lateral joint space narrowing (mJSN, lJSN) and pain. Automatic segmentations and estimated parameters were evaluated for accuracy using manual delineations of the menisci in 88 pathological knee MR examinations at baseline and 12 months time-points. Results The median (95% confidence-interval) Dice similarity index ( 2*|Auto ∩Manual|/(|Auto|+|Manual|)*100 2*|Auto∩ Manual|/(|Auto|+|Manual|)*100) between manual and automated segmentations for the MM and LM were 78.3%(75.0—78.7), 83.9%(82.1—83.9) at baseline and 75.3%(72.8—76.9), 83.0%(81.6—83.5) at 12 months. Pearson coefficients between automatic and manual segmentation parameters ranged from r=0.70 to r=0.92. MM in rOA/mJSN knees had significantly greater subluxation and smaller tibial-coverage than no-rOA/no-mJSN knees. LM in rOA knees had significantly greater volumes and tibial-coverage than no-rOA knees. Conclusion Our automated method successfully segmented the menisci in normal and osteoarthritic knee MR images and detected meaningful morphological differences with respect to rOA and JSN. Our approach will facilitate analyses of the menisci in prospective MR cohorts such as the OAI for investigations into pathophysiological changes occurring in early OA development.
    Osteoarthritis and Cartilage 01/2014; · 4.26 Impact Factor
  • Source
    Yeyang Yu, Jin Jin, Feng Liu, Stuart Crozier
    [Show abstract] [Hide abstract]
    ABSTRACT: Compressed Sensing (CS) has been applied in dynamic Magnetic Resonance Imaging (MRI) to accelerate the data acquisition without noticeably degrading the spatial-temporal resolution. A suitable sparsity basis is one of the key components to successful CS applications. Conventionally, a multidimensional dataset in dynamic MRI is treated as a series of two-dimensional matrices, and then various matrix/vector transforms are used to explore the image sparsity. Traditional methods typically sparsify the spatial and temporal information independently. In this work, we propose a novel concept of tensor sparsity for the application of CS in dynamic MRI, and present the Higher-order Singular Value Decomposition (HOSVD) as a practical example. Applications presented in the three- and four-dimensional MRI data demonstrate that HOSVD simultaneously exploited the correlations within spatial and temporal dimensions. Validations based on cardiac datasets indicate that the proposed method achieved comparable reconstruction accuracy with the low-rank matrix recovery methods and, outperformed the conventional sparse recovery methods.
    PLoS ONE 01/2014; 9(6):e98441. · 3.53 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: l1-SPIRiT is a fast magnetic resonance imaging (MRI) method which combines parallel imaging (PI) with compressed sensing (CS) by performing a joint l1-norm and l2-norm optimization procedure. The original l1-SPIRiT method uses two-dimensional (2D) Wavelet transform to exploit the intra-coil data redundancies and a joint sparsity model to exploit the inter-coil data redundancies. In this work, we propose to stack all the coil images into a three-dimensional (3D) matrix, and then a novel 3D Walsh transform-based sparsity basis is applied to simultaneously reduce the intra-coil and inter-coil data redundancies. Both the 2D Wavelet transform-based and the proposed 3D Walsh transform-based sparsity bases were investigated in the l1-SPIRiT method. The experimental results show that the proposed 3D Walsh transform-based l1-SPIRiT method outperformed the original l1-SPIRiT in terms of image quality and computational efficiency.
    Magnetic Resonance Imaging. 01/2014;
  • [Show abstract] [Hide abstract]
    ABSTRACT: Parallel imaging (PI) is widely used for imaging acceleration by means of coil spatial sensitivities associated with phased array coils (PACs). By employing a time-division multiplexing technique, a single-channel rotating radiofrequency coil (RRFC) provides an alternative method to reduce scan time. Strategically combining these two concepts could provide enhanced acceleration and efficiency. In this work, the imaging acceleration ability and homogeneous image reconstruction strategy of 4-element rotating radiofrequency coil array (RRFCA) was numerically investigated and experimental validated at 7T with a homogeneous phantom. Each coil of RRFCA was capable of acquiring a large number of sensitivity profiles, leading to a better acceleration performance illustrated by the improved geometry-maps that have lower maximum values and more uniform distributions compared to 4- and 8-element stationary arrays. A reconstruction algorithm, rotating SENSitivity Encoding (rotating SENSE), was proposed to provide image reconstruction. Additionally, by optimally choosing the angular sampling positions and transmit profiles under the rotating scheme, phantom images could be faithfully reconstructed. The results indicate that, the proposed technique is able to provide homogeneous reconstructions with overall higher and more uniform signal-to-noise ratio (SNR) distributions at high reduction factors. It is hoped that, by employing the high imaging acceleration and homogeneous imaging reconstruction ability of RRFCA, the proposed method will facilitate human imaging for ultra high field MRI.
    Journal of Magnetic Resonance 11/2013; · 2.30 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: MRI-LINAC is a new image-guided radiotherapy treatment system that combines magnetic resonance imaging (MRI) and a linear particle accelerator (LINAC) into a single unit. Moving (i.e. rotating or translating) the patient inside the strong magnetic field of the split MRI-LINAC magnet can potentially induce high levels of electric fields and corresponding current densities in the conducting tissues. The prediction and assessment of patient safety in terms of electromagnetic field exposure has received very little attention for a split cylindrical MRI magnet configuration, especially in the vicinity of the gap region. In this novel numerical study, based on the quasi-static finite-difference (QSFD) method, rotation-induced electric fields and current densities are calculated considering a split 1 T magnet and a tissue-accurate 2 mm-resolution human body model. The patient was modelled in both axial and radial orientations relative to the magnet gap in a number of treatment/imaging scenarios. It was found that rotating the patient in the radial orientation produced an order of magnitude larger field exposure in the central nervous system than when the patient was rotated in the axial orientation. Also, rotating the patient with periods lower than about Trot = 43.3 s may result in field exposures above the limits set out in the international safety guidelines. The novel results of this investigation can provide useful insights into the safe use of the MRI-LINAC technology and optimal orientations of the patient during the treatment.
    IEEE transactions on bio-medical engineering 11/2013; · 2.15 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Accurate bone segmentation in the hip joint region from magnetic resonance (MR) images can provide quantitative data for examining pathoanatomical conditions such as femoroacetabular impingement through to varying stages of osteoarthritis to monitor bone and associated cartilage morphometry. We evaluate two state-of-the-art methods (multi-atlas and active shape model (ASM) approaches) on bilateral MR images for automatic 3D bone segmentation in the hip region (proximal femur and innominate bone). Bilateral MR images of the hip joints were acquired at 3T from 30 volunteers. Image sequences included water-excitation dual echo stead state (FOV 38.6 × 24.1 cm, matrix 576 × 360, thickness 0.61 mm) in all subjects and multi-echo data image combination (FOV 37.6 × 23.5 cm, matrix 576 × 360, thickness 0.70 mm) for a subset of eight subjects. Following manual segmentation of femoral (head-neck, proximal-shaft) and innominate (ilium+ischium+pubis) bone, automated bone segmentation proceeded via two approaches: (1) multi-atlas segmentation incorporating non-rigid registration and (2) an advanced ASM-based scheme. Mean inter- and intra-rater reliability Dice's similarity coefficients (DSC) for manual segmentation of femoral and innominate bone were (0.970, 0.963) and (0.971, 0.965). Compared with manual data, mean DSC values for femoral and innominate bone volumes using automated multi-atlas and ASM-based methods were (0.950, 0.922) and (0.946, 0.917), respectively. Both approaches delivered accurate (high DSC values) segmentation results; notably, ASM data were generated in substantially less computational time (12 min versus 10 h). Both automated algorithms provided accurate 3D bone volumetric descriptions for MR-based measures in the hip region. The highly computational efficient ASM-based approach is more likely suitable for future clinical applications such as extracting bone-cartilage interfaces for potential cartilage segmentation.
    Physics in Medicine and Biology 09/2013; 58(20):7375-7390. · 2.70 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Tumors are typically analyzed as a single unit, despite their biologically heterogeneous nature. This limits correlations that can be drawn between regional variation and treatment outcome. Furthermore, despite the availability of high resolution 3D medical imaging techniques, local outcomes, (e.g. tumor growth), are not easily measured. This paper proposes a method that uses streamlines to divide a 3D region of interest (e.g. tumor) into units where local properties can be measured over the paths of growth. Parameters such as directional length and mean intensity can be measured locally at sequential time points and then compared. The method is evaluated on synthetic objects, simulated tumors, and medical images of brain tumors. The evaluations suggest that the method is suitable for mapping amorphous dynamic objects.
    IEEE transactions on bio-medical engineering 09/2013; · 2.15 Impact Factor

Publication Stats

2k Citations
448.93 Total Impact Points


  • 1989–2014
    • University of Queensland
      • School of Information Technology and Electrical Engineering
      Brisbane, Queensland, Australia
  • 2009–2013
    • Zhejiang University of Science and Technology
      Hang-hsien, Zhejiang Sheng, China
  • 2008–2013
    • The Australian e-Health Research Centre
      Brisbane, Queensland, Australia
  • 2012
    • Royal Children's Hospital Brisbane
      Brisbane, Queensland, Australia
  • 2004–2012
    • Zhejiang University
      • Department of Biomedical Engineering
      Hangzhou, Zhejiang Sheng, China
  • 2001–2012
    • University of Tasmania
      • School of Mathematics & Physics
      Newnham, Tasmania, Australia
  • 2011
    • Chongqing University
      • School of Software Engineering
      Chongqing, Chongqing Shi, China
    • The Florey Institute of Neuroscience and Mental Health
      Melbourne, Victoria, Australia
  • 2010–2011
    • Qingdao University
      Tsingtao, Shandong Sheng, China
  • 2007
    • Queensland University of Technology
      Brisbane, Queensland, Australia
  • 2005–2007
    • The Commonwealth Scientific and Industrial Research Organisation
      Canberra, Australian Capital Territory, Australia
    • National High Magnetic Field Laboratory
      Tallahassee, Florida, United States
  • 2003
    • Royal Brisbane Hospital
      • Department of Medicine
      Brisbane, Queensland, Australia