Elsevier

NeuroImage

Volume 55, Issue 4, 15 April 2011, Pages 1577-1586
NeuroImage

DTI registration in atlas based fiber analysis of infantile Krabbe disease

https://doi.org/10.1016/j.neuroimage.2011.01.038Get rights and content

Abstract

In recent years, diffusion tensor imaging (DTI) has become the modality of choice to investigate white matter pathology in the developing brain. To study neonate Krabbe disease with DTI, we evaluate the performance of linear and non-linear DTI registration algorithms for atlas based fiber tract analysis. The DTI scans of 10 age-matched neonates with infantile Krabbe disease are mapped into an atlas for the analysis of major fiber tracts — the genu and splenium of the corpus callosum, the internal capsules tracts and the uncinate fasciculi. The neonate atlas is based on 377 healthy control subjects, generated using an unbiased diffeomorphic atlas building method. To evaluate the performance of one linear and seven nonlinear commonly used registration algorithms for DTI we propose the use of two novel evaluation metrics: a regional matching quality criterion incorporating the local tensor orientation similarity, and a fiber property profile based metric using normative correlation. Our experimental results indicate that the whole tensor based registration method within the DTI-ToolKit (DTI-TK) shows the best performance for our application.

Research Highlights

► Performance evaluation of linear and seven non-linear DTI registration algorithms. ► Affine, B-spline, FSL-B-spline, Demons, Demons-log, MedINRIA, DTI-TK. ► Regional matching quality criterion. ► Fiber property profile (fractional anisotrophy) based criterion. ► DTI-TK registration method shows the best performance for our application.

Introduction

Diffusion tensor imaging (DTI) is a magnetic resonance imaging (MRI) technique that enables the measurement of restricted diffusion of water molecules in tissue to produce neural tract images. This technique, although relatively new, has become increasingly important for studies of anatomical and functional connectivity of the brain regions. DTI is now extensively used to study the fiber architecture in the living human brain via DTI tractography. This technique has proven especially of value in clinical studies of white matter (WM) integrity in the developing brain for diseases (Basser et al., 1994), such as metachromatic leukodystrophy (MLD), cerebral palsy and Krabbe (Escolar et al., 2009).

Krabbe disease (also called globoid cell leukodystrophy) is a rare, autosomal recessive neurodegenerative disorder caused by a deficiency of an enzyme called galactocerebrosidase, which aids in the breakdown and removal of galactolipids found in myelin (Wenger et al., 2001). The buildup of these galactolipids affects the growth of the nerve's protective myelin sheath and causes degeneration of myelin in both the central and peripheral nervous system. If left untreated, children with Krabbe's disease generally experience severe neurologic deterioration and death. (Escolar et al., 2005). The major forms of the disease include an early onset (infantile) form and a late onset (juvenile or adult) form. The early onset form is a more severe type and is characterized by a rapidly progressing neurological deterioration resulting in a vegetative state and typically death within the first few years of life. The infantile form is seen in 1 for every 70 000–100 000 (Wenger et al., 2001). Children with infantile Krabbe disease are seen to have hyperintense lesions within the white matter on T2-weighted MR images. Particularly the abnormal hyperintense signal is observed in the posterior limb of the internal capsule, the white matter adjacent to the lateral ventricles, the centrum semiovale, the corona radiate and the white matter and dentate nuclei of the cerebellum. Hematopoietic stem cell transplantation has shown promise as therapy for Krabbe disease based on the fact that donor leukocytes can provide the deficient enzymes to cells in the peripheral and central nervous system. Treatment at asymptomatic, neonate stage has shown to stop disease progression (Escolar et al., 2005).

Water motion in myelinated white matter is anisotropic and DTI-MR signal is sensitized to the microscopic movement of water molecules. Myelinated white matter is seen to have higher anisotropy values on DTI derived anisotropy maps (Provenzale et al., 2005). Previous studies show that patients with infantile Krabbe disease have lower fractional anisotropy (FA) across the corpus callosum (Guo et al., 2001) and along the DTI fiber bundle of internal capsules (IC) when compared with healthy age-matched controls (Escolar et al., 2009). Escolar et al. (2009) also showed a correlation of pretreatment FA measurements with post treatment gross motor function.

Based on the above research findings (Escolar et al., 2009, Goodlett et al., 2009), we use an atlas based fiber tract analysis for analyzing DTI images of Krabbe subjects. For an accurate analysis it is crucial to establish a registration based voxel-wise correspondence between a normal control neonate DTI atlas (with prior information of fiber tract locations) and the Krabbe subjects' DTI images. The research presented in this paper highlights our work to determine the best state-of-the-art approach to individually register DTI images of Krabbe subjects into the atlas space.

The registration of diffusion tensor images is particularly challenging when compared to registering scalar images as DTI data is multi-dimensional and the tensor orientations after image transformations must remain consistent with the anatomy (Alexander et al., 2001, Gee and Alexander, 2005). The application of the registration methods on DTI of Krabbe neonates makes the problem even more challenging due to the following factors. Most of the registration methods discussed in this paper are based on the intensity of the fiber tracts in the fractional anisotropy maps and as discussed earlier, the Krabbe patients have lower FA values as compared to the control group. Lower FA values are due to the anisotropy caused by the demyelination of the nerves. Relatively rapid changes occur in white matter during the first year of life restricting the control provided age matched controls to a relatively narrow age range relative to the patient. Also regional variations between FA values in white matter sites could cause inaccurate comparisons and hence the analysis needs to be performed in specific well defined white matter structures (Provenzale et al., 2005). In addition to these points, the analysis in this paper is restricted to neonates and this adds to the complexity as DTI MRI of neonates have low signal-to-noise (SNR) and poorly developed white matter tracts.

DTI registration algorithms can be broadly categorized into two groups (Zhang et al., 2006). The first kind uses scalar images derived from DTI images and performs deformable registration with traditional image registration algorithms (Schnabel et al., 2001, Joshi et al., 2004, Andersson et al., 2007, Christensen et al., 1994, Christensen et al., 1997). Although this group discards the orientation component of the data, it is the most commonly used method because of the simplicity and the ease of implementation. The second group of DTI registration algorithms directly use higher order information of diffusion tensor images like the corresponding principal eigenvectors (Yap et al., 2009), or the full tensor information (Zhang et al., 2006, Yeo et al., 2008). Due to the complexity involved and the difficulty in realizing such algorithms, this group has not been explored extensively.

In this paper, we investigate eight DTI registration approaches from both groups, available either in-house or publicly:

  • 1)

    Affine registration by Studholme et al. (1999) using normalized mutual information as a registration metric within the Image Registration Toolkit1 (referred to as Affine in this paper).

  • 2)

    B-spline based registration by Schnabel et al. (2001) using normalized mutual information as a registration metric within the Image Registration Toolkit (referred to as B-spline in this paper).

  • 3)

    B-spline based registration by Andersson et al. (2007) using weighted sum of scaled sum-of-squared differences as a registration metric via the “fnirt” implementation within FSL2 (referred to as FSL in this paper).

  • 4)

    Diffeomorphic demons3 by Vercauteren et al. (2009) using sum-of-squared differences as a registration metric3 (referred to as Demons in this paper).

  • 5)

    Log demons3 by Vercauteren et al. (2008) using sum-of-squared differences as a registration metric (referred to as Demons-log in this paper).

  • 6)

    Fluid registration by Joshi et al. (2004) using sum-of-squared differences as a registration metric (referred to as Fluid in this paper).

  • 7)

    Tensor-based registration by Zhang et al. (2006) using explicit optimization of tensor reorientation in an analytic manner within DTI-ToolKit4 (referred to as DTI-TK in this paper).

  • 8)

    Diffeomorphic tensor-based registration by Yeo et al. (2008) using the exact finite strain gradient within MedINRIA5 (referred to as MedINRIA in this paper).

The first six methods are based on normalized FA maps whereas the last two are whole tensor based registration methods. An evaluation of algorithms from both the groups will give an insight into the higher performance of one group over the other, particularly considering the complexities in registering Krabbe neonates. To evaluate the performance of the registration algorithms, we introduce two novel evaluation metrics. The first metric is based on the matching quality of the local tensor orientation and atlas anisotropy in each voxel. The voxel-wise metric values are averaged over predefined regions within the atlas (such as the genu, splenium, internal capsules and uncinates). The second evaluation metric employs a normative fiber tract profile based criterion, which computes the correlation of the FA profile along the major tracts in the registered dataset and the atlas.

Section snippets

Subjects

The studies are approved by the institutional review board at the University of North Carolina. Due to the difficulty of Krabbe data acquisition, only ten neonates with Krabbe disease identified by family history or through the New York State screening program were used in this study. The ten Krabbe neonates are aged 8 to 67 days (mean: 22 days) at the time of scan. These subjects were referred to the Program for Neuro-developmental Function in Rare Disorders (NFRD) at the University of North

Visualization results

We present detailed results for two individual representative cases, K1 (with protocol 1) and K8 (with protocol 2), as well as the summary results across the whole Krabbe population of 10 subjects. As illustrated in Fig. 1, qualitative inspection of the registration results of K1 and K8 indicate that all deformable registration algorithms show satisfactory results. The linear Affine registration method fails to map the fiber tracts of the subjects into the atlas space, as clearly seen for the

Discussion

In this paper, we evaluated one linear and seven nonlinear registration methods for use in an atlas based DTI fiber analysis framework on 10 neonates with infantile Krabbe disease. No difference was observed between the two different protocols in terms of their registration accuracy. We used visual evaluation, tensor orientation based criteria, FA profiles based criteria, the correlation of the FA values and the number of failures to evaluate the performance of the registration methods. By

Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant No.60903127); the National Alliance for Medical Image Computing (NAMIC, NIH U54 EB005149); the National Institutes of Health (NIH) Roadmap for Medical Research (U54 EB005149–01); the Autism Centers of Excellence Network at UNC-CH (NIH R01 HD055741), Penn Image Computing & Science Laboratory (PICSL) and in part by the NIH Biomedical Imaging and Bioengineering (NIBIB) and the NIH Blueprint for Neuroscience (

References (39)

  • Andersson, J.L.R., Jenkinson, M., Smith, S., 2007. Non-linear registration, aka Spatial normalization. FMRIB technical...
  • V. Arsigny et al.

    Log-euclidean metrics for fast and simple calculus on diffusion tensors

    Magnetic Resonance in Medicine

    (2006)
  • Y. Cao et al.

    Large deformation diffeomorphic metric mapping of vector fields

    IEEE Transactions on Medical Imaging

    (2005)
  • G. Christensen et al.

    3d brain mapping using a deformable neuroanatomy

    Physics in Medicine and Biology

    (1994)
  • G. Christensen et al.

    Deformable templates using large deformation kinematics

    IEEE Transactions on Medical Imaging

    (1997)
  • M.L. Escolar et al.

    Transplantation of umbilical-cord blood in babies with infantile Krabbe's disease

    The New England Journal of Medicine

    (2005)
  • M.L. Escolar et al.

    Diffusion tensor imaging detects abnormalities in the corticospinal tracts of neonates with infantile Krabbe Disease

    American Journal of Neuroradiology

    (2009)
  • J.C. Gee et al.

    Diffusion-tensor image registration

  • J.H. Gilmore et al.

    White matter development in newborns assessed with diffusion tensor imaging

    International Congress on Schizophrenia Research

    (2003)
  • Cited by (0)

    View full text