Medical image computing and computer-assisted interventionRegistration-Based Approach for Reconstruction of High-Resolution In Utero Fetal MR Brain Images
Section snippets
Super Resolution Methods
The principle of super-resolution is to combine low-resolution images to produce an image that has a higher spatial resolution than the original images (9). This is a large research field encompassing many applications; however, the majority of the work has focused on using lower-resolution data acquired on a regular grid and often assuming simple translational motion between the lower-resolution sample grids, unlike our data, which are corrupted by full three-dimensional (3D) rigid motion on a
Application to MRI
Building a high-resolution 3D MR image of the fetal brain is challenging because of the original 2D slice thickness, the intensity distortion, and the unpredictable fetal motion that can occur in any direction and can be a much as several centimeters in distance between slice acquisitions. As far as we know, this problem, which has been discussed before in the literature (18, 19), is still an open issue. Super-resolution has previously been investigated in MRI using a specialized protocol to
Methods
Our overall aim is to reconstruct an image of the fetal brain that has increased and isotropic sampling resolution, increased overall spatial resolution (reduced point spread function), and consistent tissue intensity over the field of view. The resolution of the source data is typically 1 × 1 mm in plane with 3-mm-thick slices. Multiple sets consisting of between 20 and 40 slices each are acquired over a period of around 20 seconds for each set. The slices are commonly acquired in an
Results
From the simulated motion experiments on premature neonatal data, we evaluated the RMS registration error for four points at the corners of a box within the brain tissue of size 100 mm × 100 mm for each slice. These are presented in Figure 6. The simulated starting RMS error is up to 12 mm. Each point represents the result of one simulation and its coordinates are the starting RMS error and the final RMS error. For all cases, the final overall slice alignment error was significantly reduced by
Discussion
The ability to study the developing fetal brain in high-resolution promises to provide a vital source of clinical information that could contribute directly to a number of challenging clinical questions. It will permit the use of many quantitative morphometric analysis methods, originally developed to study the adult (32) and neonatal (33, 34) brain, to be applied to examine the process of in utero brain development. Critically, high-resolution imaging is the key to seeing the process of
References (36)
- et al.
Fetal brain MR imaging
MRI Clin North Am
(2001) Fetal central nervous system MR imaging
Neuroimag Clin North Am
(2006)- et al.
Magnetic resonance techniques in the evaluation of the perinatal brainRecent advances and future directions
Semin Neonatol
(2001) - et al.
Improving image resolution using subpixel motion
Pattern Recognition Letters
(1987) - et al.
Improving resolution by image registration
Graphical Models Image Processing
(1991) - et al.
Three-dimensional magnetic resonance imaging of fetal brains
Lancet
(2001) - et al.
MRI inter-slice reconstruction using super-resolution
Magn Reson Imaging
(2002) - et al.
A survey of medical image registration
Med Image Analysis
(1998) - et al.
An overlap invariant entropy measure of 3D medical image alignment
Pattern Recogn
(1999) - et al.
A comparison of freehand three-dimensional ultrasound reconstruction techniques
Med Image Analysis
(1999)
A deformation tensor morphometry study of semantic dementia with quantitative validation
NeuroImage
Fetal magnetic resonance imaging
J Maternal-Fetal Neonat Med
Fetal MR imagingA developing modality for the developing patient
AJR Am J Roentgenol
MR imaging of fetal cerebral anomalies
Pediatr Radiol
Fetal cerebral cortexNormal gestational landmarks identified using prenatal MR imaging
AJNR Am J Neuroradiol
Fetal MRITechniques and protocols
Pediatr Radiol
Topics in multiframe superresolution restoration
Computation and analysis of image motionA synopsis of current problems and methods
Int J Comput Vision
Cited by (0)
This work was supported by a Whitaker Foundation award (RG-01-0115), NIH grant R01-MH65392, and NIH Biomedical Research Partnership grant R01-EB0822.