Unbiased Average Age-Appropriate Atlases for Pediatric Studies

TLDR: The methods used to create unbiased, age-appropriate MRI atlas templates for pediatric studies that represent the average anatomy for the age range of 4.5-18.5 years are presented, while maintaining a high level of anatomical detail and contrast.

About: This article is published in NeuroImage.The article was published on 2011-01-01 and is currently open access. It has received 1756 citations till now. The article focuses on the topics: Spatial normalization & Population.

Figures

Figure 4: RMS magnitude of the residual error vector field for each iteration (i.e., the bias in the average deformation for the current template), x axis shows the step-size in mm. On the top image, the symmetric (red circles) and asymmetric (black squares) NIHPD 04.5-18.5 models are compared. On the bottom, the different NIHPD age sub-ranges are plotted for the asymmetric atlas creation. One can see that at each iteration for each step size, the average RMS residual error magnitude is reduced, indicating that the optimization procedure is reaching a minima.

Figure 9: Comparison of probabilistic atlas of the brain tissue types (GM, WM, CSF) for the NIHPD04.5-18.5 atlas (leftmost 3 columns) and the ICBM18.5-43.0 atlas (rightmost 3 columns). The brightest voxels indicate high probability of that tissue class. Note that the skin and skull outlines are overlaid on each subimage to facilitate comparisons.

Figure 3: Average asymmetric template (04.5-18.5 y.o.) generated at each level of fitting. The grey scale images show the intensity average anatomy, while the rainbow colour scale shows the intensity standard deviation for selected iterations in the hierarchical fitting process. One can see that as fitting progresses, anatomical features become less blurred and the intensity variability is reduced. The intensity range of the average data sets runs from 0 to 100.

Figure 8: NIHPD 04.5-18.5 template (left) and ICBM 18.5-43.0 template (right), showing the T1w, T2w and PDw average templates for each group.

Figure 7: Close up of the T1w pediatric atlas data to show cortical detail.

Figure 6: NIHPD asymmetric templates (first six columns) + ICBM asymmetric template (rightmost column) for the T1w modality.

Frequently asked questions

Q1. What are the contributions mentioned in the paper "Unbiased average age-appropriate atlases for pediatric studies" ?

In this paper a method is presented to create unbiased, age-appropriate MRI atlas templates for pediatric studies that represent the average anatomy for the age range of 4. The main contribution of this paper involves the creation of anatomical T1-weighted, T2-weighted, and proton density-weighted templates for specific developmentally important age-ranges, derived from the largest epidemiological representative normal sample of the U. S. population, where each subject was carefully screened for medical and psychiatric factors and characterized using standardized rating scales and cognitive tests. 

Q2. What is the reason for using a single subject atlas?

the iterative nonlinear registration strategy used here results in templates with high anatomical detail throughout the brain, thus obviating the need to justify the use of a single subject atlas for registration. 

Q3. What is the method of creating asymmetric and symmetric templates?

Their method uses iterative refinement with successively finer scales of nonlinear registration to yield templates with a high degree of anatomical detail, even at the cortex. 

Q4. What is the way to build an atlas for a child?

Bhatia et al. (Bhatia, Aljabar et al. 2007) used an expectation-maximization framework to build an MRI atlas for 1- and 2-yearolds. 

Q5. What is the reason for the use of adult templates?

Since the developing brain is not simply a smaller version of an adult brain, the use of adult templates may introduce a bias in analysis. 

Q6. How many iterations did it take to achieve convergence?

Their experiments showed that performing four iterations for a given step size was sufficient to achieve convergence at the given level of detail, down to a 2 mm step size. 

Q7. What is the procedure used to estimate the deformation field?

this deformation field is upsampled and used as input to the next iteration of the procedure, where the blurring is reduced and the estimation of the deformation field is refined. 

Q8. What is the way to register a template?

To account for imperfections in the nonlinear registration procedure, multiple iterations are performed, each time using the new template as the registration target, until the difference between two successive templates is smaller than some threshold. 

Q9. What is the way to generate an average atlas volume?

Although this enables a user to generate an appropriate intensity average template volume for a particular study, anatomical details may be blurred in regions of high variability such as the cortex because only linear registration is used.