Presentation is loading. Please wait.

Presentation is loading. Please wait.

Sparse Shape Representation using the Laplace-Beltrami Eigenfunctions and Its Application to Modeling Subcortical Structures Xuejiao Chen.

Similar presentations


Presentation on theme: "Sparse Shape Representation using the Laplace-Beltrami Eigenfunctions and Its Application to Modeling Subcortical Structures Xuejiao Chen."— Presentation transcript:

1 Sparse Shape Representation using the Laplace-Beltrami Eigenfunctions and Its Application to Modeling Subcortical Structures Xuejiao Chen

2 Page  2 Outline Introduction Method Results Conclusion

3 Page  3 Introduction Although the atrophy of brain tissues associated with the increase of age is reported in several hundreds subjects, the finding on the atrophy of amygdalar and hippocampus are somewhat inconsistent. Many cross sectional and longitudinal studies reported significant reduction in regional volume of amygdala and hippocampus due to aging, others failed to confirm such relationship. Gender may be another factor that affects these structures.

4 Page  4 Introduction In previous volumetric studies, the total volume were estimated by tracing the ROI manually and counting the number of voxels. Limitation: it cannot determine if the volume difference is diffuse over the whole ROI of localized within specific regions of the ROI.

5 Page  5 Outline Introduction Method Results Conclusion

6 Page  6 Method Pipeline: Obtain a mean volume of a subcortical structure by averaging the spatially normalized binary masks, and extract a template surface from the averaged binary volume. Interpolate the 3D displacement vector field onto the vertices of the surface meshes. Estimate a spase representation of Fourier coefficients with l1- norm penalty for the displacement length along the template surface to reduce noise Apply a general linear model testing the effect of age and gender on the displacement.

7 Page  7 Images and preprocessing T1-weighted MRI 124 contiguous 1.2mm axial slices 52 middle-age and elderly adults ranging between 37 to 74 years.(55.52 ±10.40 years) 16 men and 36 women Trained raters manually segmented the amygdala and hippocampus structures. Brain tissues in MRI were automatically segmented using Brain Extraction Tool(BET).

8 Page  8 Perform a nonlinear image registration using the diffeomorphic shape and intensity averaging technique with the cross-correlation as the similarity metric through Advanced Normalization Tools(ANTS). A study-specific template was constructed from a random subsample of 10 subjects. Align the amygdala and hippocampus binary masks to the template space and produce the subcortical structure template. Extract the isosurface of the subcortical structure template using the marching cube algorithm.

9 Page  9 Images and preprocessing

10 Page  10 Images and preprocessing The displacement vector field is defined on each voxel. Linearly interpolated the vector field on mesh vertices from the voxels. The length of the displacement vector at each vertex is computed and used as a feature to measure the local shape variation with respect to the template space.

11 Page  11 Sparse representation using an l1-penalty In previous LB-eigenfunction and similar SPHARM expansion approaches, only the first few terms are used in the expansion and higher frequency terms are simply thrown away to reduce the high frequency noise. Some lower frequency terms may not necessarily contribute significantly in reconstructing the surfaces.

12 Page  12 Sparse representation using an l1-penalty Consider a real-valued functional measurement Y(p) on a manifold M. We assume the following additive model: Y(p) = θ(p) + ε(p) Solving Δψ j =λ j ψ j on M, we find the eigenvalues λ j and eigenfunctions ψ j. Thus we can parametrically estimate the unknown mean signal θ(p) as the Fourier expansion as Least squares estimation(LSE):

13 Page  13 Sparse representation using an l1-penalty The estimation may include low degree coefficients that do not contribute significantly. Instead of using LSE, the additional l1-norm penalty to sparsely filter out insignificant low degree coefficients by minimizing In practice, λ=1 to control the amount of sparsity. This results in 72.87 non-zero coefficients out of 1310 in average for amygdale and 133.09 non-zero coefficients out of 2449 in average for hippocampi.

14 Page  14 Sparse representation using an l1-penalty

15 Page  15 Outline Introduction Method Results Conclusion

16 Page  16 Results Traditional volumetric analysis Subcortical structure shape analysis Effect of behavioral measure on anatomy

17 Page  17 Traditional volumetric analysis The volume of a structure is simply computed by counting the number of voxels within the binary mask. The brain volume except cerebellum was estimated and covariated in GLM. The volume of amygdala and hippocampus was modeled as

18 Page  18 Traditional volumetric analysis

19 Page  19 Subcortical structure shape analysis The length of displacement vector field along the template surface was estimated using the sparse framework.

20 Page  20 Subcortical structure shape analysis

21 Page  21 Subcortical structure shape analysis

22 Page  22 Effect of behavioral measure on anatomy Pictures from the IAPS were presented to subjects with a 4sec presentation time for each picture. An EBR was induced by an auditory probe randomly at the one of the three predetermined timings. 9 trials were made for each picture condition and timing, resulting 81 trials in total. EBRs were recorded using EMG.

23 Page  23 Effect of behavioral measure on anatomy

24 Page  24 Outline Introduction Method Results Conclusion

25 Page  25 Conclusion A new subcortical structure shape modeling framework based on the sparse representation of Fourier coefficients constructed with the LB eigenfunction. The framework demonstrated higher sensitivity in modeling shape variations compared to the traditional volumetric analysis. The ability to localize subtle morphological difference may provide an anatomical evidence for the functional organization within human subcortical structures.

26


Download ppt "Sparse Shape Representation using the Laplace-Beltrami Eigenfunctions and Its Application to Modeling Subcortical Structures Xuejiao Chen."

Similar presentations


Ads by Google