1. 2006 Mouse AHM Mapping 2D slices to 3D atlases - Application of the Digital Atlas Erh-Fang Lee Laboratory of NeuroImage UCLA

2. Ultimate Goal of Mouse Atlas Project Establish a framework and working toolset that integrates anatomical and gene expression data into a centralized and easily accessible resource.

3. Purpose of 2D-to-3D registration Map the atlas information to the input image Find the standard space to house the input data

4. Real Example : The Bmp6 expression data set of P7 from GENSAT database

5. Find an approximate plane from the atlas Derive a sub-atlas which includes ROI only Workflow for 2D Data Standardization and Incorporation 2D sub-atlas labels 2D input image Atlas-based Registration/ Segmentation Data Federation

6. Ideal Implementation for Template Retrieving Modality independent. 2D orientation independent and tolerant to the variation in scale to certain range. Minimal manual data preprocessing. Quick search and comparison. Region of interest is registered.

7. Proposed Approach Find the searching range which generates the 2D sub-atlas containing the structure of interest. Digitally section over the sub-volume and generate 2D atlas planes.

8. Generate the 2D atlas planes by digitally sectioning over the brain

9. Proposed Approach Find the searching range which generates the sub- atlas containing all structures of interest. Digitally section over the sub-volume and generate 2D atlas planes. Find the plane which is most similar to the input image from the set of 2D atlases.

10. Searching approach based on brain shape comparison Modified from “Ryutarou Ohbuchi, Tomo Otagiri, Masatoshi Ibato, Tsuyoshi Takei: Shape-Similarity Search of Three-Dimensional Models Using Parameterized Statistics. 2002” The shape of the brain is modeled as features vector which is parameterized statistics of the contour along the two principle axes of inertia. The shape distance between two brains is the Euclidean distance between the feature vectors formed from these parameters.

11. Parameterized statistics for shape comparison 1.Parse the brain contours from masked data using a 4- neighbor edge detector. 2.Align each comparison set using the center of mass and the principle axes of inertia. 3.The brain is subdivided into slabs along the two principle axes of inertia. The brain shape is modeled as the combination of feature vectors composing of 1.The moment of inertia 2.The average distance of a contour point from the axis 3.The variance of distance of a contour point from the axis

12. Shape model and the definition of distance Shape model : Shape distance :

13. Parameterized statistics for shape comparison –A bunny model example of parameterized statistics along one axis Ryutarou Ohbuchi, Tomo Otagiri, Masatoshi Ibato, Tsuyoshi Takei: Shape-Similarity Search of Three- Dimensional Models Using Parameterized Statistics. 2002

14. Modification from Ohbuchi’s Model Includes options for the shape distance computation –Weighted the statistics with axis ratio –Weighted the statistics with the product of axes

15. Implementation

16. The Bmp6 expression data set of E16 from GENSAT database

17. Result of Serial Insertion of the GENSAT images

18. The Bmp6 expression data set of P7 from GENSAT database

19. Map the P0 Atlas to GENSAT data of P7

20. Observation from some experiment Better retrieving for planes from sagittal and horizontal sections Using less weight for variation term in feature vectors improves the retrieving for distorted images. Manually adjustment, if necessary, is usually within few pixels and 10 degrees to obtain a better approximation. Less robust for planes from more distal part of the brain Sensitive to the “integrity” of the brain contours

21. Possible methods to improve the accuracy of retrieving More anatomical restriction to narrow the searching range. Includes boundaries of some anatomical structures into the shape model. –Additional delineation on input image would be required

22. Conclusion Alleviate the effort on atlas navigation for users who have mild knowledge in brain anatomy. Potential Usage : –Derive a 2D reference space(s) for further nonlinear registration. –Spatially query the atlas information. –Provide a template for atlas-based segmentation and registration.

23. Find an approximate plane from the atlas Derive a sub-atlas which includes ROI only Workflow for 2D Data Standardization and Incorporation 2D sub-atlas labels 2D input image Atlas-based Registration/ Segmentation Data Federation