Presentation is loading. Please wait.

Presentation is loading. Please wait.

Yen Le Computation Biomedicine Lab Advisor: Dr. Kakadiaris 1 Automatic Multi-Region Segmentation Applied to Gene Expression Image from Mouse Brain.

Similar presentations


Presentation on theme: "Yen Le Computation Biomedicine Lab Advisor: Dr. Kakadiaris 1 Automatic Multi-Region Segmentation Applied to Gene Expression Image from Mouse Brain."— Presentation transcript:

1 Yen Le Computation Biomedicine Lab Advisor: Dr. Kakadiaris 1 Automatic Multi-Region Segmentation Applied to Gene Expression Image from Mouse Brain

2 Problem Statement Problem Statement: Segmentation anatomical regions of mouse brain gene expression images (in 2D or 3D) Data: In Situ Hybridization (ISH) images 2 Motivation: – Identify and associate the location and extent of expression of a gene in mouse brain image – Understand how genes regulate the biological process at cellular and molecular levels

3 Challenges Large variations in boundary shape 3

4 4 Large variations in the shape of the anatomical regions Challenges (2)

5 Large variations in intensity 5 Challenges (3)

6 Accomplishments to-date 2D – Geometric model to image fitting methods – Image-to-image registration method 3D – Descriptors for 3D landmark detection 6

7 3D Dense Local Point Descriptors Motivation – Need for anatomical landmarks – Need 3D local point descriptors which can: Be computed fast at densely sampled points Result in accurate landmark point detection 7

8 DAISY3D and DAISYDO – Extended from DAISY descriptor – Faster than SIFT-3D, n-SIFT at densely sampled points – Good for landmark detection on gene expression images DAISY3D vs. DAISYDO – DAISYDO requires less memory than DAISY3D – DAISYDO is faster – Comparable performance 8 3D Dense Local Point Descriptors (2)

9 3D Dense Local Point Descriptors (3) 9 DAISY’s configuration Configuration Forming DAISY feature vectorForming DAISYDO feature vector

10 10 Computational Time All methods are implemented in C++ and run in single core 1.86 GHz CPU

11 Memory Requirement 11 Memory requirements for a sample volume of size 100x100x100

12 12 Performance Evaluation Detected landmarks: voxels having the minimum - distance between its descriptor and the descriptor of referenced landmark Mean error (in voxels) for landmark detection in gene expression image

13 Publications 13 Refereed Journal Articles Yen H. Le, U. Kurkure, I. A. Kakadiaris, “Dense Local Point Descriptors for 3D Images,” Pattern Recognition (Submitted). U. Kurkure, Yen H. Le, N. Paragios, J. Carson, T. Ju, I. A. Kakadiaris, “Landmark- Constrained Deformable Image Registration of Gene Expression Images for Atlas Mapping,” NeuroImage, Elsevier Science (Submitted). Refereed Conference Articles Yen H. Le, U. Kurkure, N. Paragios, J. P. Carson, T. Ju, and I. A. Kakadiaris, “Similarity- based appearance prior for fitting subdivision mesh in gene expression image,” IEEE Computer Vision and Pattern Recognition 2012 (Submitted). U. Kurkure, Y. H. Le, N. Paragios, J. P. Carson, T. Ju, and I. A. Kakadiaris. “Landmark/image-based deformable registration of gene expression data,” In Proc. IEEE Computer Vision and Pattern Recognition, pages 1089–1096, Colorado Springs, CO, Jun. 21-23 2011. U. Kurkure, Y. H. Le, N. Paragios, J. Carson, T. Ju, and I. A. Kakadiaris, Nov. 6-13 2011, “Markov random field-based fitting of a subdivision-based geometric atlas,” In: Proc. IEEE International Conference on Computer Vision. Barcelona, Spain, pp. 2540– 2547.

14 Thank You for your attention! 14


Download ppt "Yen Le Computation Biomedicine Lab Advisor: Dr. Kakadiaris 1 Automatic Multi-Region Segmentation Applied to Gene Expression Image from Mouse Brain."

Similar presentations


Ads by Google