Elastic registration of electrophoresis images using intensity information

Elastic registration of electrophoresis images using intensity information and point landmarks Source: Pattern Recognition, 37(5), P. 1035 -1048, 2004 Authors: K. Rohr, P. Cathier, S. Worz Speaker: Chia-Chun Wu (吳佳駿) Date: 2005/02/24 1

Outline n n Introduction Proposed method Experimental results Conclusions 2

Introduction Fig. 2. Original electrophoresis image pair and marked landmarks. 3

Proposed method n Extraction of point landmarks n Elastic image registration 4

Extraction of point landmarks n A model fitting approach n 2 D Gaussian function 5

Extraction of point landmarks Fig. 1 a. Example spot from electrophoresis image: intensities (left), 3 D plots of the intensities (middle), and 3 D plots of the fitted models (right). 6

Extraction of point landmarks Fig. 1 b. Example spot from electrophoresis image: intensities (left), 3 D plots of the intensities (middle), and 3 D plots of the fitted models (right). 7

Extraction of point landmarks Fig. 1 c. Example spot from electrophoresis image: intensities (left), 3 D plots of the intensities (middle), and 3 D plots of the fitted models (right). 8

Extraction of point landmarks ơx, ơy: standard deviations a 0: background intensity a 1: peak intensity 9

Extraction of point landmarks Parametric intensity model: Minimize Let 10

Extraction of point landmarks 11

Extraction of point landmarks 12

Elastic image registration n n Registration algorithm: PASTAGA( PASha Treating Additional Geometric Attributes) algorithm[17][29] Using prominent point landmarks as geometric features 13

Experimental results n Parameter settings n n n Image size: 1024× 1024 pixels Landmark extraction: 3~10 points Size of ROI: 21× 21 or 31× 31 pixels 14

Experimental results (1) Fig. 2. Original electrophoresis image pair (easy example) and marked landmarks. 15

Experimental results (1) Fig. 5. Deformed grid according to the registration result using landmarks of the images in Fig. 2. 16

Experimental results (1) Fig. 3. Registration result of the images in Fig. 2 (contour overlay): without landmarks (left) and using landmarks (right). 17

Experimental results (1) Fig. 4. Enlarged sections of Fig. 3. 18

Experimental results (2) Fig. 6. Original electrophoresis image pair (medium example) and marked landmarks. 19

Experimental results (2) Fig. 7. Registration result of the images in Fig. 6 (contour overlay): without landmarks (left) and using landmarks (right). 20

Experimental results (2) Fig. 8. Enlarged sections of Fig. 7. 21

Experimental results (3) Fig. 9. Original electrophoresis image pair (difficult example) and marked landmarks. 22

Experimental results (3) Fig. 10. Registration result of the images in Fig. 9 (contour overlay): without landmarks (left) and using landmarks (right). 23

Experimental results (3) Fig. 11. Enlarged sections of Fig. 10. 24

Experimental results (4) Fig. 12. Original electrophoresis image pair (Compugen example) and marked landmarks. 25

Experimental results (4) Fig. 13. Registration result of the images in Fig. 12 (contour overlay): without landmarks (left) and using landmarks (right). 26

Experimental results (4) Fig. 14. Enlarged sections of Fig. 13. 27

Experimental results 28

Experimental results 29

Conclusions n n An approach for elastic registration of 2 D gel electrophoresis image using intensity and landmark information. Improve registration accuracy for images of easy and medium complexity. 30