ICA vs SPM for Similarity Retrieval of f
- Slides: 16
ICA vs. SPM for Similarity Retrieval of f. MRI Images
Overview of Sara’s and Rosalia’s Method
ICA (Sara’s Method) f. MRI raw Data fast. IC A Extract Spatial Feature Vectors 108 components IC maps Each component associates with timecourse Similarity between Component s
SPM (Rosalia’s Method) SPM Z Score with FDR=0. 05 Get SPM clusters (connected Component analysis) Compute Feature Vectors Similarity Between Two Brains
How to Compare these two methods? �Similarity between components => similarity between brains? ? �Bai etc. (2007) ◦ Maximum Weight Bipartite Matching used f. MRI brain image retrieval based on ICA components.
Optimal Assignment and Bipartite Matching Given two sets s 1 and S 1 S 2 s 2, is defined as the cost between each element i in S 1 and each element j in S 2. An optimal assignment is a permutation p = (p 1, . . . , pn) of the integers (1, . . . , n) that minimizes
Example of brains with two components 0. 7 1 1 =0. 3 =0. 2 2 Brain 1 0. 8 2 Brain 2 Minimum Cost = C 12+C 21 = 0. 2+0. 3 = 0. 5 Similarity Score =0. 5
Apply Optimal Assignment �Step 1. Convert Similarity Matrix to a cost Matrix �Step 2. Find the minimum cost between two brains using Bipartite Matching.
The comparison �SPM data: Face. Up. Vs. Fixation �ICA data: raw f. MRI data of 108 scans �Run Rosalia’s method to get a similarity Matrix M 1 �Run Sara’s method and bipartite matching to get similarity Matrix M 2
The results �Correlation between M 1 and M 2: 0. 7132? ? ? �Not a very strong correlation
Possible Reasons �Convert Similarity Matrix to Cost C 1 Matrix? �Raw f. MRI data has more than two cognitive tasks, while SPM contrast map has only two cognitive tasks. C 2 Fix Face. Up House. Up Fix House. Up Fix �Is Bipartite Matching a good measure for similarity between two brains?
Reference �B. Bai, P. Kantor, A. Shokoufandeh, D. Silver. f. MRI brain image retrieval based on ICA components. enc, pp. 10 -17, Eighth Mexican International Conference on Current Trends in Computer Science (ENC 2007), 2007 �Y. Cheng, V. Wu, R. Collins, A. Hanson, and E. Riseman. Maximum-weight bipartite matching technique and its application in image feature matching. In Proc. SPIE Visual Comm. And Image
Component Similarity Measure �Each feature is weighted according to its mean and std: � �if < threshold �Similarity Score:
SPM (Rosalia’s Method) �Preprocess t-contrast maps of particular cognitive tasks to get the activated voxels. �Cluster the resulting voxels to distinct regions. �Similarity Measure between brains ◦ Q-to-T Score = ◦ T-to-Q Score = ◦ Similarity Score =
Example The cost between S 1 and S 2 can be represented by a nxn matrix 123 C 11+C 22+C 33 132 C 11+C 23+C 32 321 C 13+C 22+C 31 231 C 12+C 23+C 31 C 22 C 23 213 C 12+C 21+C 33 C 31 C 32 C 33 312 C 13+C 21+C 32 S 2 C 11 C 12 C 13 S 1 P
Apply Optimal Assignment �Step 1. Convert Similarity Matrix (s. M)to a cost Matrix (c. M) ◦ Replace infinite value in s. M to the maximum value of all the finite values. ◦ Take top N percent of the similarity scores, set others to be zero. ◦ Set the non-zero scores to be 1 SM/maximum. ◦ set the zeros to be infinite value. �Step 2. Using Kuhn-Munkres Algorithm (Hungarian method) to find the minimum cost between two brains.