J Nucl Med. 2008; 49 (Supplement 1):379P
Instrumentation & Data Analysis: Data Analysis & ManagementData Analysis & Management Posters |
Fast and reliable parametric image formation using adaptive fuzzy clustering
Lingfeng Wen1,
Stefan Eberl2 and
Dagan Feng3
1 University of Sydney, Sydney, New South Wales, Australia;
2 Royal Prince Alfred Hospital, Sydney, New South Wales, Australia;
3 Hong Kong Polytechnic University, Hong Kong, China
1601
Objectives: Fuzzy clustering (FC) can provide improved SNR for constructing parametric images from noisy functional data, but its computational burden makes it impractical, particularly for large numbers of clusters. We proposed a novel adaptive FC (AFC) approach to substantially reduce computation time, while maintaining its superior performance and evaluated it with the model-aided GLLS method (JNM, 48:S2, 157P, 2007) in the formation of multiple parametric images.
Methods: The proposed AFC method initially clusters the voxels into a lower number of clusters, S1. For the next higher level of cluster numbers, the voxels of each of the S1 clusters are clustered into S2 clusters, giving S1*S2 clusters at this level and so on until the required number of clusters is reached. The fuzzy memberships (FM) are finally combined across the levels. Cluster centroids were fitted with model-aided GLLS and parametric images were constructed from the parameters with maximum FM. AFC was evaluated with 20 sets of simulated dynamic SPECT data using the kinetics of 5-123I-iodo-A-85380 at a moderate level of noise.
Results: The percentage bias and coefficient of variation (CV) for two parameters and three regions of interest are compared in the table below for the method using 1 (standard FC) to 4 levels.
Conclusions: The proposed AFC method provides fast and reliable parametric images for challenging SPECT data. The introduced granular levels in AFC substantially improve computational efficiency while preserving accuracy and reliability.
| |
% of bias (% of CV)
|
|
