Adaptive fuzzy clustering in constructing parametric images for low SNR functional imaging

Abstract

Functional imaging can provide quantitative functional parameters to aid early diagnosis. Low signal to noise ratio (SNR) in functional imaging, especially for single photon emission computed tomography, poses a challenge in generating voxel-wise parametric images due to unreliable or physiologically meaningless parameter estimates. Our aim was to systematically investigate the performance of our recently proposed adaptive fuzzy clustering (AFC) technique, which applies standard fuzzy clustering to sub-divided data. Monte Carlo simulations were performed to generate noisy dynamic SPECT data with quantitative analysis for the fitting using the general linear least square method (GLLS) and enhanced model-aided GLLS methods. The results show that AFC substantially improves computational efficiency and obtains improved reliability as standard fuzzy clustering in estimating parametric images but is prone to slight underestimation. Normalization of tissue time activity curves may lead to severe overestimation for small structures when AFC is applied.