Share Email Print
cover

Proceedings Paper

Improved estimation of parametric images of cerebral glucose metabolic rate from dynamic FDG-PET using volume-wise principle component analysis
Author(s): Xiaoqian Dai; Jie Tian; Zhe Chen
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Parametric images can represent both spatial distribution and quantification of the biological and physiological parameters of tracer kinetics. The linear least square (LLS) method is a well-estimated linear regression method for generating parametric images by fitting compartment models with good computational efficiency. However, bias exists in LLS-based parameter estimates, owing to the noise present in tissue time activity curves (TTACs) that propagates as correlated error in the LLS linearized equations. To address this problem, a volume-wise principal component analysis (PCA) based method is proposed. In this method, firstly dynamic PET data are properly pre-transformed to standardize noise variance as PCA is a data driven technique and can not itself separate signals from noise. Secondly, the volume-wise PCA is applied on PET data. The signals can be mostly represented by the first few principle components (PC) and the noise is left in the subsequent PCs. Then the noise-reduced data are obtained using the first few PCs by applying 'inverse PCA'. It should also be transformed back according to the pre-transformation method used in the first step to maintain the scale of the original data set. Finally, the obtained new data set is used to generate parametric images using the linear least squares (LLS) estimation method. Compared with other noise-removal method, the proposed method can achieve high statistical reliability in the generated parametric images. The effectiveness of the method is demonstrated both with computer simulation and with clinical dynamic FDG PET study.

Paper Details

Date Published: 9 March 2010
PDF: 10 pages
Proc. SPIE 7626, Medical Imaging 2010: Biomedical Applications in Molecular, Structural, and Functional Imaging, 76260H (9 March 2010); doi: 10.1117/12.843876
Show Author Affiliations
Xiaoqian Dai, Institute of Automation (China)
Jie Tian, Institute of Automation (China)
Zhe Chen, Institute of Automation (China)


Published in SPIE Proceedings Vol. 7626:
Medical Imaging 2010: Biomedical Applications in Molecular, Structural, and Functional Imaging
Robert C. Molthen; John B. Weaver, Editor(s)

© SPIE. Terms of Use
Back to Top