Poster Presentation No.: 024 A hybrid wavelet-ICA model for dynamic PET analysis Hong-Ren Su, J.A.D. Aston, M. Liou, P.E. Cheng Academia Sinica, Taipei, Taiwan Independent component analysis (ICA) has been widely used in data analysis and decomposition for neuroimaging. It, typically, aims to solve the blind source separation problem in which a set of unknown sources is mixed in some way to form the data. Wavelet analysis has also become routinely used in neuroimaging, especially in PET image analysis. Here, we investigate the feasibility of combining ICA with wavelet models in order to integrate the advantages of both underlying techniques. Wavelets are well known to have sparse signal representation, and thus it should be more feasible to decompose the underlying unknown sources (Roberts et al., 2003). ICA is not a linear technique, and hence analysis in the image domain and wavelet domain will not lead to equivalent solutions. We have investigated using ICA dynamic PET data both in the image domain and in the wavelet domain, where the data had been transformed using Battle-Lemarie wavelets as in (Turkheimer et al., 2003). That data was analyzed with ICA, resulting in two kinds of components — a spatial component and a temporal component (Bell et al., 1995). Each component pair gives the spatial distribution of that temporal component. For each pair, the mean variance of the component can be calculated and used to rank the components in a similar way to eigenvalues in principle component analysis. Hybrid wavelet-ICA should require far fewer components for signal representation than its image domain equivalent as the signal and noise are well separated. Below are the results of applying the method to a [11C]-Raclopride scan. As can be seen, better denoising Fig. 1. Comparison of ICA and wavelet-ICA for a [ 11 C]-Raclopride data set. Component pairs – spatial and temporal – are generated. Using the mean variance of each component pair, the five components with largest mean variance are preserved and used to estimate BP. Poster Presentations / NeuroImage 31 (2006) T44– T186 T67