10 May 2007 Dimensionality estimation for group fMRI data reduction at multiple levels
Author Affiliations +
Abstract
Current techniques substantially overestimate the dimensionality of group fMRI data, and this problem worsens when principle component analysis (PCA) based data reductions are applied at multiple levels. In this paper, the mechanism of the overestimation is investigated, and a new method is developed for more reliable dimensionality estimation for group fMRI data at multiple levels. Simulation suggests that small variation of the signal components within a group is a major cause of dimensionality overestimation. To obtain an improved estimation, appropriate colored noise is added into the group fMRI data in order to blur the signal component variations. The noise parameters are estimated from the original fMRI data, and the improved dimensionality is determined by applying a first-order autoregressive (AR(1)) noise fitting technique to the PCA spectrum. The proposed method was tested on group resting-state fMRI datasets acquired from 14 normal human subjects in 5 different sessions. The PCA-based data reductions were performed at 3 levels in either "individual-session-subject" or "individual-subject-session" order. Results indicate that the proposed method significantly reduces the dimensionality overestimation for multiple level data reductions. Consistency of the estimated dimensionalities is observed with different group orders of the data reduction.
© (2007) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Sharon Chen, Thomas J. Ross, Keh-Shih Chuang, Elliot A. Stein, Yihong Yang, Wang Zhan, "Dimensionality estimation for group fMRI data reduction at multiple levels", Proc. SPIE 6511, Medical Imaging 2007: Physiology, Function, and Structure from Medical Images, 651116 (10 May 2007); doi: 10.1117/12.710046; https://doi.org/10.1117/12.710046
PROCEEDINGS
10 PAGES


SHARE
Back to Top