Detection of Neural Activity in fMRI Using ... - Purdue University
extended embed settings
NeuroImage 15, 386–395 (2002) doi:10.1006/nimg.2001.0972, available online at http://www.idealibrary.com on Detection of Neural Activity in fMRI Using Maximum Correlation Modeling Ola Friman, Magnus Borga, Peter Lundberg,* and Hans Knutsson Department of Biomedical Engineering and *Departments of Radiation Physics and Diagnostic Radiology, Linköping University, Linköping, Sweden Received March 6, 2001 A technique for detecting neural activity in functional MRI data is introduced. It is based on a novel framework termed maximum correlation modeling. The method employs a spatial filtering approach that adapts to the local activity patterns, which results in an improved detection sensitivity combined with good specificity. A spatially varying hemodynamic response is simultaneously modelled by a sum of two gamma functions. Comparisons to traditional analysis methods are made using both synthetic and real data. The results indicate that the maximum correlation modeling approach is a strong alternative for analyzing fMRI data. © 2002 Elsevier Science I. INTRODUCTION Two main problems make the detection of neural activity in fMRI difficult; the noisy character of the fMRI data and the variability of the hemodynamic response. The noise problem is traditionally addressed by smoothing the fMRI images with a Gaussian filter. Referring to the matched filter theorem, it is claimed that an increase in signal to noise ratio (SNR) is obtained (Worsley et al., 1996). However, Gaussian-shaped activation patterns are rarely seen at the current resolution of the fMRI images and certainly not at the sizes of the filter kernels frequently applied. Therefore, smoothing with a Gaussian-like kernel is far from optimal when the aim is to improve SNR and the effect can very well be the opposite, e.g., some small activated area is drowned by the surrounding noise when the filter is applied. The second problem, the nature of the hemodynamic response, has drawn much attention and several models has been proposed (Friston et al., 1994; Bullmore et al., 1996; Lange and Zeger, 1997; Josephs et al., 1997). The shape of the response has been shown to vary significantly (Aguirre et al., 1998), implying that methods that incorporate this variability have greater detection sensitivity. A detection problem in general can be divided into two separate problems, feature extraction and classification. In fMRI, feature extraction amounts to the calculation of a map of some statistical parameter, for example a correlation map or a t map. The classification problem is usually solved by thresholding the statistical parameter map in order to classify each voxel as either active or nonactive. The success in solving the classification problem crucially depends on the performance of the preceding feature extraction step. Much research on detecting brain activity in fMRI data has been devoted to the classification problem, where advanced methods for finding the correct threshold has been developed (Worsley and Friston, 1995; Poline et al., 1997). In contrast, although there is an awareness that a fixed Gaussian filter may not provide optimal detection sensitivity (Kruggel et al., 1999; Shafie et al., 1998), methods such as basic cross-correlation analysis and t-tests accompanied with Gaussian smoothing have been the dominating way of producing the statistical parameter maps. The aim in this paper is to develop a method that improves the feature extraction step, making the ensuing classification part of the detection problem simpler. The technique presented is a further development of an earlier reported method based on canonical correlation analysis (Friman et al., 2001). The feature extraction problem is approached in a somewhat unconventional way by viewing the correlation between two timeseries as an objective function to be maximized with respect to some given parameters. Prior knowledge is built in by imposing constraints on these parameters. Using this approach we overcome uncertainties in hemodynamic response modelling and which spatial filter to apply in order to improve SNR. Instead of routinely smoothing with a Gaussian filter and using a fixed model for the hemodynamic response the proposed approach adaptively finds the best response model and spatial filter to obtain maximum correlation. Therefore the name maximum correlation modeling has been adopted (Tofallis, 1999). 1053-8119/02 $35.00 © 2002 Elsevier Science All rights reserved. 386
Radiation Safety Training for Use of Radioactive ... - Purdue University
An Introduction to fMRI - Research - University of California, Santa ...
A Method for Detecting Space Cluster Using Geographic Raster ...
Detection Method Using Pseudo Injured Code in LR-UWB system
Activation detection in fMRI using a maximum energy ... - IEEE Xplore
Detection and detrending in fMRI data analysis - Purdue University
Spatial mixture modeling of fMRI data - Purdue University
Detection of the early decrease in fMRI signal in ... - Purdue University
Detection of the early negative response in fMRI ... - Purdue University
Reliability of functional localization using fMRI - Purdue University
Variability of fMRI activation during a phonological - Purdue University
Assessing the sensitivity of fMRI group maps - Purdue University
Cluster analysis of fMRI data using dendrogram ... - Purdue University
Influence of acoustic masking noise in fMRI of ... - Purdue University
Characterizing the Response of PET and fMRI ... - Purdue University
Neural correlates of visual form and visual ... - Purdue University
Functional imaging and neural information coding - Purdue University
Neural correlates of auditory sensory memory ... - Purdue University
Model assessment and model building in fMRI - Purdue University
Bias and Efficiency in fMRI Time-Series Analysis - Purdue University
Detection of Consistently Task-Related ... - Purdue University
Modeling Brain Activation in fMRI Using Group MRF - Biomedical ...
Multivariate Model Specification for fMRI Data - Purdue University
Structural neural correlates of prosaccade and ... - Purdue University
Enhanced fMRI Response Detection and Reduced Latency through ...
Spatial Characterization of fMRI Activation Maps Using Invariant 3-D ...
The neural basis of executive function in working memory: an fMRI ...
Venous blood effects in spin-echo fMRI of ... - Purdue University
Simulation of human respiration in fMRI with a ... - Purdue University
Event-Related fMRI: Characterizing Differential ... - Purdue University
Magazine: Detection of Neural Activity in fMRI Using ... - Purdue University