To smooth or not to smooth? ROC analysis of perfusion fMRI data Jiongjiong Wang a,c, * , Ze Wang b,c , Geoffrey K. Aguirre b,c , John A. Detre b,c a Department of Radiology, University of Pennsylvania, Philadelphia, PA 19104, USA b Department of Neurology, University of Pennsylvania, Philadelphia, PA 19104, USA c Center for Functional Neuroimaging, University of Pennsylvania, Philadelphia, PA 19104, USA Received 3 August 2004; accepted 15 November 2004 Abstract Blood oxygenation level dependent (BOLD) contrast has been widely used for visualizing regional neural activation. Temporal filtering and parameter estimation algorithms are generally used to account for the intrinsic temporal autocorrelation present in BOLD data. Arterial spin labeling perfusion imaging is an emerging methodology for visualizing regional brain function both at rest and during activation. Perfusion contrast manifests different noise properties compared with BOLD contrast, represented by the even distribution of noise power and spatial coherence across the frequency spectrum. Consequently, different strategies are expected to be employed in the statistical analysis of functional magnetic resonance imaging (fMRI) data based on perfusion contrast. In this study, the effect of different analysis methods upon signal detection efficacy, as assessed by receiver operator characteristic (ROC) measures, was examined for perfusion fMRI data. Simulated foci of neural activity of varying amplitude and spatial extent were added to resting perfusion data, and the accuracy of each analysis was evaluated by comparing the results with the known distribution of pseudo-activation. In contrast to the BOLD fMRI, temporal smoothing or filtering reduces the power of perfusion fMRI data analyses whereas spatial smoothing is beneficial to the efficacy of analyses. D 2005 Elsevier Inc. All rights reserved. Keywords: Arterial spin labeling (ASL); Cerebral blood flow (CBF); Receiver operator characteristic (ROC); Spatial smoothing; Temporal smoothing; Functional brain imaging 1. Introduction Functional magnetic resonance imaging (fMRI) based on the blood oxygenation level dependent (BOLD) contrast has become a standard method for visualizing regional neural activation. A consensus has emerged that valid statistical inference, hence plausible interpretation of the neuronal mechanism, can only be achieved by meticulous choice of appropriate data analysis methods in BOLD fMRI. This concern regarding the validity of neuroimaging studies using fMRI arises mainly due to a robust observation that BOLD image series possess temporal autocorrelation or bsmoothness,Q manifested as elevated power in the lower frequency range of the power spectrum accompanied by less dominating broad-band components [1–5] . Various approaches have been introduced to accommodate serial correlation in the context of parameter estimation with general linear model (GLM). A general dichotomy of these methods has been proposed by Friston et al. [6], which includes (1) bwhitening Q of the time series with an exquisite understanding of the noise characteristics [7,8] and (2) bsmoothing Q (bfiltering Q) of the time series to impose a known correlation structure [9]. It has been proposed that band-pass filtering, and implicitly smoothing, provides an optimal solution to minimize bias while maintaining a reasonable degree of efficiency in BOLD fMRI. Arterial spin labeling (ASL) perfusion imaging is an emerging and alternative methodology for functional neuro- imaging studies [10,11]. In contrast to BOLD fMRI that relies on the susceptibility effects in and around the venous vasculature [12], ASL perfusion contrast is based on alternations in the longitudinal relaxation of brain tissue caused by changes in regional blood flow [13]. Conse- quently, perfusion contrast may yield more specific func- tional localization [14,15] and reduced sensitivity to static susceptibility effects [16], although the size of signal changes induced by task activation is generally smaller in ASL than BOLD fMRI. Because perfusion image series are generated by pair-wise subtraction of temporally adjacent 0730-725X/$ – see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2004.11.009 * Corresponding author. Tel.: +1 215 614 0631; fax: +1 215 349 8260. E-mail address: jwang3@mail.med.upenn.edu (J. Wang). Magnetic Resonance Imaging 23 (2005) 75 – 81