Human Bain Mapping 28:703 720 (2007) Assessing the Influence of Scanne Backgound Noise on Auditoy Pocessing. I. An fmri Study Compaing Thee Expeimental Designs with Vaying Degees of Scanne Noise Nadine Gaab, 1,2 * John D.E. Gabieli, 1 and Gay H. Glove 2 1 Depatment of Psychology, Stanfod Univesity, Stanfod, Califonia 2 Depatment of Radiology, Stanfod Univesity, Stanfod, Califonia Abstact: We compaed two expeimental designs aimed at minimizing the influence of scanne backgound noise (SBN) on functional MRI (fmri) of auditoy pocesses with one conventional fmri design. Ten subjects listened to a seies of fou one-syllable wods and had to decide whethe two of the wods wee identical. This was contasted with a no-stimulus contol condition. All thee expeimental designs had a duation of 17 min: 1) a behavio inteleaved gadients (BIG; Eden et al. [1999] J Magn Reson Imaging 41:13 20) design (epetition time, TR, ¼ 6 s), whee stimuli wee pesented duing the SBN-fee peiods between clusteed volume acquisitions (CVA); 2) a spase tempoal sampling technique (STsamp; e.g., Gaab et al., [2003] Neuoimage 19:1417 1426) acquiing only one set of slices following each of the stimulations with a 16-s TR and jitteed delay times between stimulus offset and image acquisition; and 3) an event-elated design with continuous scanning (ERcont) using the stimulation design of STsamp but with a 2-s TR. The esults demonstated inceased signal within Heschl s gyus fo the STsamp and BIG-CVA design in compaison to ERcont as well as diffeences in the oveall functional anatomy among the designs. The possibility to obtain a time couse of activation as well as the full ecovey of the stimulus- and SBN-induced hemodynamic esponse function signal and lack of signal suppession fom SBN duing the STsamp design makes this technique a poweful appoach fo conducting auditoy expeiments using fmri. Pactical stengths and limitations of the thee auditoy acquisition paadigms ae discussed. Hum Bain Mapp 28:703 720, 2007. VC 2006 Wiley-Liss, Inc. Key wods: fmri; auditoy cotex; scanne backgound noise; spase tempoal sampling; clusteed volume acquisition INTRODUCTION Contact gant sponso: National Institutes of Health (NIH); Contact gant numbe: RR09784. *Coespondence to: Nadine Gaab, Depatment of Bain and Cognitive Sciences, Massachusetts Institute of Technology, 77 Massachusetts Ave., Room NE20-423, Cambidge, MA 02138-4307. E-mail: gaab@mit.edu Received fo publication 17 Octobe 2005; Accepted 1 May 2006 DOI: 10.1002/hbm.20298 Published online 1 Novembe 2006 in Wiley InteScience (www. intescience.wiley.com). Functional MRI (fmri) povides noninvasive imaging of bain activation without the use of adioactive taces, contast agents, o electodes. Moe and moe eseach goups use this technique to exploe the human bain fo both basic eseach and clinical applications. Howeve, assessing auditoy functions in the MR envionment is poblematic because the pocess of image acquisition geneates a highamplitude MR-scanne backgound noise (SBN) [Cho et al., 1997; Counte et al., 1997; McJuy et al., 1995, 2000; Pice et al., 2001; Shellock et al., 1994, 1998]. The goal of this VC 2006 Wiley-Liss, Inc.
Gaab et al. study was to diectly compae two methods that have been developed to minimize the effects of SBN on auditoy fmri studies, and to compae these to a conventional method in fmri eseach that employs continuous scanning. The main question was whethe these vaious methods yield diffeent esults (e.g., pattens and intensity of activation) unde simila expeimental conditions. A follow-up question assessed whethe one o anothe esult appeaed moe likely to be valid, and whethe the diffeent methods included specific advantages o disadvantages. Supeio o infeio measuement of auditoy activations could have majo influences on the outcomes of auditoy fmri eseach. The SBN can be quite loud. Indeed, seveal studies epoted noise levels of moe than 100 db SPL (sound pessue level) fo the SBN, which coesponds appoximately to listening to a jackhamme fom close distance. SBN amplitude vaies depending on factos such as the pulse sequence applied o the numbe of slices acquied [fo eviews, see Amao et al., 2002; Moelke and Pattynama, 2003]. Futhemoe, seveal studies evealed a positive elation between field stengths and noise incease [e.g., Counte et al., 2000; Pice et al., 2001]. Thus, SBN becomes an inceasingly impotant facto as moe imaging centes pefom eseach at o above 3.0 T. SBN can impede auditoy functional neuoimaging in at least two ways: 1) SBN can intefee with paticipants accuately heaing auditoy stimuli, and 2) SBN can povoke activation that masks stimulus o task-diven cotical esponses of expeimental inteest. With egad to intefeence with auditoy peception of stimuli, the pesence of SBN can lead to inceased pue tone heaing thesholds as well as deceased pefomance in humans and expeimental animals [e.g., Ulme et al., 1998b], especially if auditoy stimulation contains fequencies simila to those measued fo the SBN [e.g., Scaff et al., 2004]. With egad to intefeence with activation, SBN itself (eal o ecoded) leads to inceased activations of auditoy [e.g., Bandettini et al., 1998] and language aeas [Ulme et al., 1998a], and esembles a typical hemodynamic esponse function (HRF) induced by an auditoy stimulus [e.g., Glove et al., 1999; Hall et al., 2000]. Futhemoe, the SBN-induced activation seems to be highly vaiable among subjects [Ulme et al., 1998a]. The SBN may also lead to alteed auditoy cotical esponse due to inceased baseline levels [e.g., Talavage and Edmiste, 2004]. This SBN-induced masking of the BOLD esponse descibes a eduction of the signal intensities o spatial spead within auditoy aeas as a esult of diffeences of the SBN-induced activation in the expeimental and baseline condition (e.g., a task without auditoy stimulation, a task with lowe cognitive demands, o simply silence). Simply listening to SBN evokes stong signal inceases within auditoy aeas and adding SBN to an auditoy task does not enhance the signal to the same degee, which means that the activations povoked by an auditoy stimulus and SBN ae not additive [Gaab et al., 2006; Talavage and Edmiste, 2004]. This effect of an elevated baseline leads to a eduction of signal intensities when contasting expeimental condition and baseline. Seveal studies epoted SBN-induced masking of the blood oxygenation level-dependent (BOLD) esponse in designs with inceased SBN that led to diffeences in the auditoy esponse patten as well as deceases in the spatial spead o signal intensities fo activated auditoy egions and evealed inceased signal intensities in designs with less o no SBN [e.g., Baumgat et al., 1996; di Salle et al., 2001; Hall et al., 1999; Shah et al., 1999, 2000; Yetkin et al., 2003]. Futhemoe, these SBN-induced satuation effects seem to change with the fequency of the pesented tones [e.g., Langes et al., 2005]. Nevetheless, most fmri studies contast at least two active conditions (e.g., music and language) and if SBN is pesent in both conditions it could be that the SBN effects would cancel out. That this is not the case was evealed by Tomasi et al. [2005], who vaied both the degee of SBN and the woking memoy (WM) load. Thee wee WM load-dependent inceases in BOLD signals that coelated negatively with inceased SBN thoughout the WM netwok. This suggests that two tasks diffeing in thei cognitive demands may be influenced by SBN to a diffeent degee. Futhemoe, Tomasi et al. [2005] showed that inceased scanne noise led to inceased activations in seveal extatempoal bain aeas such as the infeio, medial, and supeio fontal lingual gyus, the fusifom gyus, and the ceebellum and deceased activations in the anteio cingulate. This additional activation as well as suppession in extatempoal egions has also been shown fo the visual and moto cotex [e.g., Cho et al., 1998; Loenneke et al., 2001; Zhang and Chen, 2004]. Besides engineeing modifications of the MRI scanne hadwae and impoved headphone systems [fo eviews, see Amao et al., 2002; Moelke and Pattynama, 2003], a vaiety of diffeent scanning methods to educe SBN has been developed to ovecome the above-descibed intefeences [fo eviews, see Amao et al., 2002; Moelke and Pattynama, 2003]. These silent methods diffe in vaious aspects fom the commonly used scanning methods that employ continuous scanning such as standad block o event-elated designs with epetition times (TRs) aound 2 s (hee, ERcont). Duing an ERcont design, images ae acquied thoughout the expeiment that esults in gapless, continuous SBN. This appoach maximizes the numbe of images that can be acquied duing the time couse of the expeiment (e.g., ove 500 images in 17 min), but the influence of SBN is uncontolled and unclea. Thee ae seveal techniques that aim to acquie auditoy cotex fmri while educing the contamination fom SBN. Hee we will distinguish between designs that on each tial measue the same constant faction of the stimulus-induced HRF vs. those that use a jitteed design to measue a numbe of diffeent factions of the stimulusinduced HRF, with one faction pe tial. In the latte design, hee denoted spase tempoal sampling (STsamp), 704
Scanne Noise Influence on Auditoy Pocessing I the diffeent factions ae then combined acoss tials to sample a lage total faction of the HRF. One commonly used design that samples a constant faction of the HRF is the behavio inteleaved gadients technique (BIG) [Bilecen et al., 1996; Eden et al., 1999; Edmiste et al., 1999; Engelien et al., 2002; Le et al., 2001; Tanaka et al., 2000; Yang et al., 2000], which often implements clusteed volume acquisition (often called CVA) [e.g., Edmiste et al., 1999; Talavage et al., 1999]. CVA is chaacteized by the acquisition of all slices in apid succession at the end of one TR. This technique povides an SBN-fee peiod that allows the pesentation of auditoy stimuli without intefeences of SBN. In ode to optimize the BOLD signal, images ae usually acquied close to the hypothesized maxima of the HRF. SBN occus afte the auditoy stimulation, and duing the delayed HRF that pesumably eflects neual pocesses duing the auditoy stimulus pocessing. This technique leads to a educed numbe of images (e.g., 150 images in 17 min), but the SBN-fee pesentation of the auditoy stimuli and the patial ecovey of the SBN-induced signal may lead to impoved signal. Using a clusteed technique, Edmiste et al. [1999] showed the geatest pecent signal change within auditoy aeas fo a TR of 8 s, which is consistent with the beginning decay of the SBN-induced HRF in auditoy egions. Howeve, 8 s was the longest TR used, and it is not known if that was optimal. Shah et al. [2000] examined the ole of TR and its influence on the BOLD signal and suggest an optimal TR of 6 s. They state that longe TRs might lead to attentional effects o fatigue [see also Elliott et al., 1999], which might be accompanied by a lack of concentation due to longe measuement times. Howeve, in ode to pesent moe tials in an expeiment, many studies using BIG-CVA choose suboptimal TRs of much less than 6 8 s [e.g., Fu et al., 2002, 2005; Ojanen et al., 2005], which on top of SBN-induced masking effects may again esult in continuous auditoy stimulation (task and then SBN) and a deceased gap between task tials. A few studies have aimed to impove the sampling of the HRF within these designs by vaying the delay between the auditoy stimulus and the MR acquisition (STsamp) [e.g., Belin et al., 1999; Gaab et al., 2003; Hall et al., 1999, 2000]. Analyzing these delay times o imaging time points (ITPs) sepaately can also povide a time couse of activation [e.g., Gaab et al., 2003]. Nevetheless, due to the fact that only one image is acquied on each tial (fo example, one image in each 14 s), a educed numbe of images is collected (70 in 17 min). The question is whethe the full ecovey of the SBN-induced HRF signal and lack of suppession fom SBN, which esults in impoved signal-to-noise atio (SNR), outweigh the educed numbe of obsevations and theefoe deceased statistical powe in tems of yielding activation. The aim of the pesent study was to compae these thee diffeent expeimental designs with egad to vaying degees of SBN on signal intensity and the time couse of activation within auditoy and nonauditoy aeas. The pesent study goes beyond pio compaisons of auditoy fmri designs in two impotant ways. Fist, session length is held constant, which was not done in pio studies. This is impotant because the diffeent methods yield diffeent numbes of obsevations ove an equal session duation, and eseaches ae typically limited to a pactical session duation. Second, the numbe of stimuli and the timing was kept constant fo two of the thee designs, which allows us to diectly compae the influence of SBN without vaying the natue of the task. Thid, most pevious studies compaing designs acquied images only within auditoy cotices, and diffeences in extatempoal aeas wee not examined. The expeimental designs examined hee had TRs of 2 s (event-elated continuous design, ERcont), 6 s (behavio inteleaved gadients technique/clusteed volume acquisition design without jitteing, BIG-CVA; e.g., Eden et al. [1999]), and 16 s and a jitteed delay between auditoy stimulation and image acquisition (STsamp) [Gaab et al., 2003]. Two of the designs (ERcont and STsamp) had exactly the same auditoy stimulation, which enabled us to futhe assess the influence of SBN on signal intensity within auditoy aeas. A novel aspect of ou study is the compaison between a subset of the scan data acquied duing ERcont (ER65) that exactly matched the data sampling and auditoy stimulation timing duing STsamp, and theefoe enables us to diectly compae the influence of SBN on auditoy activation. Futhemoe, by using a spase tempoal sampling technique with jitteed delay times between stimulus offset and image acquisition, we wee able to compae the time couse of activations within auditoy and extatempoal aeas between the designs. SUBJECTS AND METHODS Paticipants Ten nomal ight-handed voluntees (age ange: 18 28, mean age: 20; five males and five females) paticipated in this study. Subjects had no histoy of neuological o heaing impaiments. Infomed consent to take pat in a study appoved by the Stanfod Univesity panel on Human Subjects in Medical Reseach was obtained fom each subject. Auditoy Setup Pocedue Auditoy stimuli wee pesented using pneumatic headphones that povide ea potection (Avotec, Stuat, FL). Tasks wee pogammed using Epime (Psychology Softwae Tools, PST, Pittsbugh, PA) unning on a PC with SoundBlaste audio cad (Ceative Technology, Singapoe). Stimuli wee sampled and pesented at 44 khz. All subjects pefomed an auditoy volume setup pocedue pio to the expeimental tasks. Afte localize and shim scans wee completed, subjects listened to wods andomly selected fom the stimulus list and wee asked to incease o decease the amplitude of the wods using a button box until they eached thei own comfotable amplitude, both while the 705
Gaab et al. scanne was unning the functional imaging scan and also while the scanne was quiescent. The amplitudes fo these two conditions wee then used fo the pesentation of the wods in the expeimental tasks accoding to whethe thee was scanne noise duing stimulus pesentation (ERcont) o not (BIG-CVA and STsamp), espectively. This pocedue optimized heaing of the wods fo all scanning conditions. Imaging Pocedue The fmri scanning was conducted with a 3.0T GE Signa scanne (Geneal Electic, Milwaukee, WI) using a custombuilt single-channel quadatue bidcage headcoil. Head motion was contolled by clamps mounted on the coil to stabilize the headphones. Sagittal T1 localize scans wee collected as well as a T1-weighted whole-bain anatomy scan (256 256 voxels, 0.86 mm in-plane esolution, 1.2 mm slice thickness) fo the puposes of nomalization of functional data into common steeotactic space. High-ode shimming was employed with a subject-specific egion of inteest (ROI) and the scanne s built-in softwae [Kim et al., 2002]. The fmri data wee collected using a spial in/out T2* pulse sequence [Glove and Law, 2001] with 30 slices coveing the entie bain (64 64 voxels, 3.43 mm in-plane esolution, TE 30 ms, 4 mm slice thickness with 0.5 mm slice skip). The imaging pocedue (TR/FA/numbe of time fames, clusteed vs. continuous acquisition) vaied fo the thee expeimental conditions (see below) but slice pesciption was kept constant, as was the appoximate duation of the scans. Duing clusteed acquisitions (BIG-CVA and STsamp), the 30 slices wee acquied in 1.995 s, while fo the continuous scan condition (ERcont) the 30 slices wee spaced evenly thoughout the 2-s TR. This choice of TR and slices kept as identical as possible the inteslice timing, gadient noise amplitude, and spectal chaacteistics fo the two acquisitions. Stimuli and Expeimental Tasks All subjects undewent thee expeimental scan uns of diffeing acquisition type and had to pefom the same task in all thee scans. They listened to a seies of fou ecoded one-syllable wods (oveall duation ¼ 4 s) and had to decide by pessing one of two buttons whethe two of the fou pesented wods wee the same o not (Fig. 1A). This expeimental condition was contasted with a silent (no-stimulus fo 4 s, Fig. 1B) contol condition in ode to measue signal intensity change within auditoy aeas due to the stimulus. This paticula task was chosen to ensue that subjects listened caefully to all fou wods and theefoe the fist and second wods wee neve identical. Nevetheless, the numbe of tials and the tempoal elation between scanne noise and auditoy stimulation vaied among the thee expeimental uns. While the button pess in the expeimental condition was not contolled Figue 1. Expeimental stimulation (A). Expeimental condition (B). Contol condition as well as image acquisition fo STsamp. The delay between the end of the auditoy stimulation and the beginning of the image acquisition was vaied ove 8 s. Each ITP coesponds to the volume acquied afte the end of auditoy stimulation, e.g., ITP0 coesponds to volumes acquied 0 s afte the end of the auditoy stimulation, and ITP5 coesponds to volumes acquied 5 s afte the end of the auditoy stimulation. in the silence condition, activation in the moto cotex was not of inteest fo this study. Oveall, 41 concete one-syllable wods spoken by a female voice wee pesented in fou-wod sequences. The ecoded wods had an aveage conceteness facto of 463 (ange: 234 614) and an aveage fequency facto of 56 (ange: 1 362) based on the MRC Psycholinguistics Database (Machine Usable Dictionay, v. 2.0). All wods wee ecoded using Audacity (http://audacity.soucefoge.net/) and wee nomalized fo thei oot-mean-squae enegy. The numbe of fou-wod sequences (n ¼ 48) did not diffe between the STsamp and the ERcont design (see below). Duing the BIG-CVA design, 106 fou-wod sequences wee pesented. The ERcont and the STsamp conditions had a minimum 4 s nonstimulus gap between auditoy stimulations, wheeas the BIG-CVA sequence had a minimum nonstimulus gap of 2 s (the duation of the SBN) if two expeimental tials followed each othe. The fequency of hit tagets was 50% fo all designs. Behavioal coelates wee obtained in pecent coect and nonpaametic tests wee used to assess possible intedesign diffeences. The thee acquisition types ae descibed below. The ode in which the thee scans wee pefomed was andomized acoss subjects. Spase tempoal sampling design (STsamp) Subjects listened to 48 sets of wods and 16 silent peiods with the same duation ove the entie scanning time. Although the TR was kept constant at 16 s (flip angle 908), the stat of the 2-s clusteed MR acquisition vaied with egad to the onset of auditoy stimulation by pseudoandomly jitteing the stat of the auditoy stimulation fame (4 s of wods 706
Scanne Noise Influence on Auditoy Pocessing I sets altogethe) was pseudoandomly inteleaved with the auditoy condition. Duing this expeimental condition 161 images wee acquied ove 16 min and 6 s. A clusteed acquisition was obtained evey 6 s duing a 2-s peiod using a flip angle of 908. Image acquisition always stated immediately afte the end of auditoy stimulation and no scanne noise was pesent duing the auditoy stimulation o contol (no stimulus) peiods. The quiescent scanne volume setting was used fo stimulus pesentation. fmri data pepocessing fmri Data Analysis Figue 2. Expeimental designs with timing of auditoy stimulation and image acquisition. A: Spase tempoal sampling (STsamp) design. B: Event-elated with continuous scanning design (ERcont). C: Clusteed volume acquisition design (BIG-CVA). Lighte gay blocks eflect auditoy/silence stimulation and dake gay (o yellow) blocks image acquisitions. D: ER65: Selecting the (in compaison to STsamp) time coesponding 65 scans (in gay/black-yellow stipes) out of the ERcont design enables the diect compaison between the two designs and the diect assessment of the influence of SBN on signal intensities. [Colo figue can be viewed in the online issue, which is available at www.intescience.wiley.com.] o silence) within the 16-s TR peiod in 1-s steps (see Fig. 2A). This jitteing pocess [fo details, see Gaab et al., 2003] esulted in eight image time points (ITPs). ITP0 coesponds to the scans acquied beginning 0 s afte the end of the auditoy stimulation, wheeas ITP5 coesponds to the images acquied beginning 5 s afte the end of the auditoy stimulation. Oveall, 66 time fames wee acquied ove the duation of 17 min and 34 s. In this case, both the contol and expeimental condition occued duing peiods of scanne inactivity, and the peviously detemined quiescent scanne volume setting was used fo stimulus pesentation. Event-elated continuous scanning design (ERcont) Fo this condition the auditoy stimulation design of the STsamp task was used but continuous scanne noise was pesent duing the entie expeiment. This method is the way in which most event-elated fmri expeiments ae pefomed (Fig. 2B). A TR of 2 s was used fo this condition with flip angle 758, which esulted in continuous scanne noise ove the entie 17 min and 12 s. Oveall, 516 time fames wee acquied in this condition. Fo these scans the volume level obtained duing setup while the scanne was opeating was used fo pesentation. Clusteed volume acquisition design (BIG-CVA) Subjects listened to 106 sets of wods duing a 2-s clusteed acquisition scan following each of the auditoy stimulations (Fig. 2C). The silent condition (4 s of silence, 53 Afte image econstuction, each set of axial images was slice time-coected, ealigned to the fist image, and coegisteed with the coesponding T1-weighted high-esolution dataset. Spatial nomalization was done in thee steps. Fist, the skull of the T1-weighted high-esolution dataset was stipped using FSL (see http://www.fmib.ox.ac.uk/ fsl/). Afte that, all T1 images wee coected using the SPM2 bias coection and then spatially nomalized to the skull-stipped SPM2 template (Monteal Neuological Institute (MNI) space). Nomalization paametes wee applied to the functional images and then functional images wee smoothed with an isotopic Gaussian kenel (4 mm fullwidth at half-maximum, FWHM). Statistical analysis of goup fmri data Statistical analysis was pefomed using paametic mapping softwae (SPM2, Wellcome Depatment of Cognitive Neuology, London, UK). The main data analysis was pefomed using a Geneal Linea Model as implemented in SPM2 [Fiston et al., 1995]. Main effects within each expeimental design. By convolving the thee diffeent task designs (using boxcas depicting each 4-s auditoy stimulus event) with a hemodynamic esponse function [Glove, 1999], thee diffeent covaiates (BIG-CVAcov, ERcontcov, STsampcov) wee developed and enteed as a egesso in thee sepaate basic model simple egession analyses. This egesso appoach was chosen to guaantee a fai compaison between the thee designs. The contast images so obtained fo each subject wee then enteed into thee sepaate one-sample t-test second-level goup analyses (one fo each expeimental design). All esults fo the andom-effects models wee coected fo multiple compaisons [FDR; Benjamini and Hochbeg, 1995] (P < 0.025; cluste size: 25 voxels). No diect statistical compaison was pefomed between the designs (see ROI analysis, below, fo compaisons). Assessing the diect influence of SBN using time coesponding images. In ode to futhe assess the influence of the scanne noise on intensity of auditoy activations, we selected 65 evenly spaced images out of the 516 images of the ERcont condition and convolved them 707
Gaab et al. with STsamp covaiate (hee efeed to as ER65). The selected scans coesponded in time with the fist 65 of the 66 images fo the STsamp condition, i.e., evey 16 s (Fig. 2D). Because the stimulus design and statistical powe of the ER65 and the STsamp condition was identical, we could diectly access the influence of the scanne noise on auditoy aeas using this ER65 design. The contast images wee then also enteed in a goup analysis (one-sample t- test). In ode to compae the STsamp and the ER65 design, we pefomed a paied t-test with the contast images. The esults ae pesented fo a theshold of P < 0.05, coected fo multiple compaisons. Time couse analysis fo STsamp and ER65. Because the gap between the end of the auditoy stimulation and the stat of the image acquisition fo STsamp (and ER65, espectively) was vaied ove 8 s, we wee able to pefom a time couse analysis even though only one set of images was acquied evey 17 s. This analysis enabled us to diectly compae the two designs ove 8 s (eight imaging time points) following the end of the auditoy stimulation to eveal possible time couse diffeences of the HRF esponse in auditoy as well as extatempoal aeas. Fo the BIG-CVA design, no time couse analysis could be pefomed due to the fixed inteval between auditoy stimulation and image acquisition. In a fixed-effects goup analysis, each of the eight ITPs fo each subject was modeled as a sepaate condition and a finite impulse esponse (FIR) basic set (ode/window length ¼ 1) as well as a high-pass filte (128 s) was applied. Contast images wee ceated fo each ITP as well as fou ITP clustes (Cluste1: ITP0-1; Cluste2: ITP2-3; Cluste3: ITP4-5; Cluste 4: ITP6-7). All esults in the time couse analysis ae coected fo multiple compaisons (P < 0.05; familywise eo (FWE) coected, cluste size ¼ 25). Movement paamete analysis In ode to compae design-elated movement fo each possible movement diection (x, y, z, pitch oll yaw; absolute movement fom fist image), we ceated a mean movement value fo each diection and each subject. The six movement paametes wee then compaed between the thee expeimental conditions using Fiedman and Wilcoxon tests. ROI Delineation Fou anatomically defined ROIs wee chosen fo this analysis using the softwae pogam WFU_pick Atlas and the implemented AAL atlas [Maldjian et al., 2003, 2004; Tzouio-Mazoye et al., 2002]. We chose sepaately fo the ight and left hemisphee the following pedefined ROIs: 1) the tansvese gyus (Heschl s gyus) coveing pimay auditoy cotex, and 2) the supeio tempoal gyus coveing highe-ode auditoy egions. Fou individual t-maps wee ceated fo each subject (BIG-CVA, ERcont, STsamp, ER65) using a single egession analysis (basic models) with the coesponding covaiates (see above). Fo the ITP analysis, a t-map fo each ITP (see detailed analysis desciption above) and each subject within the fixed-effects model was ceated fo ER65 and STsamp. Fo this analysis, standad eos wee obtained fom each subject s t-map (see below). ROIs wee applied to the t-map images fo each subject sepaately and egion-specific values wee aveaged ove all subjects fo each of the thee designs as well as the subset ER65. Then nonpaametic statistical tests (Fiedman and Wilcoxon tests) wee applied in ode to assess possible statistical diffeences between expeimental designs o hemisphees. Compaison with Theoy In any task design thee is an invese elationship between the aveage intestimulus inteval (ISI) and the numbe of time fames that can be acquied at constant total scan time. Fo a long ISI, such as in STsamp, fewe fames ae acquied with concomitantly educed degees of feedom (df) in the fixed effects modeling, with the opposite tue fo continuous scanning. Howeve, with longe aveage ISI the hemodynamic esponse ecoves moe fully to baseline condition, which povides a lage signal that in tun can offset the educed df. We calculated the expected BOLD contast-to-noise atio (CNR) fo each of the fou designs, based only on the hemodynamic esponse to the auditoy task and ignoing the influence of the SBN, i.e., nonlineaities due to masking o othe neuonal effects, but taking into account baseline shifts that might esult fom nonetun to equilibium of the HRF as well as T1 elaxation effects that vaied with the epetition time, TR. We hypothesized that if masking occued the measued data would have lowe elative signal intensities than that calculated with neual effects ignoed. Using the covaiate vectos C(n) obtained fom the expeimental design (Fig. 2) convolved with a hemodynamic esponse function [Glove, 1999], the BOLD contast B is popotional to the oot mean squae aveage of the powe in the covaiate [Papoulis, 1977]: ( B ¼ 1 X ) N 1=2 ½CðnÞ CŠ 2 N cðt1; TRÞ; ð1þ n 1 whee N is the numbe of time fames in the design, the oveba denotes aveage ove the time seies, and C is the NMR elaxation facto [Haacke, 1999] applicable when the Enst flip angle is used: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi c ¼ 1 e TR=T1 : 1 þ e TR=T1 ð2þ The Enst angle may not be optimal when inflow effects ae pesent [Lu and Golay, 2002]. Then the CNR is popotional to BHdf, so that: 708
Scanne Noise Influence on Auditoy Pocessing I Figue 3. Imaging esults fo the thee expeimental designs and the subset ER65 (P < 0.025, coected fo multiple compaisons, false discovey ate (FDR)). Each ow depicts one of the fou designs in 3D endeing (left 2 columns) and in axial slices (ight 3 columns) fo activations in Table I. [Colo figue can be viewed in the online issue, which is available at www.intescience.wiley.com.] CNR ¼ ( df ð1 e TR=T1 Þ X ) N 1=2 ½CðnÞ CŠ 2 Nð1 þ e TR=T1 Þ : ð3þ Using Eq. 3 and a T1 of 1350 ms fo gay matte, values of CNR wee computed fo the BIG-CVA, STsamp, ERcont, and ER65 designs using T1 ¼ 900 ms and df ¼ {139, 64, 256, 63}, espectively (accounting fo autocoelation due to the HRF). We compaed these calculated values with measued t-scoes (analogous to CNR) obtained fom the ROIs above as a gand aveage ove all subjects and left and ight Heschl s gyi and left and ight STG. Since the amplitude of the CNR in Eq. 3 has an implicit abitay constant multiplie, the calculations wee nomalized to the measued STsamp t-scoes to examine elative values. n¼1 RESULTS Behavioal Results All subjects pefomed above 83% coect fo all thee conditions. The mean pefomance fo ERcont was 96% coect, fo STsamp it was 99% coect, and fo BIG-CVA it was 92% coect. A Fiedman test showed significant diffeences between the thee expeimental conditions (w 2 ¼ 11.7, P ¼ 0.003). Post-hoc Wilcoxon tests evealed supeio pefomance fo the STsamp vs. ERcont condition (Z ¼ 2.5, P ¼ 0.01) and fo the STsamp vs. BIG-CVA condition (Z ¼ 2.6, P ¼ 0.009). Movement Paamete Analysis Only analysis fo the y-diection showed significant diffeences among the thee expeimental designs using a Fiedman test (w 2 ¼ 8.6; P < 0.05). A post-hoc analysis (Wilcoxon tests) showed less movement in the y-diection in the BIG-CVA (mean: 0.27 mm) vs. the ERcont (mean: 0.50 mm) condition (Z ¼ 1.98; P < 0.05). Since even the most significant diffeence was subvoxel and image coegistation was used duing all pepocessing, futhe statistics ae not epoted. Auditoy Setup Results Subjects chose on aveage an SPL amplitude of 61.6 (64.8) db (A) fo STsamp and BIG-CVA, and 72.8 (66.0) db (A) fo ERcont. The SPL fo the scanne backgound noise was measued as 116 db (A). The isolation of the headphone was not measued but was estimated to be 35 db. Imaging Results The andom-effects analyses wee coected fo multiple compaisons using the false discovey ate [e.g., Genovese et al., 2002]. We used a theshold of P < 0.05 fo the diect compaison between STsamp and the subset ER65. Fo the goup analyses assessing the main effect within each expeimental design we used a moe stingent theshold (P < 0.025) because the P < 0.05 theshold led to lage clustes that extended so much that it was difficult to discen auditoy egions. 709
Gaab et al. TABLE I. Activation table fo the thee expeimental designs and the subset ER65 MNI coodinates Region Extent x y z T CVA Left supeio tempoal gyus 865 58 8 2 15.50 Cingulate gyus 66 8 4 50 10.07 Right supeio tempoal gyus 744 56 28 16 12.89 Left caudate nucleus 73 22 8 16 12.34 Right caudate nucleus 26 18 2 16 9.03 Right dosal ceebellum 41 20 58 22 10.01 Right dosal ceebellum 64 18 68 24 8.62 Stsamp Left supeio tempoal gyus 781 60 26 12 10.61 Right supeio tempoal gyus 663 58 30 16 9.07 Cingulate gyus 294 2 2 42 8.45 Left caudate nucleus 223 24 8 14 7.75 Right caudate nucleus 176 24 24 14 7.39 Left infeio fontal gyus 98 36 26 8 7.20 Right dosal ceebellum 113 20 68 18 7.13 Left postcental gyus/infeio paietal aea 136 48 32 52 7.06 Thalamus (medial dosal nucleus) 30 10 22 10 6.10 ER65 Right supeio tempoal gyus 241 62 26 12 17.17 Left supeio tempoal gyus 56 51 22 6 11.07 Econt Cingulate gyus 81 10 16 36 16.62 Cingulate gyus 362 2 2 46 7.73 Right supeio tempoal gyus 782 66 10 4 14.32 Left supeio tempoal gyus 794 54 20 12 12.50 Left postcental gyus/infeio paietal aea 235 46 34 46 11.66 Right infeio fontal gyus 32 38 28 8 8.86 Left insula 28 38 14 14 8.07 Left caudate nucleus 65 18 8 8 7.32 Thalamus (medial dosal nucleus) 92 11 16 12 6.91 Left infeio fontal gyus 42 34 28 6 6.86 All esults ae epoted fo P < 0.025, coected fo multiple compaisons.er, event-elated; MNI, Monteal Neuological Institute; CVA, clusteed volume acquisition; Stsamp, spase tempoal sampling. The fixed-effects analysis assessing the Cluste ITP analysis was coected fo multiple compaisons using the family-wise eo ate [e.g., Nichols and Hayasaka, 2003]. Goup analysis Main effect within each expeimental design (Fig. 3, Table I). All thee designs yielded bilateal activations of supeio tempoal gyi including Heschl s gyus and the planum tempoale, the anteio cingulate, and the left caudate nucleus. The two silent designs (BIG-CVA and STsamp) also yielded activation in the ight dosal ceebellum. Only the BIG-CVA design led to activation of the ight caudate nucleus. Additional esults fo the STsamp and ERcont include the left postcental gyus/infeio paietal egion, the left infeio fontal gyus, and the thalamus. The ERcont design also evealed activations of the ight infeio fontal gyus and the left insula. Because of design limitations the BIG-CVA scan duation was 93% as long as that fo the ERcont and STsamp designs. This educed the statistical powe of the BIG-CVA activation a small amount. Howeve, we veified that no essential diffeence was obseved in the activation compaisons fo two subjects when the scan duations wee equalized by emoving data fom the ERcont and STsamp scans. Assessing the diect influence of SBN using time coesponding images. The ER65 design led to activations of bilateal supeio tempoal gyus, including Heschl s gyus and the planum tempoale (Fig. 3, Table I). Diect compaisons showed that STsamp yielded geate activation than ER65 in bilateal supeio tempoal gyus and a netwok of extatempoal bain egions including bilateal supamaginal gyus, ight pe-sma, bilateal infeio tempoal gyus, and tempoopaietal egions (Fig. 4). ROI analysis Hemispheic ROI diffeences. Thee wee significant highe mean t-values in left than ight supeio tempoal gyus fo all designs except fo ERcont (Fig. 5B, Table III). 710
Scanne Noise Influence on Auditoy Pocessing I The esults fo STsamp evealed additional activations of ight supamaginal gyus, left insula, and left thalamus, wheeas the esults fo ER65 additionally included the ight fusifom gyus (cluste 3: ITP4-5, Fig. 7C, Table II). No activations wee found fo the last cluste fo this theshold. Compaison with theoy Figue 4. Imaging esults (andom-effects analysis) fo the diect compaison between STsamp and ER65 (P < 0.05, coected fo multiple compaisons, FDR). [Colo figue can be viewed in the online issue, which is available at www.intescience.wiley.com.] Thee wee no significant hemispheic diffeences in Heschl s gyus (Fig. 5A, Table III). Intedesign ROI diffeences (Fig. 5, Table IV). Fo the ROIs in Heschl s gyus thee wee significant inceases in the mean t-values fo BIG-CVA in compaison to ERcont in the left hemisphee and to ER65 bilateally. Thee wee no significant diffeences when compaing BIG- CVA to STsamp (P < 0.79 fo left and P < 0.95 fo ight hemisphee). STsamp showed significantly highe t-values than ERcont and ER65 bilateally (Fig. 5A, Table IV). The analysis of the ROI coveing the supeio tempoal gyus evealed inceased mean t-values fo the left and ight hemisphee fo ERcont and STsamp in compaison to the ER65 bilateally. No diffeences wee found between ERcont, BIG-CVA, and STsamp, although simila tends to those obseved fo Heschl s gyus wee demonstated (see Fig. 5B, Table IV). Figue 8 shows the aveaged t-scoes fo activation in auditoy aeas compaed with CNR values calculated accoding to Eq. 1, nomalized with STsamp. The calculations pedict highe CNR fo both BIG-CVA and ERcont and geatly deceased CNR fo ER65 elative to STsamp, pimaily as a consequence of the numbe of independent samples in each, but also modulated by hemodynamic ecovey diffeences. Howeve, the measued t-scoes fo BIG-CVA ae compaable to STsamp and consideably lowe than pedicted. Futhemoe, the measued t-scoes fo ERcont ae 60% of that calculated and substantially less than STsamp even though appoximately 8 times as many time fames ae collected with the continuous design. Finally, the measued values fo ER65 ae significantly lowe than those measued fo STsamp in all ROIs (see ROI analysis, above), diectly demonstating the deleteious effects of masking by the SBN, since the two designs have exactly the same auditoy stimulation and numbe of time fames (statistical powe) but diffe only by the amount of scanne backgound noise. We veified that the calculation of signal in Eq. 2 using the Enst angle was easonable by measuing the SNR in Heschl s gyus fo one Imaging time point (ITP) analysis fo STsamp and ER65 ROI esults. Results yielded fo all fou ROIs and ITP0-5 highe mean t-values fo STsamp than ER65. Besides the oveall mean t-value diffeences, the time couses and peak ITP within each ROI diffe between the two designs (see Fig. 6A D). ITP imaging esults. Results fo cluste 1 evealed activation of supeio tempoal gyus and the cingulate gyus fo both designs and additional activation of the left pecental gyus fo STsamp in compaison to ER65 (cluste 1: ITP0-1; Fig. 7A, Table II). Results fo the second cluste evealed activation of bilateal supeio tempoal gyus, the left postcental gyus, and the cingulate gyus fo both designs. Results fo STsamp additionally yielded activation of left thalamus and left pecental gyus. The analysis fo ER65 evealed additional activation fo left infeio fontal egions (cluste 2: ITP2-3, Fig. 7B, Table II). Results fo the thid cluste evealed activation of bilateal supeio tempoal gyus and ight paahippocampal gyus. Figue 5. Results fo the ROI analysis fo the thee expeimental designs and the subset ER65: (A) Heschl s gyus (B) supeio tempoal gyus. [Colo figue can be viewed in the online issue, which is available at www.intescience.wiley.com.] 711
Gaab et al. TABLE II. Activation table fo cluste 1 3 fo STsamp and ER65 MNI coodinates Region Extent x y z T STsamp: ITP0 1 Left supeio tempoal gyus 2787 56 14 8 24.78 Right supeio tempoal gyus 2339 58 10 8 20.07 Cingulate gyus 188 4 6 46 9.03 Left pecental gyus 60 50 8 48 7.8 ITP2 3 Left supeio tempoal gyus 2554 56 14 8 20.72 Right supeio tempoal gyus 1949 64 18 8 20.56 Cingulate gyus 474 6 16 40 11.78 Pe/postcental gyus 474 42 24 52 7.87 Left thalamus 36 12 18 14 6.95 ITP4 5 Right supeio tempoal gyus 671 64 18 8 13.63 Left supeio tempoal gyus 1192 64 14 6 12.99 Right paahippocampal gyus 418 18 56 10 8.36 Right supamaginal gyus 64 62 48 26 7.17 Left thalamus 36 12 16 14 6.75 Left insula 28 40 16 4 6.16 ITP6 7 No egions ER65 ITP0 1 Left supeio tempoal gyus 1539 56 12 10 15.98 Right supeio tempoal gyus 1071 58 8 8 11.25 Cingulate gyus 200 4 12 40 9.55 ITP2 3 Left supeio tempoal gyus 1756 60 8 6 15.41 Right supeio tempoal gyus 1292 58 18 6 12.81 Cingulate gyus 397 4 18 38 9.77 Cingulate gyus 32 6 32 30 5.85 Left infeio fontal gyus 69 32 24 10 9.24 Postcental gyus 80 40 24 60 7.11 Postcental gyus 78 54 26 50 6.92 ITP4 5 Left supeio tempoal gyus 263 64 12 10 6.99 Right supeio tempoal gyus 63 64 16 4 6.99 Right paahippocampal gyus 53 26 44 12 5.90 Right fusifom gyus 34 26 62 10 5.89 ITP6 7 No egions All esults ae epoted fo P < 0.05, coected fo multiple compaisons. ER, event-elated; MNI, Monteal Neuological Institute; ITP, imaging time point; Stsamp, spase tempoal sampling. subject in both ERcont and STsamp and found a atio of 0.69, close to the 78% expected. In fact, if this measued value is used instead of the assumed calculation, the discepancy in Figue 8 between ERcont pedicted/measued and STsamp becomes even lage, futhe confiming the souce of dispaity as due to SBN. DISCUSSION This study examined the influence of thee diffeent expeimental designs with vaying degees of scanne backgound noise (SBN) on the functional anatomy, signal intensities, and time couses of activations in auditoy and extatempoal aeas. Behavioal pefomance was highe in the STsamp condition, indicating that the potentially deleteious effects of long intetial intevals on pefomance was moe than compensated fo by the SBN-fee pesentation of the auditoy mateial. The BIG-CVA design also pesented auditoy stimuli without SBN, but the inceased numbe of tials, the deceased time gap between tials, and suboptimal TR may have had led to deceased pefomance scoes. Concodantly, all thee designs (BIG-CVA, ERcont, STsamp) showed activation in a functional netwok that included supeio tempoal aeas, anteio cingulate, and caudate nucleus. Within the tempoal lobe, activation was equal bilateally in pimay auditoy cotex (Heschl s gyus), but left-latealized in seconday auditoy cotex of the supeio tempoal gyus. This is consistent with the view that hemispheic specialization is not appaent in ealy o pimay cotical 712
Scanne Noise Influence on Auditoy Pocessing I TABLE III. ROI analysis fo hemispheic diffeences ROI Design Hemispheic asymmety Z value P value Heschl s gyus Supeio tempoal gyus BIG-CVA Left ¼ Right 0.153 P ¼ 0.878 STsamp Left ¼ Right 0.474 P ¼ 0.635 ERcont Left ¼ Right 0.459 P ¼ 0.646 ER65 Left ¼ Right 0.357 P ¼ 0.721 BIG-CVA Left > Right 2.80 P < 0.01* STsamp Left > Right 2.39 P < 0.05* ERcont Left ¼ Right 1.83 P ¼ 0.06 ER65 Left > Right 2.09 P < 0.05* ROI, egion of inteest; ER, event-elated; BIG-CVA, behavio inteleaved gadient, clusteed volume acquisition; Stsamp, spase tempoal sampling. pocessing, but latealized (in this case left-latealized fo vebal mateial) in late o seconday cotical pocessing [e.g., Binde et al., 1997, 2000; Giaud and Pice, 2001; Poeppel et al., 2004]. Nevetheless, the mean t-values within auditoy and extatempoal aeas vaied among the thee designs, suggesting SBN-induced masking effects of the BOLD esponse, especially within Heschl s gyus, fo the designs with continuous SBN. Heschl s Gyus and SBN In ageement with seveal pevious studies that epoted a negative influence of SBN on signal intensities within pimay auditoy egions, ou ROI esults suggest SBNinduced masking of the auditoy cotical esponse within Heschl s gyus as a esult of inceased SBN (ERcont and ER65). This is futhe suppoted by ou compaison between TABLE IV. Intedesign ROI analysis ROI Statistical test Designs tested Hemisphee Design diffeences Z value/ chi 2 value P value Heschl s gyus Supeio tempoal gyus Fiedman All Left Yes Chi 2 ¼ 18.36 P < 0.00 All Right Yes Chi 2 ¼ 16.56 P < 0.01 Wilcoxon BIG-CVA/STsamp Left No Z ¼ 0.255 P < 0.79 BIG-CVA/STsamp Right No Z ¼ 0.051 P < 0.95 BIG-CVA/ERcont Left BIG-CVA > ERcont Z ¼ 2.09 P < 0.05 BIG-CVA/ERcont Right No Z ¼ 1.58 P < 0.11 BIG-CVA/ER65 Left BIG-CVA > ER65 Z ¼ 2.29 P < 0.05 BIG-CVA/ER65 Right BIG-CVA > ER65 Z ¼ 2.59 P < 0.01 STsamp/ERcont Left STsamp > ERcont Z ¼ 2.59 P < 0.01 STsamp/ERcont Right STsamp > ERcont Z ¼ 2.29 P < 0.05 STsamp/ER65 Left STsamp > ER65 Z ¼ 2.80 P < 0.01 STsamp/ER65 Right STsamp > ER65 Z ¼ 2.80 P < 0.01 ERcont/ER65 Left ERcont > ER65 Z ¼ 2.19 P < 0.05 ERcont/ER65 Right ERcont > ER65 Z ¼ 2.49 P < 0.05 Fiedman All Left Yes Chi 2 ¼ 11.64 P < 0.01 All Right Yes Chi 2 ¼ 13.44 P < 0.01 Wilcoxon BIG-CVA/STsamp Left No Z ¼ 0.561 P ¼ 0.57 BIG-CVA/STsamp Right No Z ¼ 0.878 P ¼ 0.87 BIG-CVA/ERcont Left No Z ¼ 0.408 P ¼ 0.68 BIG-CVA/ERcont Right No Z ¼ 0.968 P ¼ 0.33 BIG-CVA/ER65 Left No Z ¼ 1.784 P ¼ 0.74 BIG-CVA/ER65 Right No Z ¼ 1.682 P ¼ 0.93 STsamp/ERcont Left No Z ¼ 0.255 P ¼ 0.79 STsamp/ERcont Right No Z ¼ 0.663 P ¼ 0.50 STsamp/ER65 Left STsamp > ER65 Z ¼ 2.65 P < 0.01 STsamp/ER65 Right STsamp > ER65 Z ¼ 2.80 P < 0.01 ERcont/ER65 Left ERcont > ER65 Z ¼ 2.80 P < 0.01 ERcont/ER65 Right ERcont > ER65 Z ¼ 2.49 P < 0.05 ROI, egion of inteest; ER, event-elated; BIG-CVA, behavio inteleaved gadient, clusteed volume acquisition; Stsamp, spase tempoal sampling. 713
Gaab et al. SBN-induced masking of the BOLD esponse. Howeve, seveal othe factos might have influenced this compaison, such as attentional demands duing the expeiment, diffeences in the time couse of the HRF fo the designs, o delayed activation within some egions. The pesent study showed no significant diffeences between the BIG-CVA and the STsamp design fo both ROIs coveing Heschl s gyus, despite the geate numbe of acquied images as well as the quantity of wod sequences pesented fo the BIG-CVA design (106 expeimental and 55 contol fo BIG-CVA vs. 48 expeimental and 16 contol fo STsamp). This suggests that the signal gains fom STsamp measuement makes up fo the 50% eduction of obsevations. The signal gain in the STsamp design despite the educed numbe of obsevations suggests SBN-induced Figue 6. Results fo the ROI time point analysis fo STsamp and ER65 fo (A) left Heschl s gyus (B) ight Heschl s gyus (C) left supeio tempoal gyus (D) ight supeio tempoal gyus. [Colo figue can be viewed in the online issue, which is available at www. intescience.wiley.com.] the theoetical pediction of the signal intensity and the actual measued values. The design with continuous SBN, ERcont, esulted in deceased values fo measued compaed to pedicted signal intensities, which futhe suggests Figue 7. Imaging esults fo the cluste analysis fo STsamp and ER65 (P < 0.05, coected fo multiple compaisons (FWE)) (A) cluste 1 (ITP0-1), (B) cluste 2 (ITP2-3), (C) cluste 3 (ITP4-5). Cluste 4 did not eveal significant voxels fo this theshold. 3D-endeed images (left) and axial slices (ight) ae selected based on egions activated in expeimental designs as shown in Table II. [Colo figue can be viewed in the online issue, which is available at www.intescience.wiley.com.] 714
Scanne Noise Influence on Auditoy Pocessing I Figue 8. Compaison of activation measued in auditoy ROIs with CNR calculated fo each design, based on esponse to auditoy stimuli nomalized to the STsamp design. ER65 shows much lowe values than pedicted and significantly lowe values (see ROI Analysis) than STsamp due to masking and elevated baseline effects. [Colo figue can be viewed in the online issue, which is available at www.intescience.wiley.com.] masking effects of the BOLD esponse in the BIG-CVA design. This might be due to the lack of jitteing and/o the TR of 6 s, which does not allow the SBN-induced activation to ecove befoe the next acquisition. Based on the liteatue on the shape of the HRF following an auditoy stimuli [Glove et al., 1999] o SBN in paticula [Hall et al., 2000], the SBN-induced HRF should peak aound 4 6 s afte the auditoy stimulus. Duing a BIG-CVA design with a TR of 6 s, the HRF induced by image acquisition (SBN) #X would peak duing image acquisition (SBN) #X+1. Futhemoe, the measued t-scoes fo BIG-CVA ae consideably lowe than pedicted. This is consistent with the explanation that the signal is educed with the nonjitteed design of BIG-CVA because the HRF peaks afte the scan and is thus not sampled at its lagest, and is educed futhe by SBN-induced masking of the BOLD esponse and elevated baseline fom the emnant HRF of the pevious scan s SBN. Nevetheless, one othe potential souce of diffeence between the two designs might be the natue of the tasks. The BIG-CVA design uses a TR of 6 s, which leads to continuous auditoy stimulation as well as a minimum nonstimulus gap of 2 s if expeimental tials follow each othe (in compaison to 4 s o geate fo the othe designs). A BIG- CVA design with a longe TR might have educed the HRF effects due to the SBN and at the same time inceased the gap between stimulations, but then a spase tempoal sampling design with a jitteed delay (and time couse infomation) seems to be the moe appopiate design. By extacting a subset of images taken out of ERcont (ER65) and compaing diectly to STsamp, we ae able to assess the influence of SBN while keeping the expeimental auditoy stimulation as well as the quantity of obsevations constant. One diffeence between these two designs was an inceased amplitude of the auditoy stimulation fo the ER65 design. Howeve, since pevious studies have shown inceased activation and spatial extent with inceased stimulus amplitude [Jancke et al., 1998; Lasota et al., 2003], this might have led to a slight advantage fo the ER65 design. The compaison showed geate mean t-values fo STsamp in Heschl s gyus and STG in both hemisphees, which indicates stong SBN-induced masking of the BOLD esponse. Nevetheless, it emains unclea whethe this masking is a esult of the nonlineaity of the human auditoy cotex o an effect of aised signal intensities within the contol baseline condition [see Talavage and Edmiste, 2004]. Futhe expeiments ae necessay in ode to answe this question [see Gaab et al., 2006]. Extatempoal Regions and SBN Most pevious studies examined only pimay auditoy aeas, but the influence of SBN might also be eflected in nonauditoy, extatempoal aeas. BIG-CVA, STsamp, and ERcont evealed additional activation of nonauditoy aeas, including the anteio pat of the cingulate gyus. This aea has been known to be involved in attention geneally [e.g., Loose et al., 2003] and auditoy selective attention paticulaly [Sevostianov et al., 2002]. Since all thee expeimental designs showed activation of the anteio cingulate, an SBN-induced task diffeence o attention modulation within this aea between ERcont and the silent designs seems unlikely. Nevetheless, we contasted the wod sequence task with a no-stimulus condition, which is usually not the case in fmri studies. The question whethe SBN may influence attentional demands in expeimental and contol conditions diffeently cannot be answeed fom ou study and futhe studies should addess this issue. Additionally, all thee designs evealed activation of the caudate nucleus. Seveal studies have demonstated caudate activations in woking memoy tasks [Lewis et al., 2004; Owen et al., 1996; Postle and D Esposito 1999a,b; Speck et al., 2003]. Some of these studies suggested that the caudate nuclei may be moe involved in spatial than nonspatial woking memoy, suggesting a ole in the integation of spatially coded mnemonic infomation with moto pepaation to guide behavio. Howeve, Lewis et al. [2004] demonstated activation of the caudate nucleus in a nonspatial woking memoy task and the citical ole of the caudate nuclei in cognitive pocesses has also been undelined by neuodegeneative conditions such as Huntington s disease. Futhemoe, the two silent designs (BIG-CVA and STsamp) showed additional activation of the ight dosal ceebellum [Aea V+VI; Schmahmann et al., 1999] which was not pesent in the ERcont condition fo the given theshold. Seveal studies have emphasized the ole of the ceebellum in auditoy pocessing [e.g., Gaab et al., 2003; Holocomb et al., 1998; Jancke et al., 2000; Mathiak et al., 2004; Silvei et al., 1998], auditoy vebal memoy function [e.g., Andeasen et al., 1995; Desmond et al., 1997; Gasby et al., 1993; Kischen et al., 1995; Schumache et al., 1996], as well speech peception o silent speech [Ackemann et al., 2004; Mathiak et al., 2002; Silvei et al., 1998]. The 715