BR A IN RE S EA RCH 1 3 08 ( 20 1 0 ) 6 8 –78 available at www.sciencedirect.com www.elsevier.com/locate/brainres Research Report Motor area localization using fMRI-constrained cortical current density reconstruction of movement-related cortical potentials, a comparison with fMRI and TMS mapping Alberto Inuggi a , Massimo Filippi b , Raffaella Chieffo a , Federica Agosta b , Maria A. Rocca b , Javier J. González-Rosa a , Marco Cursi a , Giancarlo Comi a , Letizia Leocani a,⁎ a Department of Neurology, Neurophysiology and Neurorehabilitation, Experimental Neurology Institute, IRCCS San Raffaele, Milan, Italy Neuroimaging Research Unit, IRCCS San Raffaele, Milan, Italy b A R T I C LE I N FO AB S T R A C T Article history: The localization of human hand primary motor area (M1) has been the object of several Accepted 16 October 2009 studies during the last decades. EEG source analysis, functional magnetic resonance Available online 22 October 2009 imaging (fMRI) and focal transcranial magnetic stimulation (TMS) are non-invasive methods for localizing M1 with good accuracy compared to direct electrocorticography (ECoG) results. Keywords: EEG sources were reconstructed with Cortical Current Density (CCD) method, allowing to EEG source analysis evaluate simultaneous and distributed patterns of activation and to increase accuracy by Multimodal integration constraining on information derived from fMRI (fMRI-CCD). The aim of this study was to fMRI constrained CCD compare the M1 contribution of movement-related cortical potentials (MRCP) with TMS and fMRI results and to test the effect of constraints strength, algorithm norm and localization methods over CCD reconstruction. Seven right-handed healthy subjects underwent 64channel EEG recording of MRCP to right thumb movement, focal TMS mapping of the right abductor pollicis brevis muscle and fMRI during right hand movement. We found fMRI activations, EEG sources and TMS mapping corresponding to the anatomical landmark of the hand area in all subjects with fMRI and TMS center-of-gravity and in almost all subjects using fMRI-CCD with moderate constraint. A significant improvement was found using fMRI-CCD compared to CCD alone. This study confirms the usefulness of multimodal integration of fMRI, EEG and TMS in localizing M1 and the possibility to increase EEG spatial resolution using fMRI information. © 2009 Elsevier B.V. All rights reserved. ⁎ Corresponding author. Department of Clinical Neurophysiology, IRCCS San Raffaele, Via Olgettina 60, 20132 Milan, Italy. Fax: +39 2 2643 3085. E-mail address: leocani.letizia@hsr.it (L. Leocani). Abbreviations: BOLD, blood oxygenation level dependent; CCD, cortical current density; CCD-POSnorm,K,method, Anatomical position of CCD sources calculated with a specific norm, M1 reconstruction method and fMRI over-weighting factor strength K; CCD-fMRInorm,K,method, Distance between CCD source and fMRI maxima calculated with a specific norm, M1 reconstruction method and K fMRI over-weighting factor strength; COG, center of gravity in one single latency; COGT, average center of gravity in one period; CSF, cerebrospinal fluid; EEG, electroencephalography; fMRI, functional magnetic resonance imaging; fMRI-CCD, fMRI constrained CCD; HS, hot spot; K, strength of fMRI over-weighting factor over fMRI-CCD; M1, primary motor area; MEP, motor evoked potential; METHOD, method to calculate M1 position (HS, COG, COGT); MRCP, movement related cortical potentials; NORM, type of CCD algorithm (L1, L2); PMd, dorso-lateral premotor cortex; SMA, supplementary motor area; SNR, signal-to-noise ratio; TMS, transcranial magnetic stimulation 0006-8993/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.brainres.2009.10.042 BR A IN RE S E A RCH 1 3 08 ( 20 1 0 ) 6 8 –7 8 1. Introduction The localization of human hand primary sensory and motor areas (SM1) has been object of several studies during the last decades. Preoperative direct cortical electrical stimulation 69 (Yousry et al., 1995) and electrocorticography with recording of movement-related cortical potentials (MRCP) (Ikeda et al., 1996) localized the primary human hand motor area (M1) in BA4 on the precentral gyrus. Its position was later associated to the precentral knob region (Yousry et al., 1997). Several non- Fig. 1 – Multimodal localization of M1. Each column represents a different subject, each row a different method. (fMRI) axial slices of fMRI activation to fingers movement at fMRI maximum depth superimposed over anatomical MRI. In all subsequent rows, fMRI contours are overlaid over cortex surface (black line), intensities, either of current density and MEP values, are expressed in a color scale ranging from red (lowest values) to white (highest values). (TMS) results from MEP mapping over CSF surface. TMS COG and HS are represented with blue square and triangle respectively, fMRI maxima with a black circle. When the TMS COG and HS distance is below 3 mm only COG is shown. TMS points are projected according to the procedure described in Fig. 4. (remaining rows) CCD with both norms (L1 and L2) and two representative K: 1.0 shows unconstrained CCD and 1.4 shows sources with the K producing the lowest distance from fMRI. In L2 CCD, sources carrying currents below 50% of the largest current (full width at half maximum, FWHM) are clipped (Fuchs et al., 1999). 70 BR A IN RE S EA RCH 1 3 08 ( 20 1 0 ) 6 8 –78 invasive methods have been developed with increasing accuracy. Among these, functional magnetic resonance imaging (fMRI) measures the blood oxygenation level dependent (BOLD) contrast related to the increased blood demand in the vicinity of neuronal activity (Logothetis, 2003). fMRI of voluntary movement localized M1 area with differences ranging from 3 to 10 mm compared with intra-operative electrical stimulation (Yousry et al., 1995). The analysis of motor evoked potentials (MEPs) to focal transcranial magnetic stimulation (TMS) applied to different scalp positions over the motor cortex allows to map the scalp representation of a given muscle (Wassermann et al., 1992; Rossini et al., 1994). Cortical projection of these maps produced average localization discrepancies from 2 to 14 mm compared with fMRI activation maps (Terao et al., 1998; Herwig et al., 2002; Lotze et al., 2003; Neggers et al., 2004). High-resolution electroencephalography (EEG) has been used to investigate the spatio-temporal pattern of cortical activity by reconstructing the generators of scalp recorded potentials using dipole and distributed source models such as cortical current density (CCD) reconstruction. Several approaches have been developed to increase the spatial resolution of EEG (Graves de Peralta Menendez and Gonzalez Andino, 1998; Fuchs et al., 1999; Wagner et al., 2001a, b) that, however, remains quite poor in comparison to fMRI. EEG source reconstruction accuracy can be improved with a spatial constraints approach (Kiebel and Friston, 2004), using fMRI activated regions of interest (ROI) as priors in the solution of the inverse problem (Babiloni et al., 2000; Wagner et al., 2001b; Dale and Halgren, 2001; Babiloni et al., 2003; Liu and He, 2008). fMRI-CCD methods, to solve the inverse problem, utilize anatomo-functional correlates of cortical activity instead of arbitrary mathematical hypotheses. The analysis of MRCP (Deecke and Lang, 1996; Wildgruber et al., 1997; Ball et al., 1999; Cui et al., 1999; Toma et al., 2002, Inuggi et al., 2009) is the most investigated electrophysiological protocol for investigating M1 activity particularly close to movement onset. In this paper, for the first time, the three non-invasive methodologies (fMRI, TMS and EEG) will be applied to the same subjects in order to investigate the reciprocal relationships among these methods and the precentral knob position. A particular focusing will be given to the issue of MRCP generators reconstruction through the CCD model, providing the effects of fMRI constraints over its accuracy in real rather than simulated data as previously performed. 2. Results 2.1. fMRI All subjects showed activations in several areas, which are part of the “classical” sensorimotor network, including the primary sensorimotor cortex (SM1), bilaterally, the supplementary motor area (SMA) and the ipsilateral cerebellum. One subject also showed an additional activation in the contralateral secondary somatosensory cortex. The fMRI-maximum (ROI showing the highest T-value) was always precentral inside the ‘precentral knob’ region where the hand area is located (Yousry et al., 1997), 8 ± 5 mm deep into the precentral gyrus. In Fig. 1, fMRI maxima (row L1 1.0) and contours have been overlaid over rendered cortex and compared with TMSHS, TMSCOG and MRCP sources. No mirror movements could be observed. Table 1 – Anatomical localization of EEG sources and TMS measures. Subject Method Norm TMS L2 L1 A B C D E F G HS COG COGT HS COG COGT HS COG COGT HS COG COGT HS COG COGT HS COG COGT HS COG COGT K=1 K = 1.1 K = 1.4 K=3 K=1 K = 1.1 K = 1.4 K=3 BA6⁎ BA6⁎ BA6⁎ BA6⁎ BA6⁎ BA6⁎ BA4⁎ BA4 BA4 BA1⁎ BA1 BA1 PK BA4 BA4 BA3b BA3b BA3b BA3b BA3b BA3b BA4⁎ BA4 BA4 BA3b BA6⁎ BA6⁎ BA4⁎ BA4 BA4 BA1⁎ BA1 BA1 BA3b BA4 BA4 BA3b BA3b BA3b BA3b BA3b BA3b BA4 BA4 PK PK BA4 BA4 PK PK PK PK BA1 BA1 PK BA4 BA4 PK PK PK BA4 BA4 BA4 BA4 BA4 PK PK BA4 BA4 PK PK PK PK BA1 BA1 PK BA4 BA4 PK BA3b PK PK PK PK BA4⁎ BA3b BA3b BA6⁎ BA4 BA4 BA4⁎ BA4 PK BA4⁎ BA4 BA4 PK PK PK BA6⁎ BA6⁎ BA6⁎ BA3b BA4 BA4 BA4 BA3b BA3b PK BA4 BA4 PK BA4 PK BA4 BA4 BA4 PK PK PK BA6⁎ BA6⁎ BA6⁎ BA4 BA3b BA4 PK BA4 PK PK BA4 BA4 PK PK PK PK PK PK PK PK PK PK PK PK BA4 BA4 PK BA3b BA4 PK PK BA3b BA4 BA3b PK PK PK PK PK PK PK PK PK PK PK BA4 BA4 BA3b PK PK – BA6 PK – BA1 PK – PK PK – BA3b PK – BA6 PK – PK PK – Anatomical positions of EEG sources inside the confidence region. PK indicates that source was located in the precentral knob. EEG sources not included inside fMRI bounds are represented with the symbol ‘⁎’. BR A IN RE S E A RCH 1 3 08 ( 20 1 0 ) 6 8 –7 8 2.2. EEG Left EPB EMG analysis showed that no mirror movements were performed by any subject. The independent component analysis revealed three components with signal-to-noise ratio (SNR) values of 3.2 ± 0.9 , 1.2 ± 0.4, 0.6 ± 0.3. The first was located over mesial areas, the second located over primary sensorymotor cortex (SM1), the third pattern varied and was not used in all subjects as its SNR resulted below noise threshold (SNR < 1) during the whole interval. CCD analysis of the EEG signal reconstructed a distributed activation pattern mainly localized in contralateral primary sensorimotor cortex (M1 and S1) and SMA, but also embracing ipsilateral M1 and bilateral dorso-lateral premotor cortex. Applying fMRI constraints, activations mostly concentrated in contralateral M1 and SMA only. In Fig. 1 MRCP-CCD reconstructions with two norms and two overweighting factors are shown. CCD sources anatomical positions are summarized in Table 1 and plotted in Fig. 2. The best reconstruction methods resulted the HS with K = 3 for L1 and COGT and HS with K = 1.4 for L2 norm respectively (one M1 reconstruction in BA4 and six into the 71 precentral knob). Good performance was also achieved for L1 using the HS with K = 1.4 (two M1 reconstructions in BA4, five into PK). Significant effects of the three factors (norm, K, method) and their interactions over anatomical localization of CCD sources respect to precentral knob (CCD-POSnorm,K,method) are shown in Table 2. Concerning the interaction between norm and K strength, cortical activity resulted closer to precentral knob using K = 1.4 and K = 3 compared to K = 1.0 with both norms and using L2 compared to L1 norm with all K but K = 3. Concerning the method for M1 localization, HS performed significantly better than COG and COGT using L1 norm. Although no interaction between method and L2 norm emerged, HS resulted the only method for which L2 did not perform better than L1 in unconstrained CCD. 2.3. TMS On average, APB responses over 50 μV have been obtained in 22 ± 5.5 stimulated points. In Fig. 1 (row TMS), TMS-derived CSF maps and their corresponding measures positions (TMSCOG and TMSHS), projected orthogonally to scalp surface until CSF Fig. 2 – Distances of CCD sources from fMRI and precentral knob. Effect of norm (L1 and L2), fMRI constraints strength (K) and method (HS, COG, COGT) on CCD Euclidean distance, in mm, from fMRI maxima (left) and anatomical position respect to precentral knob (right). L1 and L2 norms are shown in left and right columns, respectively, and the different localization method (COGT, COG and HS) in upper, middle and lower row, respectively. Results are plotted with different K strengths (on the abscissas) and each subject is represented by a different symbol. 72 BR A IN RE S EA RCH 1 3 08 ( 20 1 0 ) 6 8 –78 Table 2 – Statistical analysis. Effect Distance from fMRI maxima K Norm Method norm⁎K K = 1.4, 3 < K = 1 L2 < L 1 norm⁎ method method⁎ k norm⁎K⁎ method Anatomical position K = 1.4, 3 < K = 1 L2 < L1 HS, COGT < COG all norms : K = 1.4, 3 < K = 1 K = 1, 1.1, 1.4 : L2 < L1 L1 : HS < COG, COGT all norms : K = 1.4, 3 < K = 1 all K : L2 < L1 L1 : HS, COGT << COG L2 : COG, COGT << HS all methods : L2 < L1 all methods : K = 1.4, 3 < K = 1 COG , COGT : L2 < L1 HS : K = 1.4 < K = 1, COG , COGT : K = 1.4, 3 < K = 1 K = 1.4 : HS < COG all methods (L1 ): K = 1.4, 3 < K = 1, all methods (L2 ): K = 1.4 < K = 1, COG, COGT (K = 1) : L2 < L1, K = 3 & L1: HS < COG all methods (L1) : K = 1.4, 3 < K = 1, HS, COGT (L2) : K = 1.4, 3 < K = 1, all methods (K = 1, 1.1) : L2 < L1, COG (K = 1.4, 3) : L2 < L1 Significant effect of norm, K and method on CCD distance from fMRI maxima (left) and CCD anatomical position respect to precentral knob (right), investigated with non-parametric related samples Wilcoxon signed-rank (for norm) and Friedman's several-related (for K and method) tests. Post-hoc analysis, performed with Wilcoxon signed-rank test (two related samples test, p < 0.05), is displayed. Symbol “A, B < C” indicates that measure in both conditions A and B are significantly lower than C one. surface, were overlaid over rendered cortex. The presence of both TMSCOG and TMSHS inside the following areas, precentral gyrus, precentral knob region and fMRI bounds, is summarized in Table 1 and shows that TMSCOG resulted significantly closer to precentral knob than TMS HS (TMS-POS COG = 0, TMSPOSHS = 1.14, P = 0.046). CCDL1,1.4,HS (Z = −2.366, p = 0.018), as shown in Fig. 3. Ventraldorsal differences among fMRI and CCD results showed that CCDL1,1.4,HS activation resulted more ventral than fMRI maxima. 3. 2.4. Euclidean distance of CCD sources from fMRI maxima, plotted in Fig. 2, resulted significantly affected by K and norm factors (Table 1). The post-hoc analysis showed that it resulted significantly lower in both K = 1.4 and K = 3 compared to K = 1.0 and significantly lower in L2 than L1 norm. Their interactions (norm⁎K) showed that K factor highly influenced both norms and that CCD reconstruction with L2 were significantly closer to fMRI than L1 for all values of K. Interactions between method and norm showed that COG and HS represent the worst M1 estimator for L1 and L2 norm respectively. Norm⁎K⁎method analyses confirmed the higher accuracy of L2 compared to L1, regardless of the method used, with K = 1 and K = 1.1. With K = 1.4 and K = 3 instead, such improvement occurred considering COG only. No significant differences emerged instead between TMS measures distance from fMRI maxima (TMSCOG = 8.5 ± 3 mm, TMSHS = 10.7 ± 11 mm). Comparing TMS measures with EEG ones, choosing within the latter those closer to fMRI and precentral knob with each both norm (CCD-fMRIL1,1.4,HS, CCD-fMRIL2,3,COG, CCDPOSL1,3,HS and CCD-POSL2,1.4,COGT), no significant differences emerged. 2.5. Discussion EEG and TMS distances from fMRI Coordinate differences The analysis of antero-posterior and medial-lateral position differences among TMS, fMRI and EEG results showed that fMRI maxima was significantly posterior than TMSHS (Z = −2.201, p = 0.028) and nearly significant posterior than TMSCOG (Z = − 1.859, p = 0.061) and that CCDL2,1.4,COG was posterior than The present study investigated two different aspects of hand M1 localization. First, the comparison of different noninvasive techniques that already proved to localize M1 with good precision, but that were never applied to the same subjects. Second, EEG source analysis was performed with fMRI constraints of different strength, with two different algorithm norms and with three source definition methods in order to investigate their effect, particularly those concerning fMRI constraints, over real rather than simulated data. The three non-invasive techniques used in the present study provided comparable results in localizing M1, confirming previous studies performing pairwise comparisons (Yousry et al., 1995; Terao et al., 1998; Ball et al., 1999; Hlustík et al., 2001; Toma et al., 2002; Herwig et al., 2002; Lotze et al., 2003; Neggers et al., 2004). Since we focused on anatomical M1 activation only, neither considering its temporal evolution nor the activity within secondary motor areas, differences between our metronome-paced block-design fMRI and self-initiated movement related EEG tasks should have been minor and not involving the position of the M1 cortical region generating each movement. While homologous mirror movements could be definitely excluded from EEG analysis, the fMRI paradigms ones could have instead escaped our visual monitoring and be inserted in the analysis, as in all previous studies comparing fMRI with EEG and TMS. Actually, mirror movements presence is considered a rare event in healthy adults (Cincotta and Ziemann, 2008), particularly when using their dominant hand (Leocani et al., 2000). Concerning the localization of the involved cortical areas, the assessment of different muscles (four fingers for BR A IN RE S E A RCH 1 3 08 ( 20 1 0 ) 6 8 –7 8 Fig. 3 – Position differences. Antero-posterior positions (in millimeters) of M1 localization reconstructed with, starting from the left, fMRI, TMSCOG, TMSHS, CCDL1,1.4,HS and CCDL2,1.4,COG .The asterisk (*) represents a significant difference between two measures, the plus sign (+) a trend (p = 0.061). Each subject is represented by a different symbol. fMRI and thumb extension/abduction for EEG/TMS respectively) to locate hand M1 needs some comment. Although M1 is known to be somatotopically organized (Penfield and Rasmussen, 1950), intracortical microstimulation in nonhuman primates confirmed such spatial segregation at the level of major-subdivision (arm, leg, trunk and head) but not for the single muscles areas, where a diffused overlapping could be observed (Asanuma, 1989; Humphrey, 1986; Lemon, 1990). The degree of overlapping further increases when analysed with fMRI (Sanes and Donoghue, 1997; Kleinschmidt et al., 1997) where COG distance between thumb and little finger were 2.5 ± 2 mm and the extent activation for each muscle was 46 ± 10 mm (Hlustik et al., 2001). These differences are below both EEG and TMS spatial resolution that cannot hence affordably discriminate between a finger and the thumb movement. EEG signal is in fact smeared by the non-homogeneous differences in electrical conductivities between skull and spinal fluid (Shibasaki, 2008) while the magnetic field distribution of eight-shaped TMS coil is no less than 1 cm wide (Bohning et al., 2001) and the corresponding induced electric field is even larger. Concerning fMRI and EEG comparability, previous studies investigating their respective physiologic origin (Logothetis, 2003; Kim et al., 2004) stressed that, along with a good correlation between BOLD loci and neuronal activity, the locations of neurons and involved vessels are expected to be different. Therefore, a perfect overlapping between fMRI maxima and EEG sources could not represent a CCD algorithm success because the actual active neuronal population generating the BOLD signal might not reside in its maxima and, at worst, locate around fMRI activations borders or even outside them. In order to also overweight such areas, fMRI regions were enlarged by 5 mm, before inserting them into the sources space, reducing de facto the possibility of having erroneously biased CCD to primary motor areas not specific for thumb activation. According to 73 all these considerations, differences among involved muscles and paradigms did not invalidate the correct localization of hand motor area into the precentral knob. The effect of fMRI constraints on CCD accuracy is one of the main topic of interest of the present paper. In fMRI-CCD, by rising the over-weighting factor strength (K), sources proportionally concentrate in fMRI constraints locations but a general agreement does not exist about the correct value of K that increases accuracy in over weighted ROIs without preventing to detect activations in non-over-weighted areas. Two values were considered to represent a good compromise: K = 1.4 (Wagner et al., 2001b) and K = 3 (Babiloni et al., 2003). Considering CCD distance from both fMRI maximum and precentral knob, a significant reduction was observed with both these two values compared to unconstrained CCD with both norms. The resulting CCD-fMRI distance (around 911 mm) corresponded to those obtained in a previous study comparing self paced movement CCD with fMRI (Ball et al., 1999). Notably, the fMRI constrained CCD did not simply shift the M1 localization inside the fMRI contours but also in their precentral part and, six out of seven times (considering L1 HS and L2 COGT) into the precentral knob region. While the activations shift inside fMRI bounds was expected, both the precentral and into-the-knob localization were not. In fact, since fMRI activations extended in both pre and postcentral regions and their respective loci were equally over-weighted, the algorithm did not have any further a priori information to localize the precentral knob. Such improvement was observed with K = 1.4, but did not further increase at K = 3. In addition, although not statistically, the anatomical score worsened using K = 3 with L2 norm, hence suggesting that too powerful over-weightings should be applied with caution. In a previous simulation study (Babiloni et al., 2003) in fact, the lowest reconstruction error was found with the lowest K value tested. Concerning the effect of algorithm norm, L2 norm showed a higher accuracy compared to L1 presumably for the latter high sensitivity to SNR, that in the present study was low (5.34) and only partly (36%, corresponding to a 1.92 SNR) generated by the lateral component associated to M1 activity. Most of SNR was associated instead to the mesial component, consistently with the higher SMA intensity, at the latencies investigated, resulted in previous self-paced movement studies (Ball et al., 1999; Toma et al., 2002). The low SNR of the lateral component could be also the origin of either the presence of unconstrained reconstructions outside fMRI bounds and the anatomical score high variability among subjects observed within each reconstruction. Particular for the latter phenomena, where BA4/BA6 classifications differed by only 1–2 mm from precentral knob one, a much lower distance than EEG spatial resolution. Noteworthy, the poor accuracy of unconstrained CCD could be highly increased using fMRI constraints. The M1 definition method had minor effect over CCD accuracy but consistent with the different nature of the two norms. With the focal and sparse L1 norm, HS resulted the best method to locate M1 while, using the smoothly changing distributed source activity of L2, COGT and COG performed better than HS. Coherently, source position reconstructed by L2 COG was significantly posterior to L1 HS presumably for some postcentral activity that was also averaged into L2 COG calculation. 74 BR A IN RE S EA RCH 1 3 08 ( 20 1 0 ) 6 8 –78 In the present paper, the lack of a frameless stereotactic device able to assess the real coil orientation could have limited our results, as we could not confirm that TMS pulses were actually delivered orthogonally to scalp surface, as hypothesized while projecting COG/HS according to that direction. Nevertheless, our COG distance from fMRI maxima (8.5 ± 3 mm) was consistent with 13 and 4.6 mm found in two previous TMS-PET studies (Wassermann et al., 1996; Classen et al., 1998) and with distances ranging from 2 to 14 mm in previous TMS-fMRI studies using conventional (Terao et al., 1998; Lotze et al., 2003) and neuro-navigated TMS (Herwig et al., 2002; Neggers et al., 2004). TMS measures resulted anterior than fMRI one consistently with previous study (Herwig et al., 2002; Lotze et al., 2003). Moreover, COG resulted located over the precentral knob and fMRI activation bounds in all cases and confirmed to be a better M1 estimator than HS as a result of evaluating their anatomical position differences. The variability of each single MEP, influenced by both physiological (Kiers et al., 1993; Darling et al., 2006; Rosler et al., 2008) and methodological factors, should mostly affect HS. COG calculation process instead, through its weighted averaging across several locations, might partly reduce such variability (Classen et al., 1998; Herwig et al., 2002; Neggers et al., 2004; Sparing et al., 2008) due to its good stability versus non-systematic MEP amplitude variations (Thickbroom et al., 1999). Considering that MEP could be evoked from postcentral and premotor regions also, the 1- to 2-cm resolution of the 8-shaped coil could be confirmed (Bohning et al., 2001) and may indicate that an M1 localization consistent with such resolution could be achieved with our method. On the contrary, a neuronavigator device proved to be very useful working with either more focal coils and lower stimulation intensities, in longitudinal studies that need to minimize errors in coil repositioning and to speed up stimulation procedure allowing to localize the area of interest without searching for the “hot spot.” In conclusion, the multimodal approach used in the present study demonstrated a strong relationship among fMRI, TMS and EEG results in locating human hand M1 as the three non-invasive techniques correctly located activation into the precentral knob hand region in all subjects, considering fMRI and TMS COG, and in most of them using fMR-CCD with moderate over-weighting (K = 1.4 and K = 3) and HS for L1 and COG and COGT for L2. EEG source analysis significantly improved when fMRI constraints were applied and, even if the over-weighted regions extended in nearby precentral and postcentral regions, most of the solutions concentrated into the precentral knob area confirming that the spatial resolution of CCD algorithm can be increased by fMRI information, also in case of low SNR data. In the specific case of M1 localization, the fMRI constraints could have been substituted by a manual segmentation of the precentral knob, but the main aim of this study was to test their effect over CCD and compare it with TMS and anatomy, in order to validate this methodology and support its usefulness in other paradigms. In clinical settings the non-invasive EEG localization of M1 may be useful not only for the study of brain reorganization and plasticity but also to investigate those patient that are not eligible for tests involving high magnetic fields like those produced by TMS and MRI devices. Therefore, a validated method for noninvasive M1 localization could reveal to be very useful in the functional study of motor cortex in physiological and pathological conditions. 4. Experimental procedures Seven right-handed healthy subjects (5 F, 2 M; age 26–58 years, score 10/10 at Edinburgh scale, Oldfield, 1971) first underwent EEG to self-paced movement of the right thumb, then fMRI to right hand movement and anatomical MRI and finally TMS mapping of the right abductor pollicis brevis (APB). Subjects gave their informed consent to participate in the study, which was approved by the local ethics committee. 4.1. EEG recording EEG was recorded with 64 Ag-AgCl electrodes, positioned according to the 10-20 international system, with two 32channel EEG amplifier (Synamps Amplifiers, Neuroscan Inc., Herndon, VA), using the right earlobe as reference. Electrode impedance was kept below 5 K Ohm. The 3D coordinates of the scalp electrodes and of anatomical landmarks (NAS, PAR, PAL) were obtained using the Polhemus FastTrak. Continuous EEG was recorded while the subject performed a series of 60 brisk extensions of the right thumb, at 7- to 10-s intervals (250 Hz sampling frequency, filtering DC-70 Hz). MRCPs were obtained by averaging epochs from −2.7 s to 1.5 s from EMG onset. The interval −2.7 to −2.5 s was used for baseline correction and noise estimate. Bipolar EMG from the extensor pollicis brevis (EPB) of the two sides was recorded to detect EMG onset and to monitor for relaxation; eye movements were monitored using bipolar electro-oculogram. Trials with artifacts, incomplete muscle relaxation between movements or mirror movements (MM), were excluded from further analysis. MM were counted in each subject. 4.2. MRI Using a block design (ABAB), where five periods of activation were alternated with six periods of rest (each period of activation and rest consisting of five measurements), the subjects were scanned while performing a simple motor task consisting in repetitive flexion-extension of the last four fingers of the right hand simultaneously. The movements were paced by a metronome at a 1-Hz frequency and was visually inspected by an MRI operator in order to exclude blocks with mirror movements. Using a 1.5 Tesla MR scanner (Vision, Siemens, Erlangen, Germany), fMRI scans were acquired using a T2⁎weighted single shot echo-planar imaging (EPI) sequence (TE/ TR= 66 ms/3.0 s, flip angle [FA]= 90°, matrix size= 128 × 128, field of view [FOV] = 256×256 mm, 24 axial contiguous slices, thickness= 5 mm). On the same occasion, a whole brain sagittal 3D T1-weighted magnetization-prepared rapid acquisition gradient echo (MP-RAGE) sequence (TR/TE= 11.4/4.4, FA= 15°, FOV= 256× 256, matrix size= 256 × 256, slab thickness = 160 mm, voxel size = 1 × 1 × 1 mm), including the three anatomical landmarks (nasion, NAS, right and left pre-auricular points, PAR and PAL) required for the MRI-TMS-EEG co-registration process, was also acquired. fMRI was analyzed using the statistical parametric mapping (SPM99) software (Friston et al., 1995). Prior to BR A IN RE S E A RCH 1 3 08 ( 20 1 0 ) 6 8 –7 8 statistical analysis, all images were realigned to the first one to correct for subject motion, spatially normalized into the standard SPM space, and smoothed with a 10-mm, 3D-Gaussian filter. Changes in BOLD contrast associated with the performance of the motor tasks were assessed on a pixel-by-pixel basis, using the general linear model (Friston et al., 1995) and the theory of Gaussian fields (Worsley and Friston, 1995). Specific effects were tested by applying appropriate linear contrasts. Significant haemodynamic changes for each contrast were assessed using t statistical parametric maps (SPMt). Activations below a threshold of p < 0.05 corrected for multiple comparisons were reported. MP-RAGE images were reorientered in neurological convention. Then, using the graphical interface of the vtkCISG tool (Hartkens et al., 2002), EPI mean images were coregistered to the transformed MP-RAGE images, and the resulting transformation was applied to SPMt maps of activations, to transform them from EPI coordinate system into the MP-RAGE one. 4.3. TMS Focal TMS was applied using a 70 mm eight-shaped coil (Magstim 200 stimulator, Magstim Co., Whitland, Dyfed, UK), orientated perpendicularly to scalp surface, on a grid with 75 1 cm steps drawn on a cap tightly adherent to the scalp and centered over the C3–C3A line. Bipolar EMG from the right abductor pollicis brevis (APB) muscle was recorded, TMS pulses were delivered at intensity 15% above motor threshold on its hot spot. Starting from such position, in order to improve mapping accuracy (Classen et al., 1998), adjoining grid points were stimulated until no response could be obtained. The 3D positions of the stimulation sites and the three anatomical landmarks (NAS, PAR, PAL) have been digitized using the Polhemus FastTrak digitizer (Polhemus, Kaiser Aerospace & Electronics, Colchester, VT). The average peak-to-peak amplitude values from 3 subsequent MEPs, with amplitudes over 50 μV (set as threshold value), were used for mapping. To visualize TMS results, the stimulated scalp points were projected over the cerebrospinal fluid (CSF) outer surface. This process, performed using Curry v4.6 software (Neuroscan, Inc., Herndon, VA), collapses each grid point in lateral to medial and antero-posterior direction (Fig. 4), using the three landmark points to calculate the center of such projection. Their corresponding MEP intensities were mapped by interpolation creating hundreds of CSF projected MEP values. CSF was selected for mapping being a regular surface only few mm from the cortex, allowing to clearly visualize results. In order to compare TMS results with those from fMRI and EEG, the Fig. 4 – MEPs projection process. Left: TMS montage over scalp and CSF surface. Stimulation positions over the scalp (TS) were projected to CSF according to TSC segment, connecting TS with Curry coordinate system origin, individuated from the intersection of PAL, PAR and Nasion (not shown). t and n are respectively the tangential and orthogonal planes to scalp surface in TS position. Right upper: TMS montage projection over CSF (left) and scalp (right) surfaces. Right lower: 2D representation of the 3D procedure to calculate distances among TMS scalp measures (HS and COG) and cortical EEG sources and fMRI maxima (both represented by S). COG and HS, calculated over scalp surface, were projected along the segment n, perpendicular to scalp surface (t), until its distance from the TS (TSTP) coincided with that of TS from S (TSS). The Euclidean distance TPS was then calculated. Segment TcLTnL represents the differences, at CSF level, between Curry projection (along TSC) and the one orthogonal to scalp surface (n). 76 BR A IN RE S EA RCH 1 3 08 ( 20 1 0 ) 6 8 –78 coordinates of hotspot (HS), as the stimulation site with the highest MEP, and center of gravity (COG), as the average of stimulated position coordinates weighted by MEP intensity (Lotze et al., 2003; Neggers et al., 2004), calculated as following: COGX = (ΣX MEPX × X) / ΣX MEPx, where MEPX is the intensity of the x-th MEP value at X-th grid position COGY = (ΣY MEPY × Y/ΣY MEPY , where MEPX is the intensity of the x-th MEP value at X-th grid position Considering that projection using Curry was performed according to an angle different from the one used to stimulate the scalp (perpendicular to it) and that fMRI maxima, CCD sources and neurons activated by magnetic stimulation reside into the cortex, their comparisons with COG and HS projections to CSF would have been affected by two errors. The first, acting in an antero-posterior and medial–lateral direction, was corrected by projecting them perpendicularly to scalp surface. To accomplish that, the spheres interpolating the stimulation points and the direction cosines of the radius passing from HS and COG (segment TSTP of Fig. 4) were calculated. The second bias, acting in a dorsal-ventral direction, was corrected by projecting COGs and hotspots until their distance from the original scalp point coincided with that of the compared fMRI or CCD source (Fig. 4, lower right). To allow their graphical representation, they were projected to CSF surface and overlaid on the cortical surface. 4.4. MRCP sources reconstruction A highly detailed anatomical MP-RAGE (voxel dimension 1 × 1 × 1 mm) was used to create the volume conductor consisting in a 3-compartment (scalp, skull, and dura mater triangulated with about 1000 triangles for each surface, with conductivity values of 0.33–0.0042–0.33 1/Ωm) boundary element method (BEM) and the cortex surface (3 mm discretization, ∼ 16000 investigated sources) (Fuchs et al., 2001). A CCD distributed sources model was used, with sources constrained to cortex surface and orientations fixed normal to the cortical area the elementary dipole arise from (Dale and Sereno, 1993), in order to prevent unrealistic activations. Lead-field matrix normalization was performed with an optimal componentwise depth weighting method (Fuchs et al., 1999). To improve the signal-to-noise ratio (SNR), an independent component analysis (ICA) algorithm (Hyvarinen et al., 1999) was applied to MRCP and the resulting noise-normalized components, whose SNR was below 1 across all interval of interest, were excluded by reconstruction algorithm (Inuggi et al., 2009). For fMRI constrained CCD, different lead field matrix over-weighting factor (K) were tested: K = 1.0 (no overweight), K = 1.1, K = 1.4 and K = 3 (Wagner et al., 2001b; Babiloni et al., 2003). fMRI locations were integrated into the sources space after a 5 mm enlargement. MRCP generators have been calculated using CCD with a focal L1 norm and a more distributed L2 norm (Fuchs et al., 1999; Wagner et al., 2001a,b; Babiloni et al., 2000). CCD analysis was performed from -100 to 52 ms respect to EMG onset where motor potential (MP) could be evaluated (Toma et al., 2002; Shibasaki and Hallett, 2006). The volume conductor and the sources space were co-registered to EEG montage and TMS stimulation grid by identifying the three anatomical landmarks (NAS, PAR, PAL) from the MPRAGEderived skin surface. All these operations and source reconstructions were performed using Curry v4.6 software (Neuroscan, Inc., Herndon, VA). 4.5. MRCP M1 source selection M1 position in each CCD reconstruction was calculated using different methods. A confidence region composed by pre- and postcentral gyri (BA4 and BA1), pre and postcentral banks of central sulcus (BA4 and BA3b) and dorso-lateral premotor cortex (PMd, BA6) was defined for each individual anatomical MRI. Only sources included in such region were used to calculate M1 position. Two different spatial criteria were used. In the first, M1 was associated to the source with the highest amplitude (CCDHS), in the second, to the center-of-gravity of sources activity (CCDCOG). Concerning the temporal aspects and according to the characteristic of the two norms, two different methods were selected to define the proper latency. Using L2 norm, providing more distributed and continuous activations, the latency used to locate M1 was the one showing the smallest deviations between measured and computed potentials, a measure calculated by Curry to evaluate the reconstruction accuracy (Fuchs et al., 1999). With L1 instead this approach could not be followed, as such norm produces very few and non-continuous activations and the latency with the best fit could just show the SMA source, the other cortical area intensively active during this interval (Shibasaki et al., 2006). In the latter case, the latency used to calculate M1 was the one showing the source with the highest amplitude in the region of confidence. The center-of-mass, averaged across all timepoints, of sources activity within the confidence region was also calculated (CCDCOGT) for both norms according to the following formula: COGT = X s 4Aij ij ij  X = s ; ij ij where sij is the intensity of the i-th source at timepoint j. Aij are the x, y, z position of the i-th source at timepoint j. In L2 reconstructions, only those sources carrying currents over 50% of the largest current (full width at half maximum, FWHM) were considered. fMRI constrained CCDs were then analyzed in the same latency as unconstrained one. Each reconstructed M1 position was classified, according to its anatomical localization, in three categories: inside the precentral knob, in BA4 and in the remaining areas (BA1, BA3b, BA6). 4.6. Statistical analysis Considering the fMRI maxima as the reference positions, their distances from CCD localizations and TMS measures were calculated in each subject. They will be referred as CCDfMRInorm,K,method, TMSCOG and TMSHS. In addition, a score was attributed to each CCD localization and TMS measure according to its anatomical classification: ‘0’ to sources inside the precentral knob, ‘1’ to sources inside BA4 and ‘2’ to the remaining sources. They will be referred as CCDPOSnorm,K,method, TMS-POSCOG and TMS-POSHS. To investigate the effect of norm (L1 and L2), K value (1, 1.1, 1.4, 3) and method (HS, COG, COGT) and their interactions (norm⁎method, norm⁎k, k⁎method and norm⁎k⁎method) over CCD-fMRInorm,K,method BR A IN RE S E A RCH 1 3 08 ( 20 1 0 ) 6 8 –7 8 and CCD-POSnorm,K,method, non-parametric related samples tests (Wilcoxon signed-rank and Friedman's several-related for norm and K/method respectively) were performed. Posthoc analysis for the latter factors was performed with a nonparametric pairwise comparison (two-related samples Wilcoxon signed-rank test). The same test was used to compare TMSCOG with TMSHS and TMS-POSCOG with TMS-POSHS. Anterior-posterior and medial-lateral position differences among TMS COG, TMS HS, fMRI maximum and CCD sources and ventral–dorsal position differences for fMRI and CCD measures were separately analysed using two related samples Wilcoxon signed-rank test. Acknowledgments This study was supported by the Ministry of Health, project “Tecniche robotizzate per la valutazione ed il trattamento riabilitativo delle disabilità motorie dell’arto superiore” no. ICS 030.8/RF01.175. Authors would like to thank Eng. E. Pagani and Eng. M. Cursi for their helpful advices and Dr. Anna Cercignani for her help in the language editing of the manuscript. REFERENCES Asanuma, H, 1989. The Motor Cortex. Raven Press, New York. Babiloni, F., Carducci, F., Cincotti, F., Del Gratta, C., Roberti, G.M., Romani, G.L., Rossini, P.M., Babiloni, C., 2000. Integration of high resolution EEG and functional magnetic resonance in the study of human movement-related potentials. Methods Inf. Med. 39, 179–182. Babiloni, F., Babiloni, C., Carducci, F., Romani, G.L., Rossini, P.M., Angelone, L.M., Cincotti, F., 2003. Multimodal integration of high-resolution EEG and functional magnetic resonance imaging data: a simulation study. Adv. Neurol. 19, 1–15. Ball, T., Schreiber, A., Feige, B., Wagner, M., Lucking, C.H., Kristeva-Feige, R., 1999. The role of higher-order motor areas in voluntary movement as revealed by high-resolution EEG and fMRI. Neuroimage 10, 682–694. Bohning, D.E., He, L., George, M.S., Epstein, C.M., 2001. Deconvolution of transcranial magnetic stimulation (TMS) maps. J. Neural Transm. 108, 35–52. Cincotta, M., Ziemann, U., 2008. Neurophysiology of unimanual motor control and mirror movements. Clin. Neurophysiol. 119, 744–762. Classen, J., Knorr, U., Werhahn, K.J., Schlaug, G., Kunesch, E., Cohen, L.G., Seitz, R.J., Benecke, R., 1998. Multimodal output mapping of human central motor representation on different spatial scales. J. Physiol. 255, 163–179. Cui, R.Q., Huter, D., Lang, W., Deecke, L., 1999. Neuroimage of voluntary movement: topography of the Bereitschaftspotential, a 64-channel DC current source density study. Neuroimage 9, 124–134. Dale, A.M., Halgren, E., 2001. Spatiotemporal mapping of brain activity by integration of multiple imaging modalities. Curr. Opin. Neurobiol. 11, 202–208. Dale, A.M., Sereno, M., 1993. Improved localization of cortical activity by combining EEG and MEG with MRI cortical surface reconstruction: a linear approach. J. Cogn. Neurosci. 5, 162–176. Darling, W.G., Wolf, S.L., Butler, A.J., 2006. Variability of motor potentials evoked by transcranial magnetic stimulation depends on muscle activation. Exp. Brain Res. 174, 376–385. 77 Deecke, L., Lang, W., 1996. Generation of movement-related potentials and fields in the supplementary sensorimotor area and the primary motor area. Adv. Neurol. 70, 127–146. Friston, K.J., Holmes, A.P., Poline, J.B., Grasby, P.J., Williams, S.C., Frackowiak, R.S., Turner, R., 1995. Analysis of fMRI time-series revisited. Neuroimage 2, 45–53. Fuchs, M., Wagner, M., Kohler, T., Wischmann, H.A., 1999. Linear and nonlinear current density reconstructions. Adv. Neurol. 16, 267–295. Fuchs, M., Wagner, M., Kastner, J., 2001. Boundary element method volume conductor models for EEG source reconstruction. Clin. Neurophysiol. 112, 1400–1407. Graves de Peralta Menendez, R., Gonzalez Andino, S.L., 1998. Distributed source models: standard solutions and new developments. In: Uhl, C (Ed.), Analysis of Neurophysiological Brain Functioning. Springer Verlag, New York, pp. 176–201. Hartkens, T., Rueckert, D., Schnabel, JA., Hawkes, DJ., 2002. VTK CISG Registration Toolkit: an open source software package for affine and non-rigid registration of single- and multimodal 3D images. BVM2002. Springer-Verlag, Leipzig. Herwig, U., Kolbel, K., Wunderlich, A.P., Thielscher, A., von Tiesenhausen, C., Spitzer, M., Sch?nfeldt-Lecuona, C., 2002. Spatial congruence of neuronavigated transcranial magnetic stimulation and functional neuroimaging. Clin. Neurophysiol. 113, 462–468. Hlustík, P., Solodkin, A., Gullapalli, R.P., Noll, D.C., Small, S.L., 2001. Somatotopy in human primary motor and somatosensory hand representations revisited. Cereb. Cortex 11, 312–321. Humphrey, D.R., 1986. Representation of movements and muscles within the primate precentral motor cortex: historical and current perspectives. Adv. Neurol. 45, 2687–2699. Hyvarinen, A., 1999. Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans. Neural Netw. 10, 626–634. Ikeda, A., Luders, H.O., Collura, T.F., Burgess, R.C., Morris, H.H., Hamano, T., Shibasaki, H., 1996. Subdural potentials at orbitofrontal and mesial prefrontal areas accompanying anticipation and decision making in humans: a comparison with Bereitschaftspotential. Adv. Neurol. 98, 206–212. Inuggi, A., Amato, N., Magnani, G., González-Rosa, J.J., Chieffo, R., Comi, G., Leocani, L., 2009. Cortical control of unilateral simple movement in healthy aging. Neurobiol Aging. Kiebel, S.J., Friston, K.J., 2004. Statistical parametric mapping for event-related potentials: I. Generic considerations. Neuroimage 22, 492–502. Kiers, L., Cros, D., Chiappa, K.H., Fang, J., 1993. Variability of motor potentials evoked by transcranial magnetic stimulation Electroencephalogr. Clin. Neurophysiol. 89, 415–423. Kim, D.S., Ronen, I., Olman, C., Kim, S.G., Ugurbil, K., Toth, L.J., 2004. Spatial relationship between neuronal activity and BOLD functional MRI. Neuroimage 21, 876–885. Kleinschmidt, A., Nitschke, M.F., Frahm, J., 1997. Somatotopy in the human motor cortex hand area. A high-resolution functional MRI study. Adv. Neurol. 9, 2178–2186. Lemon, R., 1990. Mapping the output functions of the motor cortex. In: Edelman, G., Gall, W., Cowan, W. (Eds.), Signal and Sense. Wiley-Liss, New York, pp. 315–355. Leocani, L., Cohen, L.G., Wassermann, E.M., Ikoma, K., Hallett, M., 2000. Human corticospinal excitability evaluated with transcranial magnetic stimulation during different reaction time paradigms. Brain 123, 1161–1173. Liu, Z., He, B., 2008. fMRI-EEG integrated cortical source imaging by use of time-variant spatial constraints. Neuroimage 39, 1198–1214. Logothetis, N.K., 2003. The underpinnings of the BOLD functional magnetic resonance imaging signal. J. Neurosci. 23, 3963–3971. Lotze, M., Kaethner, R.J., Erb, M., Cohen, L.G., Grodd, W., Topka, H., 2003. Comparison of representational maps using functional 78 BR A IN RE S EA RCH 1 3 08 ( 20 1 0 ) 6 8 –78 magnetic resonance imaging and transcranial magnetic stimulation. Clin. Neurophysiol. 114, 306–312. Neggers, S.F., Langerak, T.R., Schutter, D.J., Mandl, R.C., Ramsey, N.F., Lemmens, P.J., Postma, A., 2004. A stereotactic method for image-guided transcranial magnetic stimulation validated with fMRI and motor-evoked potentials. Neuroimage 21, 1805–1817. Oldfield, R.C., 1971. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9, 97–9113. Penfield, W., Rasmussen, T., 1950. The Cerebral Cortex of Man. A Clinical Study of Localization of Function. Macmillan, New York. Rosler, K.M., Roth, D.M., Magistris, M.R., 2008. Trial-to-trial size variability of motor-evoked potentials. A study using the triple stimulation technique. Exp. Brain Res. 187, 51–59. Rossini, P.M., Barker, A.T., Berardelli, A., Caramia, M.D., Caruso, G., Cracco, R.Q., Dimitrijević, M.R., Hallett, M., Katayama, Y., Lücking, C.H., 1994. Non-invasive electrical and magnetic stimulation of the brain, spinal cord and roots: basic principles and procedures for routine clinical application. Report of an IFCN committee. Electroencephalogr. Clin. Neurophysiol. 91, 79–92. Sanes, J.N., Donoghue, J.P., 1997. Static and dynamic organization of motor cortex. Adv. Neurol. 73, 277–296. Shibasaki, H., 2008. Human brain mapping: hemodynamic response and electrophysiology. Clin. Neurophysiol. 119, 731–743. Shibasaki, H., Hallett, M., 2006. What is the Bereitschaftspotential? Clin. Neurophysiol. 117, 2341–2356. Sparing, R., Buelte, D., Meister, I.G., Paus, T., Fink, G.R., 2008. Transcranial magnetic stimulation and the challenge of coil placement: a comparison of conventional and stereotaxic neuronavigational strategies. Hum. Brain Mapp. 29, 82–96. Terao, Y., Ugawa, Y., Sakai, K., Miyauchi, S., Fukuda, H., Sasaki, Y., Takino, R., Hanajima, R., Furubayashi, T., Pütz, B., Kanazawa, I., 1998. Localizing the site of magnetic brain stimulation by functional MRI. Adv. Neurol. 121, 145–152. Thickbroom, G.W., Byrnes, M.L., Mastaglia, F.L., 1999. A model of the effect of MEP amplitude variation on the accuracy of TMS mapping. Adv. Neurol. 110, 941–943. Toma, K., Matsuoka, T., Immisch, I., Mima, T., Waldvogel, D., Koshy, B., Hanakawa, T., Shill, H., Hallett, M., 2002. Generators of movement-related cortical potentials: fMRI-constrained EEG dipole source analysis. Neuroimage 17, 161–173. Wagner, M., Fuchs, M., Kastner, J., 2001a. fMRI-constrained dipole fits and current density reconstruction. BIOMAG 2000. Helsinki university of Technology, Espoo, pp. 785–788. Wagner, M., Kohler, T., Fuchs, M., Kastner, J., 2001b. An extended source model for current density reconstructions. BIO-MAG 2000. Helsinky University of technology, Espoo, pp. 749–752. 2001. Wassermann, E.M., McShane, L.M., Hallett, M., Cohen, L.G., 1992. Noninvasive mapping of muscle representations in human motor cortex. Electroencephalogr. Clin. Neurophysiol. 85, 1–8. Wassermann, E.M., Wang, B., Zeffiro, T.A., Sadato, N., Pascual-Leone, A., Toro, C., Hallett, M., 1996. Locating the motor cortex on the MRI with transcranial magnetic stimulation and PET. Adv. Neurol. 3, 1–9. Wildgruber, D., Erb, M., Klose, U., Grodd, W., 1997. Sequential activation of supplementary motor area and primary motor cortex during self-paced finger movement in human evaluated by functional MRI. Neurosci. Lett. 227, 161–164. Worsley, K.J., Friston, K.J., 1995. Analysis of fMRI time-series revisited—again. Neuroimage 2, 173–181. Yousry, T.A., Schmid, U.D., Jassoy, A.G., Schmidt, D., Eisner, W.E., Reulen, H.J., Reiser, M.F., Lissner, J., 1995. Topography of the cortical motor hand area: prospective study with functional MR imaging and direct motor mapping at surgery. Radiology 195, 23–29. Yousry, T.A., Schmid, U.D., Alkadhi, H., Schmidt, D., Peraud, A., Buettner, A., Winkler, P., 1997. Localization of the motor hand area to a knob on the precentral gyrus. A new landmark. Brain 120, 141–157.