NeuroImage 16, 434 440 (2002) doi:10.1006/nimg.2002.1098, available online at http://www.idealibrary.

com on

Minds Ear in a Musician: Where and When in the Brain 1

Martin Schu rmann,* , Tommi Raij,* Nobuya Fujiki,* and Riitta Hari*
*Brain Research Unit, Low Temperature Laboratory, Helsinki University of Technology, FIN-02015 HUT, Finland; and Institute of Physiology, Medical University Lu beck, D-23538 Lu beck, Germany Received August 16, 2001

The temporospatial pattern of brain activity during auditory imagery was studied using magnetoencephalography. Trained musicians were presented with visual notes and instructed to imagine the corresponding sounds. Brain activity specic to the auditory imagery task was observed, rst as enhanced activity of left and right occipital areas (average onset 120 150 ms after the onset of the visual stimulus) and then spreading to the midline parietal cortex (precuneus) and to such extraoccipital areas that were not activated during the visual control condition (e.g., the left temporal auditory association cortex and the left and right premotor cortices). The latest activations, with average onset latencies of 270 400 ms clearly separate from the earliest ones, occurred in the left sensorimotor cortex and the right inferotemporal visual association cortex. These data imply a complex temporospatial activation sequence of multiple cortical areas when musicians recall rmly established audiovisual associations. 2002 Elsevier Science (USA) Key Words: auditory imagery; magnetoencephalography (MEG); minimum current estimate (MCE); precuneus; auditory association cortex.

trol condition (sounds) served for comparison. We expected (a) that presentation of notes during the imagery task would show specic activation (beyond that elicited by closely similar visual control stimuli); (b) that such activations would be restricted to certain brain areas dened in terms of functional anatomy; and (c) that a clear temporal sequence would be manifested in source-specic onset or peak latencies of activations during the imagery task. Comparison of imagery and control conditions revealed activity specic to auditory imagery in occipital and extraoccipital areas. Marked differences in average onset latencies (from around 120 ms to around 400 ms) allowed distinction between early occipital activations and intermediate and late extraoccipital activations. The observed spatiotemporal pattern apparently reects spread of activation from the visual cortex to several extrastriatal areas. SUBJECTS AND METHODS Eleven experienced professional-level musicians (19 to 37 years, mean age 26 years; two left-handed; 6 males and 5 females) were studied with informed consent during four conditions (Fig. 1). First, in a visual control condition, black dots were projected to an empty screen, with the instruction to look at dots. Next, in an auditory control condition, piano sounds corresponding to the notes g 1, a 1, h 1, and c 2, were presented binaurally, with the instruction to listen to sounds. The third and fourth conditions were presented in alternating blocks of 20 stimuli: (a) The note-sound associations were rehearsed, by presenting the 4 visual notes (on staff lines which were displayed continuously) together with the respective piano sounds. (b) In the auditory imagery task, notes were again presented on stafines but without auditory stimuli while the subjects were instructed to imagine the corresponding piano sounds. After the experiment, the subjects were asked to give their own estimate of how well they managed to imagine the sounds (scale 0, not at all, to

When trained musicians read musical score they may hear the corresponding sounds, even complex ones, in their minds ear. This process has been studied earlier as an example of audiovisual association, using methods that permitted identication of the spatial pattern of brain activation but could not provide any timing information (Sergent et al., 1992). Here we present magnetoencephalographic (MEG) data that shed light on the time course of brain activation sites during auditory imagery in musicians. In the main experimental condition, trained musicians were presented with visual notes and were instructed to imagine the corresponding sounds. A visual control condition (note-like stimuli) and auditory conPresented in abstract form at 12th International Conference on Biomagnetism, Espoo, Finland, August 2000 (book of abstracts, p. 65).
1053-8119/02 $35.00 2002 Elsevier Science (USA) All rights reserved.
1

434

MINDS EAR IN A MUSICIAN

435

identied during 120 500 ms, i.e., in a time window immediately following the main activations in the auditory and visual control condition. For verication of the grand average sources (which can be biased by an exceptionally high amplitude in a single subject) and for statistical assessment of source strengths and timing, additional evaluation was done at the level of individual subjects with the following steps: 1. Denition of imagery-related sources. For each individual, MCEs were dened in the auditory imagery condition. This analysis resulted in altogether 139 individual sources in the 10 subjects. 2. Specicity of imagery-related activation (in terms of amplitudes). Source strengths were measured in terms of rms amplitude in a time window within which activation exceeded 50% of the peak activation of the respective source. From these values, the prestimulus rms value 2 was subtracted to correct for noise. For assessment of task specicity, source strengths were compared between auditory imagery, visual control condition, and auditory control condition (ANOVA, within-subject factor CONDITION, in each individual subject, with post-hoc t tests). 3. Topography of imagery-specic sources. The individual sources were examined for spatial overlap that would indicate consistent activation across subjects. For this purpose, regions of overlap were identied as follows: (a) Source volumes. For each individual source, the smallest sphere containing the whole source area was computed in 3-D Talairach coordinates. (b) Overlap in Talairach space. For each point in Talairach space, the number of source spheres containing this point was counted, and regions where spheres of at least 5 sources overlapped were identied as regions of overlap. Altogether 10 regions of overlap were identied, each containing on average 8.7 individual source volumes. Therefore, 87 of the original 139 source volumes were included in the analysis after this step. (c) Time window criterion. Fourteen sources with peaks later than 500 ms were discarded to match the time window of interest dened for the grand average sources. Altogether 73 sources survived this step. (d) Balancing subjects contributions to regions of overlap. Whenever more than one source per subject belonged to one region of overlap, only the earliest source was kept. This procedure, aimed to prevent any bias toward a specic subject, resulted in removal of 11 sources, and thus 62 sources were left for further analysis.

FIG. 1. Stimuli and conditions. For further details, see text. Attend to faces task is not shown in diagram.

10, perfect). The estimates varied between 7 and 10 (median 8). In an additional control condition, presented between the second and the third/fourth condition to 5 of the 11 subjects, notes on stafines, identical to those in the following auditory imagery task, were presented as frequent stimuli to be ignored. Subjects were given a task which required attention to infrequent target stimuli (black-and-white drawings of faces, approximately equal in size to the notes stimuli; attend to faces task). The aim of this control condition was to rule out that the activation effects during auditory imagery would be due to attention to visual stimuli rather than the auditory imagery task. Throughout the experiment, stimuli were presented at an irregular rate once every 1.52 s. The visual stimuli were transmitted to the magnetically shielded MEG room via a data projector, and the sounds were delivered via earpieces. MEG responses were recorded with a 306-channel Vectorview whole-scalp neuromagnetometer (Neuromag Ltd., Finland). Data from one subject were discarded due to excessive eye blinking (reducing the size of the control group with attend to faces task to 4). Surface electrodes above larynx served to rule out possible vocalization. To estimate the source congurations, minimum current estimates (MCEs, Uutela et al., 1999), optimally explaining the signal patterns, were computed for each individual. The MCE estimates were averaged across subjects after alignment to a standard brain (Roland and Zilles, 1996), and grand average sources were

436

RMANN ET AL. SCHU

FIG. 2. Results of grand average MCE source analysis. Source names refer to M(idline), L(eft), R(ight); F(rontal), P(arietal), T(emporal), or O(ccipital) areas. (AD) Source volumes (A: listen to sounds; B and C: look at dots; D: auditory imagery). Areas activated during auditory imagery are highlighted by circles (averaged over time within 120 500 ms, only those areas most consistently observed across subjects are highlighted by circles). Gray circles indicate occipital activation during auditory imagery (for comparison, these circles are shown in C as well); white circles show extraoccipital activation during auditory imagery. Source areas activated during the control conditions (60 120 ms) are highlighted by black circles (for easy reading, the black circles from A have also been drawn in the gures for auditory imagery). The highlighted areas in A are smaller than those in B, C, and D; this either reects less spatial variation across subjects in the former condition, deeper sources for the latter two conditions, or a real size difference between the activated regions. Calibration 0 2 nAm (nanoAmpere X meter) for A and 0 0.35 nAm for B, C, and D. (The amplitude differences between averaged MCEs in panels B, C, and D and the individual MCE source strengths in Table 1 are accounted for by spatial variation across subjects.) Posterior view in A and lateral views in B have been omitted due to the small amplitude of the respective activations. The asterisks below the back views in panels C and D mark a strong activation which is most probably due to an artifact (the inferior surface of the model is sensitive to such artifacts). (E) Timing of activation for individual sources in the regions of overlap corresponding to the grand average source volumes shown in Panel D. Vertical bars show average onset latencies, horizontal bars extend from earliest to latest individual onset latencies. Shading shows classication into Early, Intermediate, and Late onset latency groups according to Table 3.

(e) Relationship between regions of overlap and grand average source volumes. Each of the 10 regions of overlap corresponded spatially to one of 10 grand average source volumes shown in Fig. 2. 4. Timing of topographically dened imagery-specic sources. Using time windows dened in step 2, the onset and peak latencies of all 62 sources constituting the regions of overlap were dened to quantify the dynamics of cortical activation. The values were subjected to ANOVA. Post-hoc t tests served to identify possible categories of sources with similar onset latencies. RESULTS Grand average MCEs during different tasks. For the listen to sounds condition, the main activation

occurred in the supratemporal auditory cortices (Hari, 1990; Talairach coordinates: left 52 21 9 and right 54 18 11; Talairach and Tournoux, 1988), and during the look at dots condition in several posterior brain areas (Figs. 2A2C). During auditory imagery, activation was observed in areas similar to those activated during the visual control condition but also in several further areas (Fig. 2D). Task-specicity of activation during auditory imagery assessed at individual level. In each subject, several sources were activated during the auditory imagery condition. In all but 2 of the 139 sources observed in the 10 subjects, the activation was stronger during auditory imagery than visual control condition. Similarly, all but 2 of 139 sources showed stronger activation for auditory imagery than for auditory control condition. Such imagery-specic activations were observed in all 10 subjects (P 0.001 or better in ANOVA

MINDS EAR IN A MUSICIAN

437

TABLE 1 MCE Amplitudes in Individual Subjects

N of sources (imagery visual) 10 (10) 18 (17) 16 (16) 12 (11) 13 (13) 15 (15) 14 (14) 12 (12) 19 (19) 10 (10) rms (imagery task) 4.3 5.5 4.5 5.3 6.2 6.9 6.3 5.1 5.2 5.4 rms (visual control) 0.8 1.4 1.5 1.2 0.4 0.8 0.5 0.3 1.1 0.8 rms (auditory control) 0.3 0.1 0.4 0.3 0.1 0.3 0.7 0.2 0.5 0.1 ANOVA (factor CONDITION) P P P P P P P P P P 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 post-hoc t-test (imagery vs. visual control) P P P P P P P P P P 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 post-hoc t-test (imagery vs. auditory control) P P P P P P P P P P 0.001 0.001 0.001 0.001 0.001 0.001 0.002 0.001 0.001 0.001 post-hoc t-test (visual control vs. auditory control) n.s. n.s. P 0.05 n.s. n.s. n.s. n.s. n.s. n.s. P 0.05

Subject 1 2 3 4 5 6 7 8 9 10

Note. The rms values are given as source strengths in nAm (nanoAmpere meter) and are the averages across all sources found in a particular subject.

with P 0.002 or better in post-hoc tests for imagery vs visual control and imagery vs auditory control, see Table 1). Further analysis steps addressed source topography and timing of activation. Topography of imagery-specic activations. Imagery-specic sources (dened in the grand average according to criteria given above and conrmed by spatial correspondence with regions of overlap; see Table 2) were found in occipital as well as in extra-occipital areas. Sources in the left and right occipital areas (LO/RO) and in the midline occipito-parietal area (MOP) were also active during the look at dots condition (as seen in the comparison of Figs. 2D and 2C). However, activation of individual sources was on average stronger for auditory imagery than for look at dots.

In contrast to task-specic modulation of occipital source strengths, extra-occipital sources were activated only during auditory imagery (cf. Fig. 2D vs 2C). In the left hemisphere, auditory imagery activated the anterior and posterior parts of the superior temporal area (LTa, LTp), the frontoparietal area (LFP), and the frontal area (LF). In the right hemisphere, the frontal area (RF), the supramarginal parietal area (RP), and the inferior temporal area (RT) were activated (see Table 2 for grand average source coordinates). Timing of imagery-specic activations. First, LO and RO activations started on average at 150 and 120 ms (Fig. 2E) and peaked at 160 and 180 ms. Subsequently, activations of MOP, LTp, LTa, LF, RF, and RP started on average between 170 and 250 ms (Fig. 2E)

TABLE 2 Areas Activated during Auditory Imagery

Talairach coordinates Region: topographical (Brodmann); functional Occipital sources LO left occipital (BA 19); visual association cortex RO right occipital (BA 19); visual association cortex MOP midline precuneus (BA7, 19) Extraoccipital sources LTa left superior temporal (anterior, BA22); auditory association cortex LTp left superior temporal (posterior, BA22); auditory association cortex LFP left frontoparietal (BA1,2,3); primary somatosensory cortex LF left frontal (BA6); premotor cortex RF right frontal (BA6); premotor cortex RP right parietal (BA40); anterior supramarginal cortex RT right inferior temporal (BA21/37); visual association cortex x 33 35 4 58 40 38 43 31 54 53 y 71 75 70 22 56 25 3 1 19 40 z Number of subjects

2 3 30 4 15 44 40 43 22 5

7 9 7 8 5 5 4 7 5 5

Note. Areas identied in grand average MCE results and conrmed as areas of overlap of individual sources. Topographical anatomy (notation as in Fig. 2), the most likely Brodmann areas, functional anatomy, and Talairach coordinates (Talairach and Tournoux, 1988) of the centres of gravity. The rightmost column gives the number of subjects contributing to the areas of overlap corresponding to the respective grand average source volume (see text for details).

438

RMANN ET AL. SCHU

TABLE 3 Statistically Signicant Differences (and Tendencies) between Onset Latencles of 62 Sources during Auditory Imagery in 10 Individuals
Early LO N7 Early LO RO MOP LTp* LTa LF* RP RF LFP RO N9 n.s. MOP N7 n.s. 0.094 LTp* N5 n.s. n.s. n.s. Intermediate LTa N8 0.037 0.012 n.s. n.s. LF* N4 n.s. n.s. n.s. n.s. n.s. RP N5 0.069 0.030 n.s. n.s. n.s. n.s. RF N7 0.060 0.024 n.s. n.s. n.s. n.s. n.s. LFP N5 0.002 0.001 0.024 n.s 0.045 n.s. 0.023 n.s. Late RT N5 0.002 0.001 0.065 n.s. n.s. n.s. 0.018 n.s. n.s.

Intermediate

Late

Note. Results of post-hoc t tests following ANOVA, F(9,52) 2.95, P 0.007. Categories Early, Intermediate, and Late were dened in a way that there are at least 4 signicant differences for all possible comparisons between groups (cells marked in gray) but no signicant differences within groups (italics). *Sources LTp and LF, the only sources without any signicant differences, classied as intermediate on the basis of average latency alone (see Fig. 2E).

and peaked between 220 and 320 ms. Finally, RT and LFP were activated (average onset latencies of 280 and 400 ms, Fig. 2E; average peak latencies 340 and 440 ms). Differences in onset as well as peak latencies were statistically signicant (ANOVA for 62 sources contributing to regions of overlap corresponding to sources LORT: onset latencies, F(9,52) 2.95; P 0.007; peak latencies, F(9,52) 2.57, P 0.016). Onset latencies rather than peak latencies were chosen for further analysis in terms of post-hoc t tests for all pairs of the 10 sources LORT. Eight of 10 sources (LO, RO, MOP, LTa, RF, RP, LFP, RT) contributed to at least one signicant difference as shown in Table 3. Sources were divided into Early, Intermediate, and Late groups so that all within-group onset differences were statistically nonsignicant, whereas most of the between-group differences were signicant (most consistent differences were seen between LO, RO, and LFP, RT, all pairs P 0.002 or better). Sources LTp and LF (the only ones without any signicant differences to others sources) were grouped together with MOP, LTa, RF, RP, to which they corresponded most closely in terms of average onset latencies. The resulting classication (Early: LO, RO; Intermediate: MOP, LTp, LTa, LF, RF, RP; Late: LFP, RT) is shown in Table 3 and Fig. 2E. Additional control experiments run in four subjects to rule out attentional effects during auditory imagery. During the attend to faces task, infrequent drawings of faces served to make the subjects ignore the visual notes. Consequently, visual notes activated auditory imagery-related sources less intensely when presented during the attend to faces task. The order of activation strengths for 61 individual sources in 4 subjects

was auditory imagery attend to faces (P 0.001) visual control (P 0.01, repeated measures ANOVA). DISCUSSION Our results show that the temporospatial precision of MEG adds functionally essential information to previous PET, fMRI, and EEG studies (Sergent et al., 1992; Beisteiner et al., 1994; Petsche et al., 1996; Zatorre et al., 1998; Halpern and Zatorre, 1999). Brain activation during auditory imagery was found to be imagery-specic and not only due to attentional modication of visual responses (as demonstrated by comparison with similar visual stimuli during the look at dots condition and with physically identical stimuli during the attend to faces task). Such imagery-specic activation was either manifested as enhancement of activity that was also observed during the visual control condition or as activation of areas that were silent during the visual control condition. Activity of occipital areas, also seen during the visual control condition, was enhanced during imagery from as early as 120 or 150 ms onwards (RO, LO in Early latency group). With an average onset at 210 ms, the precuneus region (MOP, Intermediate latency group) followed. Our auditory imagery task required analysis of the spatial location of the notes with respect to the stafines. Accordingly, precuneus has previously been associated with visuospatial and visuomotor processing (Andersen et al., 1997) and with visual imagery (Fletcher et al., 1995; Salenius et al., 1995). Activation of this area has also been reported during detection of pitch changes in a sequence of sounds, possibly sug-

MINDS EAR IN A MUSICIAN

439

gesting a visual processing strategy (Platel et al., 1997). Imagery-specic activity was also observed in extra-occipital areas that were silent during the visual control condition. Of seven such sources, ve were in the Intermediate group (average onset latency 170 250 ms): (1) LTp, situated in the left posterior superior temporal sulcus, is close to a region which is activated when subjects read aloud visually presented words (Paulesu et al., 2000) and possibly has a role in grapheme-to-phoneme conversion. Reading musical score might utilize a similar mechanism. (2) LTa is close to the left anterior superior temporal sulcus, an auditory association area which has been found to participate in the conversion of visual letters to auditory phonemes (Raij et al., 2000). Activation of the temporal lobe close to the auditory cortex and at the posterior end of the lateral sulcus has been found in subjects with absolute pitch who recalled the names of visually presented notes (Hirata and Kuriki, 2000). (3) RF is situated near the dorsolateral frontal cortex which is active in tasks requiring the recall of previousy learned associations, e.g., pitch judgment (Zatorre et al., 1998). Similarly, our imagery task required association of the visual note image with the corresponding sound. Also, neighbouring areas have been found to be active during auditory imagery triggered by auditory cues, interpreted as motor planning associated with a subvocal singing or humming strategy (Halpern and Zatorre, 1999). (4) Activation of LF in the left premotor cortex may reect association (Zatorre et al., 1998) of the visual note with the sound (similar to its right-side counterpart RF). (5) RP, located in the right anterior supramarginal area, might have a role in visual-toauditory mapping; it is close to regions that are activated when musicians read musical score while listening to the respective music (Sergent et al., 1992), and lesions in this region can lead to alexia with agraphia (Sergent et al., 1992). The remaining two extraoccipital sources were in the Late group: (1) Activation of LFP (average onset latency 400 ms) in the left sensorimotor area could reect the musicians automatic association on how the sounds would be played on an instrument. (2) RT (average onset latency 280 ms) is close to the right inferior temporal region, a visual association area which has been shown to be active in multimodal processing, such as visual-verbal association in color-word synaesthesia (Paulesu et al., 1995), and in a pitch judgement task (Zatorre et al., 1998). Early activation specic to auditory imagery, with MEG peaks around 150 and 250 ms after a visual cue (hammer striking an anvil), has been reported recently (Hoshiyama et al., 2001); activation was found in the right inferior frontal sulcus and in the right insular region but no consistent left-hemisphere activation

and no modulation of occipital activity were observed. Our results are clearly different, showing consistent left-hemisphere activation, notably in the auditory association cortex (found in 8 of 10 subjects; average onset latency 250 ms). In contrast with the present study, the hammer-on-anvil task, with an instruction to keep the hammering sound in mind during soundless sequences, does not require transformation of spatial visual information to a specic sound. Furthermore, the very meaning of the musical note as the element of a semantically and syntactically dened notation system was the key element in the present task, but absent in the hammer-on-anvil task. Finally, Hoshiyama et al. (2001) modeled the sources with a single moving equivalent dipole whereas we were able to assess with MCE multiple sources even with temporal overlap. In conclusion, neural activation specic to visually triggered auditory imagery showed a well-dened temporal sequence in professional musicians. First, activity in the occipital cortex was enhanced in an imageryspecic manner. The subsequent activations were observed in precuneus, left superior temporal, left and right frontal, and right parietal brain areas and, nally, in left frontoparietal and right inferior temporal areas. Therefore the cortical network activated during auditory imagery, triggered by visually presented musical notation, includes auditory association areas that are involved in forming and recalling rmly established audiovisual associations. ACKNOWLEDGMENTS
We thank Dr. G. Curio for valuable discussions. This study was supported by the Academy of Finland, the Sigrid Juse lius Foundation, the Japan Society for the Promotion of Science, and the European Unions HCM Program through the Large-Scale Facility NeuroBIRCH III at the Brain Research Unit of the Low Temperature Laboratory, Helsinki University of Technology.

REFERENCES
Andersen, R. A., Snyder, L. H., Bradley, D. C., and Xing, J. 1997. Multimodal representation of space in the posterior parietal cortex and its use in planning movements. Annu. Rev. Neurosci. 20: 303330. Beisteiner, R., Altenmu ller, E., Lang, W., Lindinger, G., and Deecke, L. 1994. Watching the musicians brain. Eur. J. Cogn. Psychol. 6: 311327. Fletcher, P. C., Frith, C. D., Baker, S. C., Shallice, T., Frackowiak, R. S., and Dolan, R. J. 1995. The minds eyePrecuneus activation in memory-related imagery. NeuroImage 2: 195200. Halpern, A. R., and Zatorre, R. J. 1999. When that tune runs through your head: A PET investigation of auditory imagery for familiar melodies. Cerebral Cortex 9: 697704. Hari, R. 1990. The neuromagnetic method in the study of the human auditory cortex. In Auditory Evoked Magnetic Fields and Electric Potentials (F. Grandori, M. Hoke, and G. Romani, Eds.), pp. 222 282. Karger, Basel.

440

RMANN ET AL. SCHU Raij, T., Uutela, K., and Hari, R. 2000. Audiovisual integration of letters in the human brain. Neuron 28: 617 625. Roland, P. E., and Zilles, K. 1996. The developing European computerized human brain database for all imaging modalities. NeuroImage 4: S39 S47. Salenius, S., Kajola, M., Thompson, W. L., Kosslyn, S., and Hari, R. 1995. Reactivity of magnetic parieto-occipital alpha rhythm during visual imagery. Electroencephalogr. Clin. Neurophysiol. 95: 453 462. Sergent, J., Zuck, E., Terriah, S., and MacDonald, B. 1992. Distributed neural network underlying musical sight-reading and keyboard performance. Science 257: 106 109. Talairach, J., and Tournoux, P. 1988. Co-planar Stereotaxic Atlas of the Human Brain. Thieme, New York. Uutela, K., Ha ma la inen, M., and Somersalo, E. 1999. Visualization of magnetoencephalographic data using minimum current estimates. NeuroImage 10: 173180. Zatorre, R. J., Perry, D. W., Beckett, C. A., Westbury, C. F., and Evans, A. C. 1998. Functional anatomy of musical processing in listeners with absolute pitch and relative pitch. Proc. Natl. Acad. Sci. USA 95: 31723177.

Hirata, Y., and Kuriki, S. 2000. Music score elicits neuronal activation in the temporal region. Book of Abstracts, 12th International Conference on Biomagnetism, p. 64. Espoo, Finland. Hoshiyama, M., Gunji, A., and Kakigi, R. 2001. Hearing the sound of silence: A magnetoencephalographic study. NeuroReport 12: 1097 1102. Paulesu, E., McCrory, E., Fazio, F., Menoncello, L., Brunswick, N., Cappa, S. F., Cotelli, M., Cossu, G., Corte, F., Lorusso, M., Pesenti, S., Gallagher, A., Perani, D., Price, C., Frith, C. D., and Frith, U. 2000. A cultural effect on brain function. Nature Neurosci. 3: 9196. Paulesu, E., Harrison, J., Baron-Cohen, S., Watson, J. D., Goldstein, L., Heather, J., Frackowiak, R. S., and Frith, C. D. 1995. The physiology of coloured hearing. A PET activation study of colourword synaesthesia. Brain 118: 661 676. Petsche, H., von Stein, A., and Filz, O. 1996. EEG aspects of mentally playing an instrument. Cogn. Brain Res. 3: 115123. Platel, H., Price, C., Baron, J. C., Wise, R., Lambert, J., Frackowiak, R. S., Lechevalier, B., and Eustache, F. 1997. The structural components of music perception. A functional anatomical study. Brain 120: 229 243.