Conceived and designed the experiments: AB CL. Performed the experiments: CL. Analyzed the data: AB CL. Wrote the paper: AB CL.
The authors have declared that no competing interests exist.
To investigate the neural substrates that underlie spontaneous musical performance, we examined improvisation in professional jazz pianists using functional MRI. By employing two paradigms that differed widely in musical complexity, we found that improvisation (compared to production of over-learned musical sequences) was consistently characterized by a dissociated pattern of activity in the prefrontal cortex: extensive deactivation of dorsolateral prefrontal and lateral orbital regions with focal activation of the medial prefrontal (frontal polar) cortex. Such a pattern may reflect a combination of psychological processes required for spontaneous improvisation, in which internally motivated, stimulus-independent behaviors unfold in the absence of central processes that typically mediate self-monitoring and conscious volitional control of ongoing performance. Changes in prefrontal activity during improvisation were accompanied by widespread activation of neocortical sensorimotor areas (that mediate the organization and execution of musical performance) as well as deactivation of limbic structures (that regulate motivation and emotional tone). This distributed neural pattern may provide a cognitive context that enables the emergence of spontaneous creative activity.
A significant number of recent studies have used functional neuroimaging methods to investigate the perception of musical stimuli by the human brain
Creativity is a quintessential feature of human behavior, but the neural substrates that give rise to it remain largely unidentified. Spontaneous artistic creativity is often considered one of the most mysterious forms of creative behavior, frequently described as occurring in an altered state of mind beyond conscious awareness or control
In this study, we used functional MRI to study improvisation, which is the hallmark of jazz music
Since musical improvisation is an extraordinarily complex human behavior, we felt that it should be examined using paradigms that, while amenable to experimental constraint, are of high ecological validity (as argued by Burgess and colleagues; see
In the upper portion of the figure, the non-ferromagnetic MIDI piano keyboard that was used during functional MRI scanning is shown. This keyboard had thirty five full-size piano keys which triggered high-quality piano sound samples generated outside of the scanner, which were immediately routed back to the musicians using audiophile quality electrostatic earphone speakers. During scanning, subjects were randomly cued to play either the over-learned control condition or to improvise spontaneously. For Scale's control condition, subjects repeatedly played a one octave ascending and descending C major scale in quarter notes for the duration of the block (ScaleCtrl, upper left). For Scale's improvisation condition, subjects improvised in quarter notes only, selecting all notes from within one octave and from the C major scale notes alone (example shown under ScaleImprov, upper right). For Jazz's control condition, subjects played a novel melody that was memorized prior to scanning (JazzCtrl, lower left). For Jazz's improvisation condition, subjects improvised using the composition's underlying chord structure as the basis for spontaneous creative output (example shown under JazzImprov, lower right). Note that for JazzCtrl and JazzImprov, eighth notes are typically performed with a “swing” feel that is not accurately represented using standard musical notation, in both the control and improvisation conditions. Audio samples of the four musical excerpts shown here are provided in Supporting Information.
In Scale's control condition (referred to hereafter as ScaleCtrl), subjects repeatedly played a one-octave C major scale in quarter notes. During the corresponding improvisation condition (referred to as ScaleImprov), subjects improvised a melody, but were restricted to the use of C major scale quarter notes within the same octave. In the Jazz paradigm, we aimed to reproduce the high degree of musical richness of a jazz performance. Subjects were asked to memorize an original jazz composition (
All notes were recorded using MIDI (Musical Instrument Digital Interface) technology and measures derived from these recordings—total number, rate and range of musical notes and finger/hand movements—were statistically compared off-line. Thus, for each paradigm, motor activity and lower level auditory features in both conditions could be matched, with the only difference being whether the musical output was improvised or over-learned (see
The statistical analysis of piano MIDI performance data by paired T-tests revealed no significant difference between total number or weighted distribution of notes played during improvisation or control conditions for either Scale or Jazz paradigms (
Scale | Jazz | |||||
Control | Improv | Control | Improv | |||
348.67 (1.03) | 349.17 (1.47) | 0.076 | 755.33 (20.76) | 787 (184.7) | 0.66 | |
23.45 (0.01) | 23.53 (0.20) | 0.37 | 23.13 (0.12) | 24.55 (1.76) | 0.11 |
Data in
Functional imaging data were analyzed using SPM99 through standard contrasts (and inclusive masking where appropriate), conjunctions between paradigms, and comparison of hemodynamic response functions (see Experimental Procedures for further details). In order to be deemed significant, clusters of activation associated with improvisation were required to demonstrate both greater activity levels vs. resting baseline as well as greater activity levels vs. control conditions; clusters of deactivation were required to show both lower activity levels vs. resting baseline as well as lower activity levels vs. control conditions. This additional masking allowed us to distinguish true experimental activations from relative activations caused by deactivation during the control condition.
Both paradigms yielded strikingly similar results (
In both paradigms, spontaneous improvisation was associated with widespread deactivation in prefrontal cortex throughout DLPFC and LOFC, combined with focal activation in MPFC. In addition, increases in sensorimotor activity and decreases in limbic activity were seen in both paradigms. Activations were identified through inclusive masking of the contrast for [Improv–Control] with the contrast for [Improv–Rest], and deactivations were identified through inclusive masking of the contrast for [Control–Improv] with the contrast for [Rest–Improv] for both Scale and Jazz paradigms. The scale bar shows t-score values and the sagittal section shows an anatomical representation of slice location; both scale bar and sagittal slice insets apply equally to Scale and Jazz data. Labels refer to axial slice z-plane in Talairach space.
Medial prefrontal cortex activation, dorsolateral prefrontal cortex deactivation, and sensorimotor activation can be seen. The scale bar shows the range of t-scores; the axes demonstrate anatomic orientation. Abbreviations: a, anterior; p, posterior; d, dorsal; v, ventral; R, right; L, left.
Region | BA | Left Hemisphere | Right Hemisphere | ||||||
t-score | t-score | ||||||||
Polar MPF-ventral | 10 | - | - | - | - | 15.97 | 12 | 57 | −6 |
Polar MPF-middle | 10 | 11.26 | −27 | 53 | −2 | 11.26 | 7 | 61 | 3 |
Polar MPF-dorsal | 10 | 15.68 | −27 | 63 | 15 | 14.04 | 3 | 63 | 12 |
Dorsal MPFC | 8,9 | −16.23 | −12 | 48 | 36 | −18.15 | 12 | 51 | 33 |
Medial DLPFC | 46 | −7.441 | −30 | 41 | 34 | −14.71 | 51 | 30 | 27 |
Lateral DLPFC | 9 | −22.05 | −42 | 21 | 39 | −20.79 | 39 | 24 | 39 |
Superor DLPFC | 8 | −15.67 | −36 | 18 | 51 | −12.81 | 41 | 17 | 53 |
Ventral LOFC | 47,11 | - | - | - | - | −11.42 | 33 | 21 | −24 |
Mid LOFC | 11 | −14.81 | −45 | 42 | −15 | −13.51 | 33 | 39 | −15 |
All coordinates are described according to the Montreal Neurological Institute system, and were obtained using a conjunction analysis of data from ScaleImprov and JazzImprov. Activations (positive t-scores) and deactivations (negative t-scores) are shown. Abbreviations: BA, Brodmann Area; MPFC, medial prefrontal cortex; DLPFC, dorsolateral prefrontal cortex; LOFC, lateral orbitofrontal cortex
Within the prefrontal cortex, a dissociated pattern of activity was seen during improvisation. This was characterized by widespread deactivation that included almost all of the lateral prefrontal cortices, extending from lateral orbitofrontal cortex (LOFC) to the superior portions of the dorsolateral prefrontal cortex (DLPFC), as well as dorsal portions of the medial prefrontal cortex (MPFC). However, this broad pattern of deactivation was also accompanied by focal activation of the frontal polar portion of the MPFC (
Broad increases in sensorimotor activity were associated with improvisation. In neocortical sensory areas, activations were seen in anterior portions of superior and middle temporal gyri (STG and MTG), including anterior portions of the superior temporal sulcus (STS), inferior temporal, fusiform and lateral occipital gyri, as well as inferior and superior parietal lobules and the intervening intraparietal sulci. In neocortical premotor and motor areas, selective activation during improvisation was seen in both ventral and dorsal lateral premotor areas, supplementary motor area and portions of the primary motor cortex. The anterior cingulate cortex, cingulate motor area, right lateral cerebellar hemisphere, and vermis were activated as well (
Widespread attenuation of activity in limbic and paralimbic regions was seen during improvisation. Selective deactivations were in this case detected in the amygdala, entorhinal cortex, temporal pole, posterior cingulate cortex, parahippocampal gyri, hippocampus and hypothalamus (
Region | BA | Left Hemisphere | Right Hemisphere | ||||||
t-score | t-score | ||||||||
Frontal operculum-p. triangularis | 45 | 6.51 | −51 | 33 | 3 | - | - | - | - |
Frontal operculum-p. opercularis | 44 | 11.42 | −52 | 8 | 17 | - | - | - | - |
Dorsal frontal operculum | 44/6 | 16.54 | −54 | 0 | 30 | 7.14 | 51 | 6 | 27 |
Dorsal Lateral PMC | 4/6 | 11.34 | −30 | −15 | 64 | 9.18 | 30 | −6 | 63 |
SMA proper | 6 | 16.93 | −3 | 0 | 63 | 10.39 | 3 | −4 | 68 |
Dorsal MI | 4 | 14.58 | −27 | −15 | 54 | 10.04 | 27 | −9 | 51 |
STG | 22 | - | - | - | - | 6.14 | 63 | −33 | 9 |
Ant MTG-STS | 21 | 11.19 | −63 | −27 | −9 | 10.72 | 57 | −21 | −9 |
Ant MTG-ITG | 20/21 | 10.39 | −51 | −15 | −24 | 6.41 | 45 | −15 | −18 |
Fusiform-ITG | 37 | 15.74 | −48 | −66 | −21 | - | - | - | - |
SMG | 40 | 11.34 | −53 | −41 | 41 | 12.44 | 48 | −41 | 41 |
IPS | 40/7 | 16.05 | −42 | −45 | 45 | 17.46 | 45 | −42 | 51 |
SPL | 7 | 20.62 | −18 | −75 | 51 | 14.09 | 21 | −77 | 55 |
Inf OG | 18 | 7.01 | −36 | −90 | −5 | ||||
Mid OG | 18/19 | 7.87 | −27 | −94 | 14 | 11.18 | 36 | −75 | 18 |
Sup OG | 19 | 10 | −21 | −94 | 29 | 7.638 | 35 | −83 | 25 |
ACC D | 32/24 | 10.71 | −5 | 8 | 49 | - | - | - | - |
Dentate | - | - | - | - | - | 7.96 | 21 | −63 | −30 |
Post Hemisphere | - | - | - | - | - | 7.94 | 3 | −78 | −39 |
Vermis | - | - | - | - | - | 6.22 | 6 | −67 | −17 |
All coordinates are described according to the Montreal Neurological Institute system, and were obtained through a conjunction analysis of data from ScaleImprov and JazzImprov.
Region | BA | Left Hemisphere | Right Hemisphere | ||||||
t-score | t-score | ||||||||
Hypothalamus | - | −8.51 | −11 | −6 | −9 | −11.12 | 9 | −6 | −6 |
Amygdala | - | −14.64 | −24 | 0 | −18 | −7.205 | 28 | −1 | −16 |
HPC/PHPC-ventral | - | - | - | - | - | −12.49 | 27 | −24 | −18 |
HPC/PHPC-dorsal | - | −9.71 | −24 | −36 | −3 | −10.06 | 15 | −42 | 3 |
PHPC gyrus | 35,36 | −13.08 | −36 | −27 | −21 | −11.34 | 24 | −27 | −24 |
Posterior cingulate | 23,31 | −13.92 | −3 | −51 | 24 | −18.14 | 3 | −57 | 30 |
Temporal polar | 38,20 | −14.27 | −30 | 3 | −24 | −12.99 | 33 | 0 | −39 |
Ventral striatum | - | −15.41 | −30 | −12 | −9 | −13.12 | 27 | 6 | −9 |
Caudate | - | −7.03 | −12 | 15 | 11 | −10.37 | 9 | 15 | 3 |
Putamen | - | −6.61 | −29 | −3 | 6 | −10.76 | 27 | −15 | 6 |
Ant insula/pyriform cortex | - | −6.61 | −33 | 15 | −11 | −10.86 | 33 | 15 | 12 |
Mid Insula | - | −6.38 | −33 | 4 | 13 | −11.11 | 33 | 0 | 12 |
Post Insula | - | −12.85 | −33 | −24 | 9 | −6.02 | 42 | −11 | 7 |
Posterior STS | 21 | −17.77 | −51 | −57 | 18 | ||||
Angular gyrus | 39 | −17.12 | −45 | −69 | 30 | −9.98 | 51 | −54 | 24 |
All coordinates are described according to the Montreal Neurological Institute system, and were obtained through a conjunction analysis of data from ScaleImprov and JazzImprov.
As highly trained professional right-handed jazz pianists constitute a relatively select study population, the present study was limited to six musicians. To address the issue of a small sample size, we also utilized a multi-subject conjunction analysis to examine functional imaging data obtained from the piano improvisation experiments
Our results strongly implicate a distinctive pattern of changes in prefrontal cortical activity that underlies the process of spontaneous musical composition. Our data indicate that spontaneous improvisation, independent of the degree of musical complexity, is characterized by widespread deactivation of lateral portions of the prefrontal cortex together with focal activation of medial prefrontal cortex. This unique pattern may offer insights into cognitive dissociations that may be intrinsic to the creative process: the innovative, internally motivated production of novel material (at once rule based and highly structured) that can apparently occur outside of conscious awareness and beyond volitional control.
In jazz music, improvisation is considered to be a highly individual expression of an artist's own musical viewpoint
In comparison, the lateral prefrontal regions (LOFC and DLPFC), which were deactivated during improvisation, are thought to provide a cognitive framework within which goal-directed behaviors are consciously monitored, evaluated and corrected. The LOFC may be involved in assessing whether such behaviors conform to social demands, exerting inhibitory control over inappropriate or maladaptive performance
In light of these distinct roles, we believe that the dissociation of activity in MPFC and LOFC/DLPFC observed here during improvisation is highly meaningful. If increased activity in the MPFC serves as an index of internally motivated behavior, concomitant decreases in the LOF and DLPFC suggest that self-generated behaviors (such as improvisation) occur here in the absence of the context typically provided by the lateral prefrontal regions. Whereas activation of the lateral regions appears to support self-monitoring and focused attention, deactivation may be associated with defocused, free-floating attention that permits spontaneous unplanned associations, and sudden insights or realizations
While the results of some previous studies
It has also been suggested that deactivation of the lateral prefrontal regions represents the primary physiologic change responsible for altered states of consciousness such as hypnosis, meditation or even daydreaming
Since improvisation was also accompanied by changes in sensorimotor and limbic systems, it is tempting to speculate that these changes might be causally related, triggered in a top-down fashion by changes initiated in the prefrontal cortex. Increased activity in some of the sensory areas involved might be explained by their role in processing complex stimuli in the auditory modality. For example, the anterior temporal regions (anterior STG, MTG, and intervening STS) that were selectively activated during improvisation appear to play an integral role in processing complex features of highly structured acoustic stimuli, including music
Previous studies of music perception have reported both increases and decreases in limbic activity. Because of the presumed relationship between musical creativity and emotion, involvement of the limbic system was anticipated here. The deactivation of the amygdala and hippocampus we observed may be attributable to the positive emotional valence associated with improvisation, consistent with studies that have reported these limbic structures to be less active during perception of music that is consonant
In an intriguing neuroimaging study of musical improvisation in classically trained pianists, Bengtsson et al.
Because our experiments were performed in highly trained musicians, it remains to be clarified whether or not our findings have characterized a higher qualitative level of musical output (as opposed to that which might be produced by less skilled performers). However, the similar findings seen for both Scale and Jazz paradigms, despite the musical simplicity of the former, strongly suggest that our findings are attributable to neural mechanisms that underlie spontaneity more broadly rather than those specific to high-level musicality alone. Taken together, the consistency of findings reported here suggests that the dissociation of activity in medial and lateral prefrontal cortices is attributable to the experimentally constant feature of improvisation and may be a defining characteristic of spontaneous musical creativity.
Six right-handed, normal hearing healthy male musicians (age range 21–50 years, mean 34.2±10.4 s.d.) participated in the study. All were full-time professional musicians (either as working performers or music professors) that were highly proficient in jazz piano playing. None of the subjects had any history of neurologic or psychiatric disorders. Informed consent was obtained for all subjects, and the research protocol was approved by the NINDS/NIDCD Institutional Review Board of the NIH.
Two block-design test paradigms were used to assess musical improvisation (see Supporting Information
In the second paradigm (Jazz), a musically rich context was provided for improvisation. Prior to arrival for the scan session, all subjects received sheet music of a jazz melody (“Magnetism”, twelve-bar blues form) that was composed by one of the authors (C.J.L) to ensure novelty for the subjects (
A non-ferromagnetic piano keyboard (MagDesign, Redwood, CA) was custom-built with plastic keys and casing, which contained 35 full size piano keys, and sent out Musical Instrument Digital Interface (MIDI) information only (
All studies were performed at the NMRF Imaging Facility at the NIH. Blood oxygen level dependent imaging (BOLD) data were acquired using a 3-Tesla whole-body scanner (GE Signa; General Electric Medical Systems, Milwaukee, WI) using a standard quadrature head coil and a gradient-echo EPI sequence. The scan parameters were as follows: TR = 2000 ms, TE = 30 ms, flip-angle = 90°, 64×64 matrix, field of view 220 mm, 26 parallel axial slices covering the whole brain, 6 mm thickness. Four initial dummy scans were acquired during the establishment of equilibrium and discarded in the data analysis. 270 volumes were acquired for each subject during the Scale paradigm and 760 volumes were acquired for each subject during the Jazz paradigm. In addition to the functional data, high-resolution structural images were obtained using a standard clinical T1-weighted sequence. BOLD images were preprocessed in standard fashion, with spatial realignment, normalization, and smoothing (9 mm kernel) of all data using SPM99 software (Wellcome Trust Department of Imaging Neuroscience, London, U.K.)
For the MIDI piano data, the total number of notes played by each subject was tabulated for each condition. The range of notes from low to high was computed for each subject by analysis of the raw MIDI data. As a quantitative measure that reflected not only the absolute range of notes but also the distribution of keyboard notes played (and to a limited extent, the physical movements required), a weighted distribution of notes was calculated. The weighted distribution was computed by taking a mean of the MIDI pitch value of all notes played (in reference to the keyboard's 35-note range), weighted by the number of times each individual note was played. Paired t-tests were used to compare piano output during control and improvised conditions for both Scale and Jazz paradigms.
For fMRI analysis, data from all six subjects were entered into a group-matrix within SPM99. Fixed-effects analyses were performed with a corrected threshold of p<0.01 (or <0.001 where noted) for significance. Contrast analyses were performed for activations and deactivations across all conditions (Improv and Ctrl), and conjunction analyses were performed for results across Jazz and Scale paradigms (p<0.01 corrected). Multi-subject conjunctions for all six subjects were also performed for each paradigm. To perform the multi-subject conjunctions, individual subject contrasts (eg. [Improvisation]–[Control]) were calculated for each subject; all individual contrasts were then subjected to a conjunction analysis without Bonferrini correction (p<0.001) that identified only those areas strictly activated (or deactivated) in all subjects
Areas of activation during improvisation were revealed by standard contrast analyses, with the application of inclusive masking of contrasts for increased specificity. Contrasts for [improvisation (I)>control (C)] were masked with contrasts for [I>rest (R)], p<0.001 corrected. This inclusive masking was used to identify areas with greater net activity during [I] than [C] attributable to increased activity during [I] within each paradigm (as opposed to decreased activity during [C]). Areas of deactivation during improvisation were revealed by inclusive masking of contrasts for [C>I] with [R>I], p<0.001 corrected; ie. areas with greater net activity during [C] than [I] attributable to deactivations during [I] within each paradigm. For example, to show activations during the Scale paradigm associated with improvisation, the contrast for [ScaleImprov>ScaleCtrl] was masked inclusively with the contrast for [ScaleImprov>ScaleRest]. An analogous method was used to identify areas of activation and deactivation associated with control conditions. Conjunction analyses were used to identify commonalities shared across paradigms for each condition. For example, to show areas activated during improvisation for both Scale and Jazz paradigms, we performed a conjunction of the results for the contrasts of [JazzImprov>JazzCtrl] masked inclusively by [JazzImprov>JazzRest] and [ScaleImprov>ScaleCtrl] masked inclusively by [ScaleImprov>ScaleRest]; the same method was applied to identify common areas of deactivation across paradigms.
15s excerpt of control condition, Scale paradigm
(0.26 MB WMV)
15s excerpt of improvisation condition, Scale paradigm
(0.26 MB WMV)
30s excerpt of control condition, Jazz paradigm
(0.48 MB WMV)
30s excerpt of improvisation condition, Jazz paradigm
(0.48 MB WMV)
Multi-subject conjunction analyses for Scale and Jazz paradigms. These conjunctions reveal broad deactivation of dorsolateral prefrontal cortex for both paradigms (n = 6) as well as focal activation of the medial prefrontal cortex in Jazz (n = 5) and Scale (n = 4) paradigms. Data are presented at a statistical threshold of p<0.001 without Bonferrini correction.
(7.25 MB TIF)
The authors thank Steve Wise and Alex Martin for their review of the data and comments, Brian Rabinovitz for technical support and Jim Zimmerman for discussions going back many years. We also thank the jazz musicians who participated in the study.