Functional neuroimaging of visual creativity: a systematic review and meta‐analysis

Abstract Introduction The generation of creative visual imagery contributes to technological and scientific innovation and production of visual art. The underlying cognitive and neural processes are, however, poorly understood. Methods This review synthesizes functional neuroimaging studies of visual creativity. Seven functional magnetic resonance imaging (fMRI) and 19 electroencephalography (EEG) studies were included, comprising 27 experiments and around 800 participants. Results Activation likelihood estimation meta‐analysis of the fMRI studies comparing visual creativity to non‐rest control tasks yielded significant clusters in thalamus, left fusiform gyrus, and right middle and inferior frontal gyri. The EEG studies revealed a tendency for decreased alpha power during visual creativity compared to baseline, but comparisons of visual creativity to non‐rest control tasks revealed inconsistent findings. Conclusions The findings are consistent with suggested contributions to visual creativity of prefrontally mediated inhibition, evaluation, and working memory, as well as visual imagery processes. Findings are discussed in relation to prominent theories of the neural basis of creativity.

The neural basis of visual creativity was assessed using activation likelihood estimation (ALE) meta-analysis of fMRI studies, in addition to qualitative synthesis of findings from this and other neuroimaging modalities. We also evaluated support for existing accounts of cognitive and neural contributions to creativity, including right hemispheric dominance, PFC involvement, and the role of alpha power. A further aim was to expand on previous reviews by assessing evidence for (1) effects of participants' visual creative ability on the neural or electrophysiological correlates of visual creativity and (2) differences in the neural basis of visual creativity according to whether tasks emphasized the functionality, esthetics, or originality of generated visual solutions (Dietrich & Kanso, 2010;Gonen-Yaacovi et al., 2013).

| Search strategy
This systematic review and ALE meta-analysis followed PRISMA guidelines ) and synthesized studies recording neural activity during active generation of visual-based creative (i.e., novel and useful) ideas (Runco & Jaeger, 2012). Tasks involving only passive viewing of visual creative forms or their retrieval from memory were not included. Convergent thinking, problem-solving or insight tasks, which typically have a single, fixed solution, can engage creative thinking (Abraham, 2013). Divergent thinking or open-ended tasks with multiple possible solutions, however, typically provide a closer approximation to the creativity involved in visual design, art, and innovation (e.g., Ellamil et al., 2012;Kowatari et al., 2009), and also meet standard definitions of creativity (Runco & Jaeger, 2012).
F I G U R E 1 Flowchart of article selection, following PRISMA guidelines. Adapted from Moher et al. (2009). fMRI, functional magnetic resonance imaging; EEG, electroencephalography The article selection procedure is summarized in Fig. 1. Using the above search terms, 3489 records were identified and 46 were identified through reference lists of relevant studies. Following deduplication and screening for inclusion criteria (see Table 1), 26 articles, comprising 27 experiments, were included in the review, of which six fMRI studies were included in the ALE meta-analysis. No limitations were placed on the date of publication.

| Quality assessment
Included experiments were rated according to quality assessment criteria adapted from Whiting, Rutjes, Reitsma, Bossuyt, and Kleijnen's (2003) QUADAS quality assessment tool: (1) clear description of participant selection criteria and demographics; (2) visual creativity task compared against a non-rest/fixation control task (hereafter, "control task"); (3) sufficient detail on task procedure for reproducibility; (4) sufficient detail on the neuroimaging procedure and outcome measures for reproducibility; (5) sufficient information on analyses and results for reproducibility; (6) conclusions justified based on analyses, for example, appropriate multiple comparisons corrections; (7) no substantial confounds between groups/conditions. Criterion (2) was selected as comparisons against a constrained non-rest control task that elicits overlapping processes are thought to better facilitate isolation of processes that are unique to the task of interest than an unconstrained rest/fixation condition (Abraham, 2014;Lazar, 2008). For each experiment, a score of 0 (criterion not met) or 1 (criterion met) was assigned for each criterion, and the percentage of criteria met was calculated. Scores are summarized in Table S1.

| Activation likelihood estimation
A coordinate-based ALE meta-analysis was conducted using Brainmap GingerALE 2.3 (http://www.brainmap.org/ale). ALE metaanalysis uses peak coordinates from published studies to calculate brain regions in which the convergence across studies is greater than expected by chance if the included foci were independently distributed (Eickhoff, Bzdok, Laird, Kurth, & Fox, 2012;Eickhoff et al., 2009). Each included activation focus is modeled as the center of a 3D Gaussian distribution, the full width at half maximum of which is determined by the study's sample size (Eickhoff et al., 2009).
Modeled activation (MA) maps are calculated by computing the maximum across the Gaussian distribution of each focus (Turkeltaub et al., 2012). The ALE map resulting from combining the MA maps is then compared against an ALE null distribution map. A random effects model was employed (Eickhoff et al., 2009), and significance thresholds on the ALE scores were set via cluster-level inference . A cluster-level threshold of p < .05 and cluster-forming threshold of p < .001 were used to set the minimum cluster volume at 192 mm 3 , via 1000 permutations. The smaller, more conservative mask size was selected.
The meta-analysis was conducted in MNI space. In GingerALE 2.3, anatomical labels were assigned to ALE peaks which surpassed the voxel and cluster-level thresholds using the Talairach Daemon, after transformation of significant coordinates using icbm2tal (Lancaster et al., 2007).

| RESULTS
Included studies comprised 7 fMRI and 19 EEG experiments. No NIRS, MEG, ERP, or PET studies met inclusion criteria. Information on participants, creative and control tasks, analyses, and results are summarized in Table 2 for fMRI studies, and Table 4  Whole-brain: no differences between experts and novices in design or control activity (vs. baseline); no differences between design and control tasks in experts or novices. ROIs in PFC and parietal cortex: R > L in experts but not novices. In experts, R versus L difference in PFC positively associated with originality of pen designs. (continues)

| Study characteristics
Of the 26 reviewed articles, 10 have to our knowledge not been included in previous systematic reviews or meta-analyses. The 26 articles comprised 27 experiments and around 800 participantsthis is approximate as the studies of Bechtereva and Nagornova (2007) and Nagornova (2007), of Petsche (1996) and Petsche, Kaplan, Von Stein, and Filz (1997), and of Volf, Tarasova, and Razumnikova (2010a) and    , Park et al., 2015;Saggar et al., 2015) or distinct (e.g., Volf et al., 2010a) phases. Tasks included those in which solutions must fulfill a specified function (25.9%), those emphasizing the originality/fluency of solutions (51.9%; these are combined as typically task instructions emphasized both criteria, for example, "generate as many original solutions as possible"), those emphasizing the esthetics of solutions (7.4%), and tasks giving no instructions as to the desired characteristics of solutions (14.8%). On average, studies satisfied 67% of quality criteria (SD = 21; range 14-100%; Table S1).

| Functional magnetic resonance imaging study characteristics
The participants, procedure, and main findings of the reviewed fMRI studies are summarized in participants were asked to draw visual representations of presented words (e.g., "graduate," "snore").
In the final study (Aziz-Zadeh et al., 2013), the desired features of visual solutions were not emphasized, and brain activity was recorded from architects while they mentally combined three presented shapes to create an image. Activity during this task was compared with activity during a mental rotation task.

| ALE meta-analysis findings
Six fMRI studies (Aziz-Zadeh et al., 2013;Ellamil et al., 2012;Gilbert et al., 2010;Huang et al., 2013;Park et al., 2015;Saggar et al., 2015), including 77 foci from seven contrasts were included in the ALE meta-analysis (Section 2.3). All foci were associated with greater activity during visual creativity compared to control conditions. An additional study (Kowatari et al., 2009) met inclusion criteria but as no differences were found between visual creativity and control tasks in experienced or novice designers, no foci were included in the meta-analysis. Three included studies employed tasks emphasizing the function of solutions, two emphasized originality/fluency, and one had no clear focus-the numbers of studies in each of these categories were insufficient for analysis of effects of task focus. The meta-analysis revealed seven clusters that surpassed the significance threshold (see Section 2.3, for thresholding and analysis). Results are summarized in Table 3 and significant clusters are displayed in Fig. 2.
Regions showing significant ALE activity included thalamocortical nucleus, right middle and inferior frontal gyri, cingulate gyrus, and left fusiform gyrus.
For each significant ALE cluster, only two studies from a subset of three (Ellamil et al., 2012;Park et al., 2015;Saggar et al., 2015) contributed foci which fell within the cluster boundaries. This meets the previously suggested quality criterion of a contribution of 33% of included studies for reporting ALE maxima (Brooks et al., 2012;Van der Laan, De Ridder, Viergever, & Smeets, 2011), and further foci from other studies which were out with the cluster boundaries may still have contributed to their significance (Brooks et al., 2012). However, due to the possibility that only a minority of studies contributed to the meta-analysis findings, a qualitative synthesis of fMRI findings is reported below.

| Electroencephalography study characteristics
Nineteen EEG studies comprising 20 experiments were reviewed.
The main findings are summarized in Table 4 Most experiments employed measures of EEG power (25%), coherence (35%), or both (35%). EEG power refers to the amplitude of a particular frequency band, while coherence, or phase synchrony, instead reflects functional cooperation between cortical regions.
Studies varied substantially in the control tasks employed, and the focus of key contrasts. Several compared activity during visual creativity to a verbal creativity or verbal and/or visual control task, often involving memory or convergent thinking (e.g., Jausovec, 2000;Nagornova, 2007 Figure 3 summarizes the numbers of studies where a substantial majority of significant effects on EEG power across electrodes were
These outcomes are summarized for each of the main contrast types (Section 3.3.1).

| Visual creativity versus baseline
In the low-frequency delta and theta bands and the lower and upper alpha bands, a consistent pattern of decreases in EEG power during visual creativity compared to baseline fixation or rest-task-related

| Visual creativity versus control
Several studies compared activity during visual creativity to that during verbal creativity (Jausovec & Jausovec, 2000;Razumnikova et al., 2009Razumnikova et al., , 2010, or during verbal and/or visual convergent problem-

| Comparisons of high-and low-creativity participants
In the alpha band, greater power (Jausovec, 2000;Sviderskaya, Taratynova, & Kozhedub, 2006)  Findings of power changes during visual creativity versus baseline are displayed in blue; power changes versus control tasks in red; differences between high-and low-creativity participants in green; and differences between production of original versus standard images in purple reported opposite effects of creative ability in male and female participants. In males, high creativity (based on originality scores) was associated with greater upper alpha TRD compared to low-creativity participants, while females of low creativity showed greater TRD compared to high-creativity females. A tendency for greater TRD in posterior compared to anterior sites was also reported-in men, this was exhibited by high-creativity individuals only in the lower and upper alpha bands. However, as Volf et al. (2010a) divided participants based on median splits of originality scores performed separately for males and females, it is unclear if originality scores were comparable between high-and low-creativity men and women, and so differential effects of creativity in each group must be interpreted with caution.
In the theta and beta bands, Sviderskaya et al. (2006)

| Findings-Electroencephalography coherence
The numbers of studies showing predominant coherence increases, decreases, or no clear pattern for the main contrast types are displayed in Fig. 4.

| Visual creativity versus baseline
For the low-frequency delta and theta bands, a tendency toward wide- indicating that individual differences in creativity or strategy use, in addition to task differences contribute to discrepancies between studies. Across both high-and low-creativity groups, however, task-related upper alpha coherence decreases were observed (Volf et al., 2010b).

| Visual creativity versus control
Sviderskaya (2011b) reported overall increases in delta and alpha coherence during visual creativity compared to a visual convergent thinking task. Bechtereva and Nagornova (2007)  figures. This was largely driven by right hemisphere intrahemispheric increases in female participants.

| Comparisons of high-and low-creativity participants
Both studies examining effects of creative ability on delta coherence during visual creativity versus baseline reported greater coherence in high-compared to low-creativity individuals (Bhattacharya & Petsche, 2005;Sviderskaya et al., 2006). Bhattacharya and Petsche (2005) found that these coherence differences were focused on posterior occipitotemporal sites, and took the form of inter-and intra- (2010b) observed greater intrahemispheric coherence in high-versus low-creativity participants due to the tendency of the former group to show increased intrahemispheric coherence versus baseline, while the latter showed reduced task-related coherence versus baseline.
The picture was less consistent for the higher frequency ranges.
For upper alpha, there were two reports of greater coherence in highversus low-creativity participants (Sviderskaya et al., 2006;Volf et al., 2010b), one of reduced coherence (Jausovec, 2000), and Petsche et al. (1997) reported no differences between groups. Bhattacharya and Petsche (2005) reported overall coherence reductions at frontal sites in artists versus novices across the alpha band. Sviderskaya et al. (2006) and Bhattacharya and Petsche (2005) reported increased coherence in high-compared to low-creativity participants across the beta range, although in the latter study this effect was restricted to right temporal sites, and reduced beta coherence was observed in artists versus novices over frontal sites. This study also provided the only examination of creative ability effects on gamma coherence, reporting reduced coherence over frontal sites in artists versus novices. Sviderskaya (2011a) and Sviderskaya et al. (2006) examined effects of creative ability on spatial synchronization (SS) and/or spatial disordering (SD), measures of coherence based on linear and nonlinear relationships, respectively. Both found that artists compared to novices showed greater task-related increases in coherence versus baseline, particularly over right frontal and occipital sites.

| Trends across contrasts
Collapsing across the four main contrast types, the percentages of studies reporting mainly coherence increases, decreases, or neither did not differ according to frequency band (p = .35, Fisher's exact test). There was no consistent trend toward coherence increases or decreases in any frequency band, aside from the delta band where 80% of studies reported increased coherence during visual creativity.
Visual creativity-related effects on alpha coherence did not differ according to the contrast employed, for the lower or upper alpha band (p = .75; p = .86, Fisher's exact test).

| Hemispheric lateralization
No clear pattern of laterality of power or coherence effects emerged.
Most studies examining EEG power effects reported bilateral effects, in several studies of further bilateral occipitotemporal regions (e.g., inferior temporal gyrus, lateral occipital cortex), is consistent with predictions that visual creativity compared to control tasks is associated with greater visual processing, including visual imagery (e.g., Kosslyn & Thompson, 2003).

| Overview of electroencephalography findings
Where visual creative tasks were compared against baseline fixation, the theta and lower and upper alpha frequency bands consistently showed reduced power, while higher frequency beta and gamma bands typically showed increased power relative to baseline.
The theta band findings are at odds with reports that this frequency band typically shows increased power with increasing task demands (Klimesch, 1996;Klimesch, 1999). The studies reporting theta TRD (Razumnikova et al., 2009;Sviderskaya, 2011a;) did not fully explain procedures for collecting baseline data (see Table S1; Section 4.5), and so it is possible that this result is due to lengthy baseline rest periods resulting in high levels of cognitive activity and thus theta power at baseline . The findings of alpha TRD are consistent with a role of semantic and attentional processes during visual creativity (Klimesch, 2012). Increased beta power is indicative of increased alertness and active concentration during visual creativity (e.g., Gola, Kamiński, Brzezicka, & Wróbel, 2012;Klimesch, 1999). Such findings relative to fixation are, however, uninformative as to the neural mechanisms specific to visual creativity, as such changes may be observed in any task requiring greater cognitive resources than fixation.
Despite this, relatively few EEG studies directly compared activity during visual creativity and during appropriate control tasks, and those that did revealed largely inconsistent and contradictory findings in the alpha and beta bands. A contributing factor in these inconsistencies is likely the wide variety of control tasks employed,

| Prefrontal cortex functions
The proposed contribution of PFC functions to creativity has received consistent support from neuroimaging studies to date (e.g., Dietrich, 2004;Dietrich & Kanso, 2010;Gonen-Yaacovi et al., 2013). Accounts of creative cognition have proposed a role of PFC-mediated executive functions in creative idea generation, including updating of working memory, inhibition of irrelevant ideas, monitoring, and selection of generated solutions (Dietrich, 2004;Mumford et al., 2012). Previous The reviewed EEG studies did not employ source localization, but the observation that visual creativity-related coherence changes were often focused on frontal sites is again consistent with a contribution of frontal lobe functions (Dietrich, 2004). Foci of alpha and beta coherence increases included frontal sites in most studies examining this outcome (Bechtereva & Nagornova, 2007;Bhattacharya & Petsche, 2005;Jausovec & Jausovec, 2000;Sviderskaya, 2011b), indicative of increased functional connectivity between frontal regions and further cortical sites (Thatcher et al., 1986). This may involve top-down modulation of downstream processes including perceptual, mnemonic, or attentive processes (e.g., Gazzaley et al., 2007). Petsche Petsche et al. (1997) reported reduced interhemispheric frontal alpha coherence during visual creativity, interpreting this as increased independence of left and right PFC functions. Some studies also reported that task-related power increases or decreases were particularly evident at frontal sites (Jaarsveld et al., 2015;Jausovec & Jausovec, 2000). Further behavioral and neuroimaging studies including EEG studies employing source localization in addition to appropriate control tasks are necessary to establish the subregions of PFC and associated cognitive processes contributing to visual creativity. Mihov et al.'s (2010) meta-analysis of EEG and fMRI studies of creativity reported right hemispheric dominance in visual and verbal creativity This, however, contrasts with other reviews and meta-analyses (Dietrich & Kanso, 2010;Gonen-Yaacovi et al., 2013;Wu et al., 2015) which reported no evidence of lateralization of creativity. Here, the Of the 20 reviewed EEG experiments, only a minority reported effects of hemisphere (Section 3.3.4). Bhattacharya and Petsche (2005) observed greater task-related coherence increases in the theta, alpha, and beta bands in the right compared to the left hemisphere, but Kozhedub et al. (2007) in contrast reported greater probability of task-related changes in alpha coherence in the left compared to right hemisphere. In the majority of EEG studies, visual creativity-related effects on power and coherence were largely bilateral, with no evidence of hemispheric dominance. No evidence was revealed of the alpha power asymmetry effects that have been associated with positive versus negative affect (Davidson, 1992) or response inhibition (Wacker, Chavanon, Leue, & Stemmler, 2010). In addition, the above studies reporting hemispheric effects did not assess whether these effects remained when comparing visual creativity to matched control tasks, and so it is unclear whether such effects are specific to visual creativity.

| Hemispheric lateralization
Taken together, the findings of the current review do not provide support for theories of hemispheric lateralization of visual creativity.

| Role of semantic and episodic memory processes
A number of the left-lateralized regions identified in visual creativity contrasts in the reviewed fMRI studies have been implicated in semantic retrieval. A meta-analysis of 120 functional neuroimaging studies (Binder, Desai, Graves, & Conant, 2009) found left MFG, SFG, and IFG and left inferior parietal lobe to be involved in semantic processing, regions which showed greater activity during visual creativity compared to control tasks in several of the reviewed fMRI studies (Aziz-Zadeh et al., 2013;Ellamil et al., 2012;Huang et al., 2013;Park et al., 2015;Saggar et al., 2015). Left IFG, particularly pars orbitalis, has been consistently associated with semantic processing and retrieval (Binder et al., 2009;Liakakis, Nickel, & Seitz, 2011), and supports controlled access to conceptual representations (Badre, Poldrack, Paré-Blagoev, Insler, & Wagner, 2005). This region showed greater activity during visual creativity compared to control tasks in several studies (Aziz-Zadeh et al., 2013;Ellamil et al., 2012;Huang et al., 2013;Saggar et al., 2015), but was not identified in the ALE meta-analysis. Activity during visual creativity in regions which have been linked to semantic processing does not directly support a role of the latter in visual creativity (Poldrack, 2006), but such a role is consistent with proposals that semantic retrieval and association are core components of creative ideation (Abraham & Bubic, 2015;Beaty et al., 2014;Mednick, 1962;Mumford et al., 2012).
Episodic memory, memory for personally experienced events bound with context (Tulving, 1983), is thought to facilitate generation of creative ideas through a constructive process involving elements of previously experienced events Runco & Chand, 1995). Consistent with this, two of the reviewed fMRI studies of visual creative ideation (Ellamil et al., 2012;Park et al., 2015), in addition to studies of verbal creative ideation (e.g., Fink et al., 2009) reported greater activity during creative tasks in the hippocampus and parahippocampal cortex, regions strongly associated with mnemonic processing (e.g., Dickerson & Eichenbaum, 2010). The mediodorsal thalamic nucleus region revealed in the fMRI meta-analysis has also been linked to recollection and familiarity in episodic memory (Zola-Morgan & Squire, 1993;Zoppelt, Koch, Schwarz, & Daum, 2003), and is thought to relay inputs to and from hippocampal and prefrontal memory processing regions (Markowitsch, 1982;Xu & Sudhof, 2013).

| Visual imagery and visual processing
ALE meta-analysis revealed activity for the contrast of visual creativity versus control tasks in the left fusiform gyrus. The majority (5/7) of the reviewed fMRI studies reported greater activity during visual creativity compared to control tasks in this and further occipitotemporal regions, including lateral and middle occipital cortex and middle and inferior temporal gyri (Aziz-Zadeh et al., 2013;Ellamil et al., 2012;Huang et al., 2013;Park et al., 2015;Saggar et al., 2015). These findings are consistent with a greater role of processing of visual information during visual creativity. As the idea generation phase of each fMRI study involved visual input, whether verbal instructions (Gilbert et al., 2010;Saggar et al., 2015), images/image fragments (Aziz-Zadeh et al., 2013;Huang et al., 2013;Kowatari et al., 2009), or sketches drawn by the participant (Ellamil et al., 2012), this activity may simply reflect perceptual and conceptual processing of visual input (Cowell, Bussey, & Saksida, 2010;Tyler et al., 2013). However, a further, not mutually exclusive possibility is that greater visual cortical activation is associated with greater engagement of visual imagery processes. Visual imagery engages many of the same or highly similar regions of occipitotemporal cortex as visual perception, including bilateral inferior and middle temporal gyri and middle occipital cortex (Ganis, Thompson, & Kosslyn, 2004;Ishai, Haxby, & Ungerleider, 2002), regions that were identified in several of the reviewed fMRI studies.
The left fusiform gyrus region revealed in the meta-analysis has been repeatedly linked to visual imagery (Ganis et al., 2004;Kosslyn & Thompson, 2003). Consistent with suggestions that visual creativity, in particular visual design, engages manipulation of visual imagery, fMRI meta-analyses have found overlapping left fusiform gyrus regions to be engaged in mental rotation (Tomasino & Gremese, 2015;Zacks, 2008). An overlapping region has also been implicated in retrieval of the semantic representations required to support visual imagery (Kan et al., 2003). The left lateralization of the observed fusiform activity is consistent with studies reporting that visual imagery predominantly engages the left hemisphere (D'Esposito et al., 1997;Sack, Camprodon, Pascual-Leone, & Goebel, 2005), but as none of the reviewed studies formally compared effects in corresponding regions of left and right hemispheres, this account is not directly supported.
When visual creative tasks were compared to baseline fixation, a consistent pattern of lower and upper alpha power decreases was observed. This is consistent with greater cortical activation, and greater semantic and attentional processing during visual creativity versus fixation (e.g., Klimesch, 2012). Power changes versus baseline, however, provide a limited contribution to understanding of the neural basis of visual creativity, as they do not inform as to whether this response is specific to visual creativity (Arden et al., 2010)-similar patterns may emerge in response to any number of other tasks that are more cognitively demanding than fixation. Reduced alpha power compared to baseline has, for example, been elicited during working memory (Stipacek, Grabner, Neuper, Fink, & Neubauer, 2003), recognition (Dujardin et al., 1993), and visual classification (Pfurtscheller & Klimesch, 1990). Such findings along with early reports of reduced alpha power simply when eyes are open compared to closed (e.g., Klimesch, 1999) have consolidated the view that alpha suppression reflects cortical activation. Fink and colleagues have, however, consistently observed alpha power increases during verbal creative ideation (e.g., , and a selective review by Fink and Benedek (2014) reported overall support for a role of alpha TRS in creative ideation. The majority of the evidence reported by Fink and Benedek (2014), however, also referred to studies of verbal ideation, and so these contradictory findings could be reconciled if inhibitory processes, manifested by alpha TRS, are more often engaged during verbal compared to visual creativity, the latter involving greater semantic and attentional processing.
Three of the four studies comparing alpha power in participants of high and low creativity reported increased lower and upper power in the former group. However, due to the small number of studies and as these increases reflected both reduced TRD (Petsche et al., 1997) and increased absolute power (without reference to baseline; Jausovec, 2000;Sviderskaya et al., 2006), it is difficult to arrive at a

| Methodological issues in reviewed studies
The qualitative synthesis of EEG studies revealed relatively few consistent findings, and despite several significant clusters emerging in the ALE meta-analysis, findings of the fMRI studies also differed substantially. This lack of consistency may stem from substantial heterogeneity in the visual creative and control tasks, contrasts conducted, and outcome measures recorded (see Amabile, 1983). Even where the same creative task was employed, for example, TTCT-IF, it was compared against a variety of control tasks, ranging from simple line tracing to more cognitively demanding visual and verbal problem-solving and memory tasks. Evidence of a common neural or electrophysiological basis of visual creativity may be obscured by comparisons against tasks eliciting widely differing cognitive processes (Arden et al., 2010).
Tasks also differed in their focus, with visual design tasks highlighting the functionality of generated solutions (e.g., Ellamil et al., 2012;Gilbert et al., 2010;Kowatari et al., 2009); artistic tasks emphasizing esthetics (e.g., Bhattacharya & Petsche, 2005;Petsche, 1996); and others emphasizing the originality or fluency of solutions (e.g., Kozhedub et al., 2007;Volf et al., 2010a). Greater consistencies in the neural or electrophysiological correlates of visual creativity may be detectable by subdividing studies according to these goal-related factors, that is, tasks requiring generation of solutions that are (1) functional, (2) esthetically pleasing, or (3) original. However, heterogeneity in procedures, populations studied, contrasts conducted, and outcome measures recorded meant that such subdivisions were unfeasible here due to low numbers of comparable studies within each category.
A further key issue is that of the timing and duration of sampling of neural activity associated with visual creativity. Most reviewed studies recorded and averaged neural activity across the duration of the visual creativity task, but in a subset of studies (e.g., Aziz-Zadeh et al., 2013;Gilbert et al., 2010;Jaarsveld et al., 2015) participants were asked to signal when the task was complete, and activity was averaged from the start of the task until the response. Both methods are likely to capture the cognitive and neural processes involved in visual creative ideation, and likely also idea evaluation, but due to the long sampling periods (typically ~30 s) are likely also to include further cognitive processes both related and unrelated to visual creativity, for example, comprehension of task instructions, maintenance of visual representations, default mode activity (Fink, Benedek, Grabner, Staudt, & Neubauer, 2007;Fox & Raichle, 2007), potentially reducing the signal to noise ratio, and ability to detect processes specific to visual creativity (Abraham, 2013).

| Quality assessment
Quality assessment of the reviewed studies (Section 2.2, Table S1) revealed that most did not meet all quality criteria. Many did not provide complete descriptions of participant selection and demographic information (41%), task procedure (33%), neuroimaging procedure and outcome measures (7%), and analyses and results (15%). This not only precludes replication, but also leads to difficulties in directly comparing findings across studies (Whiting et al., 2003). A further critical issue is that 37% of the 27 experiments did not conduct appropriate multiple comparisons corrections, or in the case of EEG studies, correction for violation of sphericity, limiting the reliability of reported findings.
Lack of controls in 48% of experiments of factors such as task difficulty or duration between experimental and control tasks (e.g., Jausovec, 2000;Nagornova, 2007) introduced further potential confounds. In 60% of EEG studies, visual creativity was simply compared against baseline fixation/rest and not a matched non-rest control task, leading to the inability to infer whether effects are specific to visual creativity or are observed during multiple cognitive processes.
Another difficulty in synthesizing results across EEG studies stemmed from differences in outcome measures-several reported differences in raw measures of power, while others reported task-related power corrected for baseline power.

| Future directions
To more clearly establish the neural basis of visual creativity, it is necessary to address the above methodological issues and ensure the quality criteria outlined in Section 2.2 are met. It is important to introduce measures to ensure control of confounds between visual creative and control tasks (Abraham, 2013). Greater standardization of the control tasks employed or use of several control tasks within the same sample will better enable identification of commonalities in neural activity between studies and between visual creative tasks.
In fMRI studies, examination of functional overlap between regions identified in contrasts of visual creativity against multiple appropriate control tasks would enable identification of regions that are reliably engaged in, and are specific to, visual creativity.
It is also important to acknowledge that visual creativity is a composite, nonunitary construct and likely consists of multiple distinct cognitive and neural processes (Dietrich & Kanso, 2010)-a common neural basis may not be readily detectable. The mechanisms underlying visual creativity may differ according to task-specific features such as focus on (1)  The current meta-analysis revealed little evidence of overlap in the cortical regions engaged compared to Boccia et al.'s (2015) metaanalyses of musical and verbal creativity, aside from an overlapping region of left medial frontal gyrus (BA32) here and in the musical creativity meta-analysis. This may be due to lack of power in the current meta-analysis due to small numbers of studies, but also reinforces Boccia et al.'s (2015) findings of domain-specific as well as domaingeneral cortical contributions to creativity. However, to directly contrast visual creativity with other forms of creative ideation it will be necessary for future studies to directly compare visual and nonvisual creativity within the same participants. A small number of the reviewed EEG studies reported power and/or coherence effects versus baseline of similar magnitude and in the same direction for both visual and verbal divergent thinking (Jausovec & Jausovec, 2000;Molle et al., 1999;Razumnikova et al., 2009Razumnikova et al., , 2010. However, as these findings refer to baseline contrasts, comparable effects may be observed with any number of tasks requiring cognitive effort (Section 4.3.5).
Consistent with Arden et al.'s (2010) suggested psychometric approach to creativity, given suggestions that visual creativity relies on semantic, executive, and visual imagery processes, these claims could be evaluated by assessing whether ability in these cognitive domains predicts visual creative ability, or neural activity elicited during visual creativity. Such associations could be compared across multiple domains of creativity and across task foci.
Machine learning algorithms (see Brouwer, Zander, van Erp, Korteling, & Bronkhorsst, 2015;Mwangi, Tian, & Soares, 2014, for reviews) offer promising avenues in identification and classification of EEG and fMRI features associated with visual creativity compared to control tasks, or in classification of features associated with visual creativity emphasizing functionality, esthetics, and originality. In fMRI, multivariate pattern recognition algorithms may aid in identifying not only which cortical regions show involvement in visual creativity, but also which regions show evidence of representing visually generated creative ideas, and which regions differentiate between the generation of functional, esthetic, and original visual solutions (Mur, Bandettini, & Kriegeskorte, 2009).
The inherent difficulty in temporal isolation of the processes directly relevant to creativity has been noted (Abraham, 2013), leading Fink et al. (2007) to suggest a method via which participants indicate the moment of idea generation, and the activity immediately preceding the button press is examined. The issue of selection of an arbitrary sampling duration is not fully avoided using this method, but in future studies, adoption of a common method of isolating activity associated with creative ideation will aid comparability of findings across studies.
These suggestions for future research are summarized below: 1. Ensure greater between-study consistency in the nature of creative and control tasks employed, and adequate control of confounds between creative and control tasks.
2. Directly examine effects of task focus (e.g., function, esthetics, originality) on the neural basis of visual creativity.

3.
Directly contrast and compare the neural and cognitive basis of visual compared to verbal and musical creativity (Arden et al., 2010).

4.
Capitalize on advancements in machine learning and multivariate pattern analysis techniques to identify features associated with representation of visual creative ideas.

5.
Employ standard methods across studies of isolating the time period to be examined, for example, following Fink et al.'s (2007) approach of examining neural activity directly preceding pressing of an "idea button."

| CONCLUSIONS
Meta-analysis of six fMRI studies revealed, across studies, greater activity in regions of right middle and inferior frontal gyri during visual creativity compared to non-rest control tasks, and EEG power and coherence effects during visual creativity were often focused on frontal sites. These findings are consistent with theories of creative cognition that propose an integral role of PFC functions including working memory, inhibition of task-irrelevant information, selection among competing representations, and monitoring and evaluation of solutions. Meta-analysis of fMRI studies and qualitative synthesis of fMRI and EEG studies also supported a role of occipitotemporal regions in visual creative task performance, consistent with a role of increased visual processing, including visual imagery and visual image manipulation, during visual creativity.
Neither fMRI nor EEG studies provided clear support for the notion of right hemispheric dominance in visual creativity, although the meta-analysis findings demonstrated greater cross-study consistency in the right compared to left PFC regions engaged. Synthesis of the EEG studies did not provide consistent support for suggestions that either increases or decreases in alpha power contribute to visual creativity.