Dystonia is the third most prevalent movement disorder worldwide, characterized by abnormal postures or movements in specific regions of the body resulting from persistent muscle contractions (1, 2). The global incidence of dystonia is approximately 30.85 per 100,000 individuals, significantly affecting patients’ quality of life and imposing considerable social and economic burdens (3). Idiopathic dystonia (ID), a common subtype of dystonia, occurs independently of other neurological or genetic conditions (4). While dysfunction in the basal ganglia or cerebello-thalamo-cortical circuits is widely regarded as the primary cause of ID (5, 6), its precise neuropathological and physiological mechanisms remain incompletely understood (7).

Voxel-based morphometry (VBM) facilitates automated comparisons of gray matter (GM) volume or density in T1-weighted magnetic resonance imaging (MRI) scans at the whole-brain level between different groups (8). Unlike the traditional region-of-interest (ROI) approach, VBM offers greater objectivity and efficiency (8, 9). Previous studies have investigated GM alterations in ID using VBM, but the results have been inconsistent and occasionally contradictory. For instance, Chirumamilla et al. and Bianchi et al. reported increased GM volume in the anterior cingulate cortex in patients with ID (10, 11), whereas Piccinin et al. found a decrease in GM in the same region (12). Additionally, some studies have reported no significant GM abnormalities in ID (13–17). Therefore, further research is required to identify consistent and reliable patterns of GM alterations, which could provide deeper insights into the pathophysiological mechanisms underlying ID.

Coordinate-based meta-analysis (CBMA) enables the integration of multiple neuroimaging studies, facilitating the derivation of more generalizable and robust findings (18, 19). To date, five CBMAs have summarized whole-brain VBM studies on ID and its subtypes, yet inconsistencies in their results have been observed (20–24). This discrepancy may be attributed to the use of different CBMA algorithms. Zheng et al. and Wu et al. employed activation likelihood estimation (ALE) techniques (21, 24), whereas Huang et al. and Zhang et al. utilized anisotropic effect size-based signed differential mapping (AES-SDM) techniques (20, 22, 23). In contrast to ALE, AES-SDM can incorporate both significant and non-significant results along with their effect sizes but adopts a less stringent statistical approach (25, 26). Disparities in literature search strategies, time frames, and the focus on specific ID subtypes may have further contributed to these inconsistencies. A comprehensive summary of the information from these five CBMA studies is provided in Table 1.

In line with the latest guidelines and recommendations, several aspects of the previous five CBMAs require improvement (27–29). First, the studies by Gellea et al. and Pantano et al. should be excluded, as they employed ROI techniques rather than whole-brain approaches in their VBM analyses (30, 31). Additionally, the inclusion of studies that employed small-volume correction (SVC) should be reassessed (32–36), as this method may exaggerate results in specific brain regions and introduce bias in region selection (27). Furthermore, Zheng et al. applied an older version of ALE with false discovery rate (FDR) correction (24), which has been demonstrated to be potentially suboptimal for CBMA (37). Moreover, the inclusion of ALE analyses with a limited number of studies (9 and 14 studies, respectively) (21, 24), while a minimum of 17 studies is recommended for ALE analysis to ensure adequate statistical power (38). These factors likely compromised the robustness of the CBMA results. Notably, the number of VBM studies focusing on ID has increased in recent years. Thus, an updated CBMA of existing VBM studies on ID is crucial for identifying consistent and robust GM alterations.

The latest version of CBMA, seed-based d mapping with permutation of subject images (SDM-PSI), applies threshold-free cluster enhancement (TFCE)-based family-wise error (FWE) correction to control for multiple comparisons (39). This method is advantageous because it simultaneously addresses both FWE and cluster-level significance without requiring a predefined cluster size threshold, thus preserving statistical sensitivity while reducing the likelihood of false positive findings. It has demonstrated effectiveness in a range of neurological and psychiatric disorders, including schizophrenia and Parkinson’s disease, and has exhibited high statistical efficacy (40, 41). Therefore, we conducted an updated CBMA to identify consistent and robust GM alterations in ID, employing SDM-PSI in line with current guidelines and recommendations (27–29). Additionally, meta-regression analyses were performed to examine the potential influence of participant demographic and clinical characteristics on GM alterations.

This updated CBMA did not exhibit any consistent GM alterations in patients with ID compared to HC subjects, with jackknife sensitivity analysis confirming the robustness of the findings. Furthermore, meta-regression analysis revealed a significant impact of disease duration on GM in the right insula.

It is noteworthy that our findings were inconsistent with those in previous CBMAs (20–24), which may be attributed to the application of the latest SDM-PSI algorithm alongside more stringent statistical methods, and adherence to contemporary guidelines that excluded studies using ROI and SVC analyses. Additionally, our CBMA included a larger cohort comprising 32 original VBM datasets and 840 participants, thus enhancing the reliability of the results. Consequently, the absence of consistent GM alterations may suggest that GM is not a dependable neuroimaging biomarker for ID.

One potential explanation for the inconsistent GM abnormalities observed in patients with ID may be the absence of true GM alterations within this population. Previous autopsy studies and animal models have demonstrated that ID does not present with structural brain anomalies (64–66). Recent functional MRI studies have revealed significant abnormalities across several brain regions in patients with ID, including the primary motor cortex, supplementary motor area, cerebellum, thalamus, and putamen (67–71). Additionally, several studies have demonstrated a correlation between the severity of clinical symptoms and functional abnormalities in specific brain regions (68, 69, 71). Furthermore, some research suggests that symptomatic treatments, such as botulinum toxin injections or deep brain stimulation, can modulate abnormal brain functions (72–75). Collectively, these findings suggest that ID may be primarily characterized as a functional brain disorder that does not necessarily involve structural abnormalities.

The inconsistency in identifying GM alterations in ID may be attributed to the heterogeneity among participants. As illustrated in Table 2, the 32 datasets include 13 distinct subtypes of ID, highlighting significant diversity in demographic and clinical characteristics. Specific subtypes of ID are often associated with distinct patterns of GM abnormalities (76). For instance, Tomic et al. and Ramdhani et al. observed that, compared to non-task-specific dystonia (blepharospasm and cervical dystonia), task-specific dystonia (laryngeal dystonia and writer’s cramp) was associated with increased GM volume in the primary somatosensory cortex, middle frontal gyrus, temporal lobe, and occipital lobe (56, 63). Moreover, the varied motor symptoms observed in patients are frequently associated with increased GM volume in specific brain regions, providing insights into the potential mechanisms underlying GM alterations in ID (64, 77). As a result, GM alterations in ID present diverse patterns, with the disorder’s inherent heterogeneity likely contributing to the lack of consistent GM findings.

The meta-regression analysis indicated that disease duration has an impact on GM volume in the right insula, a region crucial for emotional perception and sensory processing (78, 79). Individuals with ID frequently experience non-motor symptoms such as chronic pain, anxiety, and depression (80–82). These symptoms have been linked to changes in brain structure in various psychiatric and neurological disorders, often correlating with disease duration (83–85). This suggests that the alterations in GM volume in the insula may reflect an adaptive response to these non-motor symptoms in individuals with ID. Although no significant modulatory effects of age or gender on GM volume were observed, caution should be exercised in interpreting this finding, as it is based on study-level rather than individual-level data. Age and gender are well-established factors that are known to influence brain structure (86–89). Other potential confounding variables, such as handedness (90–92), disease severity (12, 54, 57, 93), age of onset (93), medication history (12, 57, 63), and education level (94), may also contribute to discrepancies in brain structure. Unfortunately, the lack of sufficient data from the original studies limits the ability to systematically investigate these potential variables. Consequently, the demographic and clinical heterogeneity of participants in VBM studies may account for the failure to identify consistent GM alterations.

In addition to participant heterogeneity, the inconsistency in GM alterations may also arise from variations in imaging protocols. As outlined in Table 3, the 27 VBM studies employed a range of MRI scanners (GE, Philips, Siemens), magnetic field strengths (3.0 and 1.5 Tesla), head coil channels (2, 8, 32, and unreported), pulse sequences, and voxel sizes. Focke et al. and Takao et al. found that variations in MRI scanners influenced VBM analysis outcomes (95, 96), and other studies have demonstrated that magnetic field strength also influences results (97–99). Furthermore, differences in head coil channels, pulse sequences, and voxel sizes have been associated with discrepancies in GM measurements (77, 100–102). Moreover, inadequate management of head movement during image acquisition may compromise image quality and result in inaccuracies in GM quantification (103, 104). Notably, only a limited number of original studies rigorously controlled for head movement during MRI scanning or conducted comprehensive visual inspections and manual corrections of images (11–14, 17, 21, 54, 55, 57–59, 61).

The inconsistency in GM alterations may also result from variations in image processing procedures, which are a crucial aspect of VBM analysis. Initially, brain images from all subjects underwent segmentation and registration into standard space, followed by modulation to compensate for image deformation, and a smoothing process (8). Subsequently, statistical analyses were conducted to identify and interpret differences between groups. The VBM studies included in the CBMA employed various software platforms (different versions of SPM and FSL), applying diverse techniques for image segmentation, registration, modulation, and smoothing, which could potentially influence GM measurements (105–111). Correcting for multiple comparisons is crucial in neuroimaging studies to prevent inflated positive results and bias (112). However, seven out of the 32 analyzed datasets employed uncorrected thresholds (12, 16, 46, 50, 52, 55). Moreover, factors such as age, gender, and total intracranial volume serve as important covariates in VBM analysis (113, 114); however, six studies failed to incorporate any covariates in their analyses (10, 12, 14, 15, 49, 52). Therefore, the multitude of methodological decisions involved in each step of VBM analysis may contribute to inconsistencies in results.

Finally, a small sample size can constrain the ability to identify significant findings in neuroimaging studies, potentially reducing the robustness of the results (115). Fusar-Poli et al. observed that VBM studies with smaller sample sizes generally produce fewer significant findings (116). However, the current ID datasets have an average of only 26.25 participants, which reduces the statistical power necessary to detect significant GM differences. Furthermore, the non-normal distribution of the data complicates the situation further, as smaller studies are more prone to producing false positives compared to larger ones (117). This susceptibility compromises the generalizability and stability of the results. Consequently, the limited sample size likely plays a major role in the discrepancies observed in VBM findings, explaining the lack of significant GM abnormalities identified in this CBMA.

Taken together, the inconsistency in observed GM alterations may primarily arise from demographic and clinical heterogeneity, variations in imaging acquisition and analytical methods, and small sample sizes. To enhance the robustness and reproducibility of future studies, the following recommendations are proposed: (1) conduct power analyses prior to VBM studies to determine appropriate sample sizes; encourage multicenter collaborations or data sharing to increase sample sizes; (2) conduct comprehensive assessments of clinical population characteristics to minimize the potential impact of confounding factors on results; and (3) establish and implement standardized imaging acquisition and analytical methodologies, and apply rigorous statistical strategies for analysis.

It is essential to acknowledge several limitations of this study. First, CBMA relies on peak coordinates and corresponding effect sizes reported in VBM studies rather than on original datasets. Some scholars have suggested that image-based meta-analyses or hybrid meta-analyses combining images and coordinates may improve result accuracy (118, 119), although this approach depends on the availability of raw data shared by researchers. Second, the datasets included in this CBMA are incomplete, as unpublished studies, non-English publications, and studies lacking crucial information were excluded, which may have introduced selection bias. Third, the limited number of studies focusing on specific ID subtypes impeded the conduct of subgroup CBMA analyses to explore the variability within the disorder. Fourth, we did not analyze functional MRI studies in ID. Integrating structural and functional neuroimaging could help bridge the gap between the inconsistent GM changes observed and the functional abnormalities reported, which may be beneficial in revealing the neuropathological mechanisms of ID.

The CBMA of whole-brain VBM studies failed to reveal a consistent and reliable pattern of GM differences between patients with ID and HC subjects. This finding may indicate that GM is not a reliable neuroimaging marker for ID. It is possible that ID may primarily be a functional disorder. Another explanation for CBMA’s inability to detect consistent GM alterations could be the demographic and clinical heterogeneity among participants, along with variations in image acquisition, processing techniques, and small sample sizes. To improve the accuracy of future findings, upcoming VBM studies must rigorously assess potential confounding factors, adhere to standardized protocols for image acquisition and analysis, and aim to increase sample sizes.