Brain Structural Abnormalities in Addiction Disorders: Transdiagnostic Meta‐Analyses

ABSTRACT

Addictions are commonly classified as either substance‐use disorders (SUD) or behavioral addictions (BA). SUDs are well established in taxonomies of mental‐health disorders. BAs are only recently recognized as mental‐health disorders, with limited and conflicting categorizations. Within each category, individual diagnoses are based on objects of addiction. Despite addiction prevalence, no comprehensive coordinate‐based meta‐analysis (CBMA) has examined both SUD and BA. The overall goal of this study was to identify GM alterations associated with SUD and BA, both independently and jointly. To this end, a total of 108 peer‐reviewed VBM studies reporting case–control contrasts in addiction disorders were identified: 83 reports across five SUD diagnoses (8206 participants); and 25 BA reports across two diagnoses (1572 participants). On these datasets, eight transdiagnostic alteration likelihood estimation (ALE) CBMAs were performed. Three primary ALEs computed cross‐study neuroanatomical convergence for: (1) SUD; (2) BA; and (3) Pooled Addictions (SUD + BA). These three primary ALEs were supplemented by five ALE‐to‐ALE comparison analyses: interaction (SUD + BA to SUD and BA), overlap (SUD on BA), conjunction (SUD on BA), and contrast (SUD > BA; BA > SUD). The independently performed ALEs demonstrated largely distinct anatomical‐alteration patterns, overlapping only in the caudate nucleus. The pooled ALE, however, yielded many (n = 24/34) alteration foci not identified in either independent ALE. All newly emerging foci were contributed to by both conditions (SUD and BA) and thus termed “interactive.” Functional decoding of all three ALEs consistently identified reward and reward‐related cognitive operations as being performed by the brain‐regions exhibiting structural alterations in addiction disorders.

Brain Regions Common to Substance Use Disorders and Behavioral Addictions. Voxel‐based morphometry coordinate‐based meta‐analysis of Substance Use Disorders and Behavioral Addictions reveals a shared pattern of gray matter alterations across categories of Addiction.

1. Introduction

Addictions are commonly dichotomized into two categories: substance‐use disorders (SUD; APA 2013) and non‐substance‐use disorders (Robinson and Adinoff 2016), also termed behavioral addictions (BA). Both SUD and BA are characterized by behaviors being compulsively repeated in response to rewards, despite harmful consequences (APA 2013). High prevalence rates of SUD and BA have led to devastating effects across personal and societal domains (SAMHSA 2021; Sussman and Sussman 2011). The negative impact of SUD and BA is compounded by the lack of effective interventions (Fleury et al. 2016) and high rates of relapse (Brecht and Herbeck 2014). Improved treatments for addiction disorders are sorely needed. An achievable therapeutic goal for neuroimaging studies of addictions is to identify candidate neuroanatomical targets for neuromodulatory therapies including deep brain stimulation (DBS), transcranial magnetic stimulation (TMS), transcranial direct current stimulation (tDCS), and focused ultrasound (FUS).

SUD is recognized in the Diagnostic and Statistical Manual of Mental Disorders 3rd (DSM‐3), 4th (DSM‐4), and 5th editions (DSM‐5; APA 2013). In DSM‐5, SUD covers 10 substance‐specific diagnoses. As well, SUD is recognized in the International Classification of Disorders 10th (ICD‐10) and 11th editions (ICD‐11; WHO 2018), also with numerous substance‐specific diagnostic codes. Similarly, SUD is recognized by the Hierarchical Taxonomy of Psychopathology (HiTOP; Kotov et al. 2017), with substance‐specific categories. Distinct scientific communities address SUD by investigating the best possible interventions for SUD (Connery et al. 2020), the pathophysiology of the disorder (Everitt 2014), and animal models of SUD (Wu and Schulz 2012). The advancement of neuroimaging methods now allows for non‐invasive studies in humans which hold promise for developing imaging‐based biomarkers and neuroanatomical targets for neuromodulatory treatment (Heilig and Leggio 2016; Heilig et al. 2021; Volkow et al. 2003).

BAs, by contrast, remain marginalized in the medical and scientific fields. In the DSM, BAs were recognized for the first time in the 5th edition (2013), with Pathological Gambling (PG) as the sole identified disorder. BAs were recognized for the first time in ICD‐11 (WHO 2018), with Internet Gaming (IG) disorder as the only behavioral‐specific category. HiTOP excludes BAs altogether. To this day, PG remains the only behavioral addiction recognized in the latest DSM (DSM‐5‐TR; APA 2022), with ICD‐11 recognizing both PG and IG (WHO 2025). Animal models of BAs exist but are limited (Proctor et al. 2014). To address these issues, clinician scientists have used advanced functional and structural medical‐imaging modalities to study and elucidate behavioral addictions (Yuan, Qin, Liu, and Tian 2011; Yuan, Qin, Wang, et al. 2011) well before their adoption by established diagnostic frameworks (DSM, ICD, HiTOP) or by government organizations (WHO; Zhu et al. 2015).

The Syndrome Model of Addiction (SMA), advanced by the American Psychological Association, offers an alternative perspective on addiction disorders (Shaffer et al. 2004, 2018; Shaffer 2012; Grant et al. 2010). The SMA considers all addictions as a single syndrome in which different objects of addiction—whether substances or behaviors—are varied manifestations of an underlying dysregulation of reward management. This transdiagnostic model unifies the proliferation of diagnoses of DSM, ICD, and HiTOP, by which each object of addiction motivates a unique diagnosis. For SUD, choice of abused substances is strongly influenced by availability and cost (Acuff et al. 2021), clear evidence against the logic of object‐of‐addiction diagnostics. A large body of evidence demonstrates that substance and behavioral addictions share many characteristics, including their development, exposure‐schedule sensitivity, symptoms, tolerance, comorbidity, genetic influences, neurobiology, and treatment responses (Grant et al. 2012; Shaffer et al. 2018; Zagaria et al. 2025). From the perspective of anatomically targeted, neuromodulatory therapeutics, brain regions implicated transdiagnostically, that is, spanning object‐based categories, are the most promising targets for clinical trials.

Voxel‐based morphometry (VBM) is a whole‐brain, voxel‐wise neuroimaging method used to detect subtle but reliable disease‐related brain alterations. VBM converts T1‐weighted MRI into images of gray matter (GM) density, comparing cases to controls as statistical parametric images (SPI). Results are reported as atlas‐referenced neuroanatomical coordinates of local maxima of statistically significant alterations (Wright et al. 1995; Ashburner and Friston 2000). VBM alterations (both atrophy and hypertrophy) have been reported in numerous neurological, psychiatric, and systemic disorders (Glahn et al. 2008; Crossley et al. 2014; Goodkind et al. 2015; Kotkowski et al. 2019; Vanasse et al. 2018, 2021). In neurological disorders, brain structural alterations can be caused by a variety of underlying neuropathologies, notably including protein phosphorylation and misfolding (Tenreiro et al. 2014), inflammation (Chiang et al. 2019), and epileptic excitotoxicity (Armada‐Moreira et al. 2020). In psychiatric disorders, the most likely pathophysiology is metabolic stress at high‐traffic nodes associated with mental/behavioral stereotypies (Crossley et al. 2014; Cauda et al. 2018; Vanasse et al. 2021). In both neurological and psychiatric disorders, the observed alterations are strongly network based, confirming an extended scope of the network degeneration hypothesis (NDH) (He et al. 2007; Glahn et al. 2008; Seeley et al. 2009; Crossley et al. 2014, Goodkind et al. 2015; Vanasse et al. 2018, 2021; Gray et al. 2020, 2023, 2025).

The first report of VBM effects in SUD appeared soon after the invention of the method (Franklin et al. 2002). The first VBM reports in BA appeared much later (Yuan, Qin, Liu, and Tian 2011; Yuan, Qin, Wang, et al. 2011; Zhou et al. 2011). As with all voxel‐wise (mass‐univariate) SPI methods, VBM requires correction for multiple comparisons. Insufficiently conservative corrections for multiple comparisons will predispose SPI analyses to false‐positive results (Woo et al. 2014). The most reliable mean to overcome this limitation is via meta‐analysis (Eklund et al. 2016). For coordinate‐reporting literatures, coordinate‐based meta‐analysis (CBMA) is the most robust and widely applied approach.

CBMA uses the tabular results of a similarly themed body of published SPI reports as input to compute between‐study consensus, expressed as an SPI of voxel‐wise effect likelihood (Fox et al. 1998, 2014). Various CBMA algorithms have been developed and validated (Salo et al. 2023), including activation/alteration likelihood estimation (ALE; Turkeltaub et al. 2002; Eickhoff et al. 2009, 2012; Turkeltaub et al. 2012), multilevel kernel density analysis (MKDA; Wager et al. 2007), anisotropic effect‐size signed differential mapping (AES‐SDM; Radua et al. 2014), and seed‐based d mapping with permutation of subject images (SDM‐PSI; Albajes‐Eizagirre, Solanes, Fullana, et al. 2019, Albajes‐Eizagirre, Solanes, Vieta, and Radua 2019). Of these, ALE is the most well‐validated and widely used (Acar et al. 2018; Tahmasian et al. 2019; Samartsidis et al. 2020; Yeung et al. 2023). For these reasons, ALE was the method of choice for our analyses.

Several prior CBMAs have compared brain alterations in healthy controls versus SUD, BA, or both. Qin et al. (2020) analyzed 20 VBM studies (505 subjects) confined solely to BA using AES‐SDM. Zhang et al. (2021) combined 36 VBM studies of SUD (1476 subjects) and 23 of BA (620 subjects), also using AES SDM. Yan et al. (2023) reported solely on SUD but combined 77 VBM studies (3457 subjects) with 39 voxel‐based physiology (VBP) studies (1444 subjects), using SDM‐PSI. Most recently, Mei et al. (2024) reported on BA alone, but combined 11 VBM studies (287 subjects) with 26 VBP studies (577 subjects), also using AES‐SDM. Despite studying the same addiction disorders and using similar analytical methods, there is much variability in regional effects reported by these four addiction CBMAs. A consistent common inference was of reward system involvement, that emerged despite combining across disorder class (SUD and BA) and imaging modality (VBM and VBP). We note, however, that CBMAs combining VBM and VBP data have shown concordant results in some disorders (major depressive disorder, Gray et al. 2020; temporal lobe epilepsy, Towne et al. 2025), but discordant results in other disorders (Alzheimer's disease, Dang et al. 2025). In addiction disorders, it remains to be demonstrated that VBM and VBP results are concordant.

This study aimed to identify GM alterations associated with SUD and BA separately, as well as those shared across both conditions (SUD + BA). To this end, eight transdiagnostic CBMAs were conducted on the addiction‐disorder VBM literature: three primary analyses and five comparative analyses. In the 1st primary analysis, ALE was performed across five SUD diagnoses. In the 2nd primary analysis, ALE was performed across two BA diagnoses. In the 3rd primary analysis, ALE was performed across seven addiction diagnoses on the pooled addiction‐disorder data set: SUD + BA. Resulting ALE‐derived images were assessed individually and compared for interaction, conjunction, and contrast (SUD > BA and BA > SUD). Interaction analysis examined the pooled SUD + BA ALE to identify effects (ALE clusters) not present in the individual‐disorder ALEs. Overlap and conjunction compared SUD and BA ALE images for spatial commonalities. Contrast compared SUD and BA ALE images for statistically significant differences (SUD > BA; BA > SUD). Based on the results of previously published CBMAs, the symptoms characteristic of addiction disorders, and the transdiagnostic perspective of the Syndrome Model of Addiction, we hypothesized that: (1) SUD alone will show significant between‐study spatial convergence; (2) BA alone will show significant between‐study convergence; (3) comparison of SUD and BA alteration patterns will exhibit spatial overlap and convergence; (4) SUD + BA jointly analyzed will exhibit additional regions of interaction, regions appearing only in the pooled analysis; and (5) functional characterization of both SUD and BA will implicate reward‐system dysregulation.

2. Methods

This study adhered to BrainMap Database (https://brainmap.org/) data‐quality standards (Fox et al. 2005, 2014), including only: (1) studies published in English‐language, peer‐reviewed journals; (2) studies performing whole‐brain image acquisitions in all subjects; (3) study data transformed from native space into an anatomically standardized space, preferably Talairach Space or MNI space; (4) data analyzed as statistical parametric images in a whole‐brain, voxel‐wise (mass univariate) manner, without sub‐sampling or regions‐of‐interest; (5) data expressed as statistical contrasts, either between condition (e.g., task‐control contrasts) or between subject groups (e.g., case–control contrasts); (6) data reported as x‐y‐z coordinates of statistically significant local maxima (center of mass or peak voxel) within the reference space. This study also complied with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses statement (PRISMA; Liberati et al. 2009).

2.1. Data Sources and Searches

Standards‐compliant data sets contrasting persons with addictions to healthy controls were identified using several strategies. The BrainMap VBM data sector (Vanasse et al. 2018) was searched using Sleuth (Version 3.0.4, https://brainmap.org/sleuth/), as a source of relevant publications already coded for meta‐analysis. The BrainMap website publications listing (https://brainmap.org/pubs/) was searched for prior, standards‐compliant coordinate based meta‐analyses papers, which typically list all included papers. Literature repositories and search engines including PubMed, Google Scholar, ResearchGate, and ScienceDirect were also searched. Search terms applied in various combinations included: MRI, imaging, brain imaging, VBM, gray matter; and, addiction, addiction disorder, drug, drug addiction, substance‐use disorder, SUD, drug addiction, polysubstance, alcohol, nicotine, smokers, smoking, stimulants, opioids, cannabis, heroin, marijuana; and, behavioral addiction, BA, pathological gambling, gambler, gambling, internet addiction, internet gaming addiction, and gaming addiction. The literature search was completed in January 2025.

2.2. Study Selection

VBM publications were systematically identified and reviewed for inclusion and exclusion criteria. Only whole‐brain studies that compared persons diagnosed with SUD or BA to healthy controls were allowed. Symptom‐score correlational studies were excluded to promote data homogeneity.

Studies were included if the target population included individuals with a history of SUD or BA and an abstinence period of less than 1 year at the time of enrollment. Studies in which patients met criteria for neurological or psychiatric comorbidities were excluded. Direction of GM change (increase or decrease) was not a selection criterion; studies which reported GM decreases, increases, or both were allowed.

Data redundancy can introduce study bias in meta‐analyses. For this reason, data were screened for duplication at the paper and experiment level. Particular attention was paid to sequential publications by the same authors. When this was observed, only the most recently reported (typically largest) cohort was selected. When selection questions arose, co‐author consensus was used to resolve uncertainty. Data was double checked to ensure no duplicated data per Müller et al. (2018). Included studies and associated details are provided in Tables S1 and S2. Study triage is illustrated in Figure 1 and the final dataset used for these analyses is reported in Table 1. Both BA and SUD met the recommended minimum number of experiments for VBM studies (Frahm et al. 2023).

FIGURE 1.

Study selection flow diagram. Published VBM studies were systematically reviewed for inclusion in the meta‐analysis. Study selection adhered to standard quality criteria of the BrainMap Project and the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses statement. For more information, visit: https://www.brainmap.org/. BA, behavioral addiction; CTh, cortical thickness; HC, healthy control; ROI, region of interest; SUD, substance use disorder; VBM, voxel‐based morphometry.

TABLE 1.

Sample sizes of included studies.

Sample sizes				Addiction types
Papers (experiments; %)	Patients	Controls
Substance use disorders
28 (27; 33)	1613	1343		Alcohol
16 (16; 19)	1086	1426		Nicotine
20 (20; 24)	916	780		Stimulants
10 (11; 13)	261	256		Opioids
9 (9; 11)	268	257		Cannabis
83 (83; 77)	4144	4062	SUD totals
Behavioral addictions
19 (19; 76)	592	578		Internet gaming disorder
6 (6; 24)	182	220		Pathological gambling
25 (25; 33)	774	798	BA totals
108 (108)	4918	4824	Combined totals

Note: Papers tallied met inclusion criteria and avoided exclusion criteria for the current study. All included papers were peer‐reviewed, English‐language reports of one or more voxel‐based morphometric analyses performed as case–control contrasts of persons with a substance use disorder (SUD) or behavioral addiction (BA) to demographically matched healthy control subjects. Experiments are the results of such statistical contrasts. Note the large difference in sample size for SUD and BA.

Abbreviations: BA, behavioral addiction; SUD, substance use disorder.

2.3. Coordinate‐Data Harmonization

Data retrieved from the BrainMap VBM sector (Vanasse et al. 2018) were exported in MNI space. Data coded using Scribe (Version 3.6; https://brainmap.org/scribe/) for BrainMap entry were coded in the as‐published reporting space and converted into MNI space prior to analysis using the Lancaster transforms (Lancaster et al. 2007), to eliminate the spatial disparity between reference frames to optimize CBMA accuracy (Laird et al. 2010).

2.4. Statistical Threshold Determination

Prior to performing ALE analyses preliminary assessments of the spatial extent of the ALE outputs were performed at three cluster‐forming thresholds. Cluster‐forming thresholding is a voxel‐level preprocessing step (uncorrected for multiple comparisons) which precedes formal significance testing of clusters. Commonly applied cluster‐forming thresholds are p < 0.001 or p < 0.005. For smaller sample sizes and for contrast and conjunction comparisons of pairs of ALE images, less stringent thresholds may be applied. Comparisons of unequal sample sizes are particularly problematic (Laird et al. 2005) and can benefit from less stringent cluster‐forming thresholding. To determine an appropriate statistical threshold for the presently available samples (Table 1), preliminary assessments of ALE outputs for each sample were computed for a range of cluster‐forming thresholds: p < 0.001, < 0.005, and < 0.01 (Table 2). Each assessment performed 1000 permutations using GingerALE (3.0.2). From these results, a voxel‐level, cluster‐forming threshold of p = 0.01 was selected. This threshold level maximized the brain volumes and numbers of foci to be formally analyzed for all 3 datasets, to mitigate sample‐size bias and allow more meaningful assessment of conjunctions and contrasts.

TABLE 2.

Statistical threshold assessment.

Level	Substance use disorders				Behavioral addictions				Pooled addictions
p	Clusters	Foci	Volume	Voxels	Clusters	Foci	Volume	Voxels	Clusters	Foci	Volume	Voxels
0.001	4	7	5200	708	0	0	0	0	5	10	4800	735
0.005	3	8	8600	1075	1	4	2150	270	3	34	10,086	1282
0.01	4	15	14,632	1829	1	6	3464	433	6	34	30,464	3800

Note: Prior to performing ALE analyses, preliminary assessments of the spatial extent of the ALE outputs were performed at three cluster‐forming thresholds. Cluster‐forming thresholding is a voxel‐level pre‐processing step (uncorrected for multiple comparisons) which precedes cluster‐level significance testing (FWE corrected for multiple comparisons). Commonly recommended cluster‐forming thresholds are p < 0.001 or p < 0.005. For smaller sample sizes, and for contrast and conjunction analyses of pairs of ALE images, less stringent thresholds may be applied. For all ALE computation in the present study, a cluster‐forming threshold of p = 0.01 was applied. This threshold level maximized the brain volumes and numbers of foci to be formally analyzed for all 3 datasets, allowing more meaningful assessment of conjunctions and contrasts. Column headings: Clusters, number of clusters observed in ALE output; Foci, number of local maxima within the output clusters; Level, voxel‐level (cluster forming), uncorrected statical threshold; Volume, spatial extent of output in mm³; Voxels, number of 2 mm (isotropic) voxels contained in the output volume.

2.5. ALE Analyses

The GingerALE application (Version 3.0.2, https://www.brainmap.org/ale/) was used for all analyses (Turkeltaub et al. 2002, 2012; Laird et al. 2005; Laird, Eickhoff, Kurth, et al. 2009; Laird, Eickhoff, Li, et al. 2009; Eickhoff et al. 2009, 2011, 2012, 2016; Kuppers et al. 2024). ALE analyses were conducted for the SUD and BA literatures both independently and pooled (SUD + BA). The ALE algorithm treated each x‐y‐z coordinate as a Gaussian probability distribution to account for spatial uncertainty, weighted by sample size (subjects per study), to form modeled alteration maps (MA maps) per experiment. From the per‐experiment MA maps, ALE computed a union of probabilities across experiments as ALE values. Significance was determined relative to a spatially random null distribution using parameters.

Correction for family‐wise error rate at the cluster‐level (cFWE) was the approach of choice (Frahm et al. 2023). Cluster‐level inference generated a simulated data set of randomly distributed foci based on characteristics of the input data set for testing the null hypothesis. To minimize bias in the comparisons of SUD and BA, a voxel‐level, cluster‐forming threshold of p < 0.01 (uncorrected) was used for all three analyses, per the preceding section (Statistical Threshold Determination). Cluster‐level significance was performed at p < 0.05 with family‐wise error correction for multiple comparisons, as is commonly recommended (Müller et al. 2018; Tahmasian et al. 2019; Yeung et al. 2023). GingerALE output for analysis included: ALE images in NifTI format (.nii) and a peaks spreadsheet (.xls). The peaks spreadsheet characterized the above‐threshold brain‐alterations sorting data into clusters, providing local maxima locations (x, y, z coordinates in MNI space) within clusters, maximum ALE value per local maxima, and anatomical labels of the nearest GM region from the Talairach Daemon (Lancaster et al. 2000). Mango software (Version 4.1, https://ric.uthscsa.edu/mango/) was used to determine the volume of each cluster and to visualize ALE results by overlaying them onto a standardized brain template (T1 weighted image in MNI space).

2.5.1. ALE 1: Substance Use Disorders Alone

This ALE analysis identified GM anatomical alterations across five SUD diagnoses. Abused substances included: alcohol, nicotine, stimulants, opioids, and cannabis (Table 1). A total of 83 SUD experiments were included, collectively representing 4144 persons with SUD and 4062 healthy controls (Figure 1, Table 1).

2.5.2. ALE 2: Behavioral Addictions Alone

This ALE identified GM anatomical alterations across two BA diagnoses. Behavioral addictions included: internet gaming disorder, and gambling addiction (Table 1). A total of 25 BA experiments were included in the analysis, collectively representing 774 persons with BA and 703 healthy controls (Figure 1, Table 1).

2.5.3. ALE 3: Pooled Addictions: Substance Use Disorders Plus Behavioral Addictions

This ALE was performed to identify a pattern of GM alteration across all addiction disorders (SUD + BA), a total of seven addiction‐object‐based diagnoses. This analysis consisted of 108 experiments: 83 (77%) SUD; 25 (23%) BA (Figure 1, Table 1).

2.6. ALE Comparison Analyses: Interaction, Overlap, Conjunction, Spatial Correlation, and Contrast

Interactions are defined statistically as effects emerging from two variables examined jointly that are not present when examined independently. To identify interactions, ALE cluster output for SUD and BA alone were compared to SUD + BA. To confirm these “new” clusters as interactive, a percent contribution analysis was performed for the pooled SUD + BA analysis to confirm that each cluster had contribution from both SUD and BA studies. Briefly, the number of foci that contributed to each cluster from each study (SUD or BA) was quantified and a percent contribution of SUD and BA studies was calculated for each cluster.

Commonalities are effects observed for both variables when examined alone. Contrasts identify effects observed within one dataset that are not observed in the other. Commonalities were determined by spatial overlap, conjunction, and spatial cross‐correlation. Overlap was identified using the Mango (Version 4.1) overlay functions. Conjunction was computed using the GingerALE (Version 3.0.2) conjunction operator. Note that both Overlap and Conjunction computed the voxel‐wise minimum of two ALE images (SUD and BA); that is, they are mathematically identical operations. Spatial cross‐correlation was performed with FSL's fslcc tool (Version 6.0.7.18; Smith et al. 2004).

Contrasts were identified using the “Contrast” functions in GingerALE (Version 3.0.2) using previously described procedures (Laird et al. 2005; Eickhoff et al. 2011; Kuppers et al. 2024). Inputs for GingerALE conjunction and contrast analyses were the 3 ALEs described in the prior section. The pooled SUD + BA ALE provided the binary mask within which contrast and conjunction were computed.

2.7. Sign of Change in Gray Matter Volume Analysis

The direction of GM volume change sign (increase or decrease) was determined for each cluster based on the contributing studies. Clusters composed exclusively of studies reporting increased GM volume were classified as increases, whereas clusters composed exclusively of studies reporting decreased GM volume (atrophy) were classified as decreases. Clusters that included studies reporting both increases and decreases in GM volume were classified as mixed (increase/decrease).

2.8. Fail‐Safe N Analyses

Despite CBMA robustness, publication bias is a threat of validity to many meta‐analytic studies that focus on finding patterns from published studies. Meta‐analyses relying on effect size (e.g., using Cohen's d or similar) are highly susceptible to missing data (file‐drawer effect), while effect‐location meta‐analyses (Fox et al. 1998), as performed by ALE, are minimally susceptible to missing data (Samartsidis et al. 2020). The fail‐safe N (FSN; Rosenthal 1979) method was introduced to correct for missing data in effect‐size meta‐analysis. Acar et al. (2018) proposed an ALE adaptation of the FSN method, which assesses the impact of null findings on the observed results, by introducing noise into the ALE algorithm. An in‐house version of the FSN method with a step‐wise workflow was applied (Gray et al. 2020, 2025). Each ALE was re‐computed iteratively with known amounts of added random noise to quantify result robustness.

2.9. Functional Network Membership Interpretive Analysis

Network membership of clusters was determined by spatial overlap with masks of canonical functional connectivity (FC) networks obtained meta‐analytically by applying the MELODIC independent components analysis application to the BrainMap Task‐Activation Database (Fox et al. 2005), as first described by Smith et al. (2009). The canonical network masks and names applied were those reported by Vanasse et al. (2021), with BrainMap input data limited to healthy controls.

2.10. BrainMap Meta‐Data Functional Interpretation Analyses

Meta‐data from the BrainMap Task Activation Database (Fox et al. 2005) filtered for healthy controls were used to infer the functional roles of brain regions that exhibited GM alterations by ALE analyses implemented as Mango plug‐ins (Version 4.1, https://ric.uthscsa.edu/mango/; Lancaster et al. 2012). Coordinates of clusters were transformed from MNI to Talairach Space (Lancaster et al. 2007) prior to using the plugins, as Talairach is the native space of the plugins.

2.10.1. Paradigm Class Analyses

Paradigm Classes designate experimental paradigms employed for controlled in‐scanner task performance, using terms such as N‐back, Stroop, Go/No‐Go. Paradigm class analyses were completed using the Paradigm Analysis plugin (Version 2.0, 115 paradigm classes). The quantitative regional behavior analysis was performed on masks created from selected brain regions of interest (ROIs) that are anatomically altered in addiction, to infer how these regions relate to healthy brain functions. Paradigm class analyses were performed for the whole image ALE results and for overlap and contrast results. Paradigm classes with the operative threshold z‐score of 3.3 or greater, as comparable to a group p‐value of 0.05, were considered statistically significant.

2.10.2. Behavioral Domain Analyses

Behavioral Domain analyses were performed using the Behavioral Analysis plugin (Version 4.0). The plugin conducts a regional behavioral analysis for each selected ROI and presents the results according to BrainMap's Behavioral Domain framework. This framework includes five primary domains, namely Action, Cognition, Emotion, Interoception, and Perception, along with 60 associated subdomains. A z‐score of 3.0 or greater was considered statistically significant (p‐value less than or equal to 0.05 with Bonferroni correction for multiple comparisons). Behavioral Domain analyses were performed for the whole image ALE results and for overlap and contrast results.

3. Results

3.1. Study Inclusion and Characteristics

A total of 108 individual VBM experiments (108 papers) comprising results from 4918 patients and 4860 healthy controls were identified for inclusion in this meta‐analysis (Table 1). Substance use disorder studies (83 experiments/papers) included five diagnoses: alcohol use disorder (1613 participants); nicotine use disorder (1086 participants); stimulant use disorder (916 participants); opioid use disorder (261 participants); and cannabis use disorder (268 participants). Behavioral addictions (25 experiments/papers) included: internet gaming disorder (592 participants); and pathological gambling (182 participants). Supporting Information provides: (1) per‐study MRI acquisition and analysis parameters (Table S1); (2) participant demographics (Table S2); and (3) source‐data citations.

3.2. ALE Analyses

3.2.1. ALE 1: Substance Use Disorders

The SUD meta‐analysis was computed from a total of 650 reported coordinates (x‐y‐z reported as foci) comprising the concatenated results of 83 experiments (Table 1). These per‐experiment regional effects were reduced by ALE to four clusters with a total of 15 local maxima (Table 3) and a total cluster volume of 14,632 mm³. Patterns of convergence were found in several regions including the right thalamus (medial dorsal nucleus), left caudate head, bilateral medial frontal gyrus (MFG) (Brodmann areas [BrA] 6, 9, 10, and 11), left cingulate gyrus (CG; BrA 32), left superior frontal gyrus (SFG; BrA 6), and bilateral anterior cingulate cortex (ACC; BrA 32) (Table 3 and Figure 2).

TABLE 3.

Substance use disorders ALE: Output clusters and local maxima.

Clusters and foci	Sign	Size	Side	Region	ALE	MNI			FSN (%)	Network	z‐score
						x	y	z
1.1	↑/↓	4784	R	Thal, DMNu	29.03	6	−10	8	> 30	THN, IC‐6	24.5
1.2			L	Caud, head	21.44	−2	2	0		FSN, IC‐14	10.4
2.1	↓	3992	L	ACG, BrA32	24.13	0	22	36	> 30	CCN, IC‐17	74.2
2.2			R	MFG, BrA 6	21.69	4	20	46		CCN, IC‐17	47.5
2.3			L	ACG, BrA 32	21.37	2	32	28		CCN, IC‐17	30.5
2.4			L	MFG, BrA 9	16.79	−2	44	26		DMN, IC‐13	16.7
2.5			L	SFG, BrA 6	16.70	−2	16	60		VMC, IC‐5	12.1
3.1	↓	3032	R	ACG, BrA 32	32.22	6	48	−2	> 30	DMN, IC‐13	16.3
3.2			R	ACG, BrA 32	21.23	4	44	12		DMN, IC‐13	17.6
4.1	↑/↓	2824	R	ACG, BrA 32	19.61	2	36	−14	< 6	DMN, IC‐13	6.1
4.2			L	ACG, BrA 32	19.18	−6	44	−12		DMN, IC‐13	9.5
4.3			R	ACG, BrA 32	18.57	8	20	−14		DMN, IC‐13	1.8
4.4			L	MFG, BrA 10	15.94	0	50	−16		DMN, IC‐13	3.7
4.5			R	MFG, BrA 11	15.14	8	36	−20		TLN, IC‐18	0.8
4.6			L	MFG, BrA 10	14.61	−2	46	−18		DMN, IC‐13	3.7

Note: In the ALE contrast of persons with substance use disorders vs. healthy controls, four gray matter clusters with 15 local maxima and a total cluster volume of 14,632 mm³ were observed. Column headings: Cluster: cluster #. Sign: ↓, patients<controls (gray matter atrophy); ↑, patients>controls (gray matter hypertrophy); ↑/↓, both atrophy and hypertrophy. Size: cluster size in mm³. Region: anatomical terms per local maxima from Talairach Daemon (Lancaster et al. 2000). ALE: peak ale score per local maxima x 10³. MNI: local‐maxima address in Montreal Neurological Institute coordinates. FSN: Fail Safe N values computed per Gray et al. (2020). Network: anatomical terms per Vanasse et al. (2021). Z‐score: cluster‐level z‐score, from GingerALE output. Region abbreviations: ACG, anterior cingulate gyrus; BrA, Brodmann area; Caud, caudate; MFG, middle frontal gyrus; SFG, superior frontal gyrus; Thal, thalamus. Network abbreviations: CCN, cognitive control network; DMN, default mode network; FSN, fronto‐striatal network; ThN, thalamic network; TLN, temporo‐limbic network; VMC, visuomotor control network.

FIGURE 2.

Substance use disorder ALE image. ALE was performed on the results of 83 peer‐reviewed reports of voxel‐based morphometry (VBM), case–control contrasts of 4144 persons with substance use disorder (SUD) against 4062 controls. Depicted results were thresholded at P < 0.05 cFWE. ALE, activation/alteration likelihood estimation; L, left; R, right.

3.2.2. ALE 2: Behavioral Addictions

The BA meta‐analysis was computed from a total of 127 reported coordinates (foci) comprising the concatenated results of 25 experiments (Table 1). These per‐experiment regional effects were reduced by ALE to one cluster with 6 local maxima (Table 4) and a total cluster volume of 3464 mm³. ALE results demonstrated convergent abnormality in only one cluster with six local maxima (Table 4). Local maxima were in the left ACC (BrA 32 and 33), right CG (BrA 24), left caudate body and head, and left lentiform nucleus (putamen) (Table 4 and Figure 3).

TABLE 4.

Behavioral addictions ALE: Output clusters and local maxima.

Clusters and foci	Sign	Size	Side	Region	ALE	MNI			FSN (%)	Network	z‐score
						x	y	z
1.1	↑/↓	3464	L	ACG, BrA 32	16.41	−10	30	20	> 30	CCN, IC‐17	12.5
1.2			R	CG, BrA 24	12.30	4	18	26		CCN, IC‐17	34.8
1.3			L	Caud, body	11.01	−10	8	4		FSN, IC‐14	31.0
1.4			L	Caud, head	10.51	−8	20	0		FSN, IC‐14	8.3
1.5			L	ACG, BrA 33	09.91	−4	22	16		CCN, IC‐17	9.1
1.6			L	Lent, Put	09.49	−18	4	6		FSN, IC‐14	32.8

Note: In the ALE contrast of persons with behavioral addictions versus healthy controls, one gray matter cluster with six local maxima and a total cluster volume of 3464 mm³ was observed. Column headings: Cluster: cluster #. Sign: ↓, patients<controls (gray matter atrophy); ↑, patients>controls (gray matter hypertrophy); ↑/↓, both atrophy and hypertrophy. Size: cluster size in mm³. Region: anatomical terms per local maxima from Talairach Daemon (Lancaster et al. 2000). ALE: peak ale score per local maxima x 10³. MNI: local‐maxima address in Montreal Neurological Institute coordinates. FSN: Fail Safe N values computed per Gray et al. (2020). Network: anatomical terms per Vanasse et al. (2021). Z‐score: cluster‐level z‐score, from GingerALE output. Region abbreviations: ACG, anterior cingulate gyrus; BrA, Brodmann area; Caud, caudate; CG, cingulate gyrus; Lent, lentiform nucleus; MFG, middle frontal gyrus; Put, putamen. Network abbreviations: CCN, cognitive control network; FSN, fronto‐striatal network.

FIGURE 3.

Behavioral addiction ALE image. ALE was performed on the results of 25 peer‐reviewed reports of voxel‐based morphometry (VBM), case–control contrasts of 774 persons with behavioral addiction (BA) against 798 controls. Depicted results were thresholded at p < 0.05 cFWE. ALE, activation/alteration likelihood estimation; L, left; R, right.

3.2.3. ALE 3: Pooled Addictions: Substance Use Disorders and Behavioral Addictions

The pooled (SUD + BA) addiction was computed from a total of 777 reported coordinates (foci) comprising the concatenated results of 108 experiments (Table 1). The per‐experiment regions effects were reduced by ALE to 6 clusters with 34 local maxima and total cluster volume of 30,604 mm³ (Table 5 and Figure 4A). The Pooled Addictions ALE exceeded the sum of the BA and SUD ALEs in total cluster number, local maxima number, and cluster volume, robustly confirming the anticipated interaction effects. This is further addressed in the following section.

TABLE 5.

Pooled addictions ALE: Clusters and local maxima.

Clusters and foci	Sign	Size	New	%	Side	Region	ALE	MNI			FSN	Network	z‐score
								x	y	z
1.1	↑/↓	10,216	N	80/20	R	ACG, BrA32	32.24	6	48	−2	> 30	DMN, IC‐13	16.3
1.2			Y	80/20	L	SFG, BrA10	32.15	−8	60	−16		DMN, IC‐13	0.4
1.3			N	80/20	L	ACG, BrA32	24.69	−10	30	20		CCN, IC‐17	12.5
1.4			N	80/20	L	ACG, BrA32	24.42	0	22	36		CCN, IC‐17	74.2
1.5			Y	80/20	R	ACG, BrA32	23.56	4	30	34		CCN, IC‐17	40.9
1.6			N	80/20	L	ACG, BrA32	22.25	2	32	28		CCN, IC‐17	30.5
1.7			N	80/20	R	MFG, BrA6	21.70	4	20	46		IC‐17	47.5
1.8			Y	80/20	L	MFG, BrA10	21.58	−12	62	−2		DMN IC‐13	2.6
1.9			N	80/20	R	ACG, BrA32	21.24	4	44	12		DMN IC‐13	17.6
1.10			Y	80/20	L	MFG, BrA9	19.81	−4	50	26		DMN IC‐13	20.6
2.1	↑/↓	5208	Y	81/19	R	IFG, Ins, BrA13	32.20	32	4	−18	> 30	TLN, IC‐18	12.9
2.2			Y	81/19	R	Lent, Put	21.15	18	6	−16		TLN, IC‐18	15.6
2.3			Y	81/19	R	Lent, GP	20.10	26	−16	−8		TLN, IC‐18	22.3
2.4			Y	81/19	R	ParaHip, Amy	17.42	34	−8	−18		TLN, IC‐18	12.8
2.5			Y	81/19	R	Lent. Put	17.18	36	−14	−6		TLN, IC‐18	10.2
2.6			Y	81/19	R	ParaHip, Hip	13.91	38	−14	−28		TLN, IC‐18	3.3
3.1	↑/↓	4704	Y	83/17	L	Ins, BrA13	24.92	−44	−8	−2	< 6	AN, IC‐7	15.8
3.2			Y	83/17	L	ParaHip, Amy	23.04	−34	−8	−20		TLN, IC‐18	12.8
3.3			Y	83/17	L	Lent, GP	21.81	−26	−14	−8		TLN, IC‐18	28.9
3.4			Y	83/17	L	Lent. Put	21.19	−28	−4	6		SN IC‐2	21.0
3.5			Y	83/17	L	MTG, BrA21	16.75	−56	−2	−14		DMN, IC‐13	6.6
3.6			Y	83/17	L	Claus	16.39	−36	−12	2		SN IC‐2	12.3
3.7			Y	83/17	L	MTG, BrA21	14.54	−40	−14	−14		TLN, IC‐18	4.9
3.8			Y	83/17	L	Claus	13.64	−40	−20	0		AN, IC‐7	14.2
4.1	↑/↓	4624	N	95/5	R	Thal, MDN	29.03	6	−10	8	> 30	THN, IC − 6	24.5
4.2	↑/↓		N	95/5	L	Caud, Head	24.19	−6	8	2		FSN, IC − 14	21.5
5.1	↑/↓	3176	Y	89/11	L	ACG, BrA32	24.47	0	36	−14	< 6	DMN, IC‐13	6.8
5.2			Y	89/11	L	ACG, BrA32	22.02	−6	42	−12		DMN, IC‐13	10.0
5.3			N	89/11	R	ACG, BrA32	19.00	8	20	−14		DMN, IC‐13	1.8
5.4			N	89/11	L	MFG, BrA10	16.08	0	50	−16		DMN, IC‐13	3.7
6.1	↑/↓	2536	Y	85/15	R	Ins, BrA13	24.01	38	6	8	< 6	SoN IC − 2	29.3
6.2			Y	85/15	R	Claus	20.30	34	14	−2		SaN, IC − 4	25.8
6.3			Y	85/15	R	Ins, BrA13	16.47	36	22	0		SaN, IC‐4	31.5
6.4			Y	85/15	R	Ins, BrA13	16.25	46	10	−4		VAN, IC‐16	14.5

Note: The Pooled Addictions analysis included both persons with substance use disorder and behavioral addiction versus healthy controls. This analysis resulted in 6 gray matter clusters with 34 local maxima and a total cluster volume of 30,604 mm³. Of the 34 local maxima, 24 are new, not being present in SUD or BA alone. Column headings: Cluster: cluster #. Sign: ↓, patients<controls (gray matter atrophy); ↑, patients>controls (gray matter hypertrophy); ↑/↓, both atrophy and hypertrophy. Size: cluster size in mm³. %: percentage contribution SUD/BA. New: Y, cluster occurring only pooled ALE; N, cluster appears in SUD or BA ALE. Region: anatomical terms per local maxima from Talairach Daemon (Lancaster et al. 2000). ALE: peak ale score per local maxima × 10³. MNI: local‐maxima address in Montreal Neurological Institute coordinates. FSN: Fail Safe N values computed per Gray et al. (2020). Network: anatomical terms per Vanasse et al. (2021). Z‐score: cluster‐level z‐score, from GingerALE output. Region abbreviations: ACG, anterior cingulate gyrus; BrA, Brodmann area; Caud, caudate; MFG, middle frontal gyrus; SFG, Superior Frontal Gyrus; Thal, thalamus. Network abbreviations: CCN, cognitive control network; DMN, default mode network; FSN, fronto‐striatal network; Thal, thalamus; TLN, temporo‐limbic network; VMC, visuomotor control network.

FIGURE 4.

Pooled addiction ALE image. ALE was performed on the pooled results of 108 peer‐reviewed reports of voxel‐based morphometry (VBM), case–control contrasts of 4918 persons with substance use disorder (SUD) and behavioral addiction (BA) against 4824 controls (A). The interactive regions ALE image represents the clusters which were present only in the pooled analysis (B). Depicted results were thresholded at p < 0.05 cFWE. ALE, activation/alteration likelihood estimation; BA, behavioral addiction; L, left; R, right; SUD, substance use disorder.

3.3. ALE Comparison Analyses: Interaction, Overlap, Conjunction, Spatial Correlation, and Contrast

3.3.1. Interaction Analysis

The Pooled Addictions ALE (SUD + BA) yielded more clusters (6 vs. 5), more local maxima (34 vs. 21), and greater cluster volume (30,604 mm³ vs. 18,095 mm³) than the sum of the two independent ALEs (SUD and BA). This is most parsimoniously understood as an interaction effect, in which voxels that were below‐threshold in each independent data set positively interacted to surpass the significance threshold in the combined analysis. As GingerALE does not provide an interaction analysis tool, two inspection‐based interaction tests were performed. First, binary masks of the Pooled Addictions ALE clusters were compared to those of the SUD and BA ALEs to identify clusters present in the former that were not present in either of the latter. These clusters are illustrated in Figure 4B and enumerated in Table 5 (columns New and %). Interactive areas include the right and left insula (BrA 13), bilateral claustrum, left SFG (BrA 10), right inferior frontal gyrus (BrA 13), bilateral lentiform nucleus (lateral globus pallidus), bilateral parahippocampal gyrus (amygdala), right parahippocampal gyrus (hippocampus), and left middle temporal gyrus (MTG, BrA 21).

In the percent contribution analyses, SUD contributed 80% to 95% of the data; BA contributed 5%–20% of the data. These percentages are in broad agreement with the relative numbers of experiments (83 [77%] SUD vs. 25 [33%] BA) and included coordinates (650 [84%] SUD vs. 127 [16%] BA) contributing to the ALE analyses.

3.3.2. Commonalities: Overlap and Conjunction Analyses

Overlap analysis was performed using binary masks of clusters from SUD and BA ALEs (Figure 5A, Table 6). Conjunction analysis was performed using the GingerALE Contrasts function (Figure 5B, Table 6). The two commonality analyses gave identical results, both finding a small region of commonality in the caudate head. In agreement with the overlap and conjunction analyses, spatial cross‐correlation was very low (r = 0.0045).

FIGURE 5.

Overlap and conjunction of SUD and BA ALE images. (A) Overlap. SUD (blue) and BA (green) ALE images spatially overlapped at caudate head (pink), indicated by arrow. (B) Conjunction of BA and SUD ALE images was also in the caudate head (pink/red), indicated by the arrow. ALE, activation/alteration likelihood estimation; BA, behavioral addiction; L, left; R, right; SUD, substance use disorder.

TABLE 6.

Overlap and conjunction analyses: Output clusters and local maxima.

Clusters and foci	Sign	Size	Side	Region	ALE	MNI			Network	z‐score
						x	y	z
Overlap ALE
1	↓	56	L	Caud, head	12.16	−6	8	2	FSN, IC‐14	21.5
Conjunction ALE
1	↓	56	L	Caud, head	8.9	−8	8	2	FSN, IC‐14	21.5

Note: Overlap was the binary mask overlap of the two source ALEs. Conjunction analysis computed regions of commonality by statistical comparison of two source ALEs (Laird et al. 2005; Eickhoff et al. 2011; Kuppers et al. 2024). In each comparison, a single cluster was observed with a single local maximum (listed) in the same location. Column headings: Cluster: cluster #. Sign: ↓, patients < controls (gray matter atrophy); ↑, patients > controls (gray matter hypertrophy). Size: cluster size in mm³. Region: anatomical terms per local maxima from Talairach Daemon (Lancaster et al. 2000). ALE: peak ale score per local maxima × 10³. MNI: local‐maxima address in Montreal Neurological Institute coordinates. FSN: Fail Safe N values computed per Gray et al. (2020). Network: anatomical terms per Vanasse et al. (2021). Z‐score: cluster‐level z‐score, from GingerALE output. Region abbreviations: Caud, caudate. Network abbreviations: FSN, fronto‐striatal network.

3.3.3. Contrast Analysis

Contrast Analysis comparing SUD and BA ALEs (using the GingerALE contrast function) revealed three clusters (all in Thalamus) that were significantly greater in the SUD ALE than in the BA ALE (Figure 6; Table 7). Contrast Analysis found no regions significantly greater in the BA ALE than the SUD ALE.

FIGURE 6.

Contrast between SUD and BA ALE images. Contrast analysis reported the thalamus (yellow arrow pointing to red ROI) as being more significantly associated with SUD versus BA (SUD>BA). No clusters were more significantly associated with BA when compared with SUD (BA > SUD). ALE, activation/alteration likelihood estimation; BA, behavioral addiction; L, left; R, right; ROI, region of interest; SUD, substance use disorder.

TABLE 7.

Contrast analysis: Output clusters and local maxima.

Cluster#	Side	Region	z‐score	MNI coordinates of local maxima			p
				x	y	z
SUD>BA
1	R	Thalamus	3.4316144	6	−16	0	0.0003
	L	Thalamus	3.2905266	2	−16	−2	0.0005
	R	Thalamus	3.1946511	4	−15.3	7.3	0.0007
2	R	Thalamus	3.1946511	7.3	−18	15.3	0.0007
3	L	Thalamus, Medial Dorsal Nucleus	3.0902321	2	−20	0	0.001
BA > SUD
No clusters found

Note: The Contrast Analysis revealed three clusters in the right and left Thalamus as being more significantly associated with SUD versus BA. No clusters were more significantly associated with BA when compared with SUD.

Abbreviations: BA, behavioral addiction; L, left; MNI, Montreal Neurological Institute; R, right; SUD, substance use disorder.

3.4. Fail‐Safe N Analyses

From the independent ALE analyses, SUD Clusters 1, 2, and 3 survived 30% added noise, while Cluster 4 did not survive 6% noise (Table 3). BA consisted of only one cluster which survived 30% added noise (Table 4). From the pooled ALE analysis (SUD + BA), Clusters 1, 2, and 4 survived 30% noise, while Clusters 3, 5, and 6 did not survive 6% noise (Table 5).

3.5. Network Membership Interpretive Analysis

Each cluster significantly overlapped (z ≥ 3) with at least one canonical functional connectivity (FC) network. In the independent SUD analysis, clusters showed associations with thalamic, fronto‐striatal, cognitive control, visuomotor, and default mode networks, with Clusters 3 and 4 mapping exclusively to the default mode network (Table 3). In the independent BA analysis, Cluster 1 corresponded to the cognitive control and fronto‐striatal networks (Table 4). The overlap/conjunction region in the caudate head was also linked to the fronto‐striatal network (Table 6). In the pooled SUD + BA analysis, clusters mapped to several networks, including default mode, cognitive control, tempero‐limbic, thalamic, somesthesis, salience, ventral attention, and auditory networks (Table 5).

3.6. Sign of Change in Gray Matter Volume Analysis

In the independent SUD analysis, Clusters 2 and 3 were driven exclusively by studies reporting decreases (atrophy) in GM volume, whereas Clusters 1 and 4 included contributions from studies reporting both decreases and increases in GM volume (Table 3). The single cluster identified in the independent BA analysis was supported by studies reporting both increased and decreased GM volume (Table 4). Similarly, all clusters derived from the Pooled SUD + BA analysis reflected contributions from studies reporting both directions of GM volume change (Table 5).

3.7. Functional Interpretation Analyses

3.7.1. Paradigm Class Analyses

Whole‐image Paradigm Class analyses of ALE output for SUD, BA, and Pooled addictions (SUD + BA) all demonstrated significant associations (z ≥ 3.3) with reward (z = 9.1, 3.8, and 9.7, respectively; Figure 7). For all three ALEs, reward was the most significant association, and for BA it was the only significant association. SUD and Pooled ALEs also exhibited significant associations with other tasks including: response inhibition (Go/No‐Go; z = 5.8 and 4.9); emotion induction (z = 5.1); sexual arousal (z = 3.8); and discriminations of pain (z = 5.6 and 6.7), facial affect (z = 4.6 and 6.9), and olfaction (z = 4.6 and 4.6). In the Overlap/Conjunction analysis of the SUD and BA ALEs, the observed cluster (caudate head; Table 6, Figure 5) showed a near‐significant association with reward (z = 2.5). In the Contrast analysis of the SUD and BA ALEs, the observed cluster (thalamus; Table 7, Figure 6) showed no significant functional associations.

FIGURE 7.

Functional interpretations of brain alterations in addiction disorders. Functional interpretation analyses using Paradigm Class (left) and Behavioral Domain (right) were performed on whole image SUD, BA, and pooled ALE results. Paradigm Class analysis showed significant associations for SUD (A) and Pooled (C) ALE with several paradigm classes, whereas BA was associated only with Reward (B). Reward was the strongest paradigm across all analyses (A, B, and C; z = 9.08, 3.83, and 9.68 respectively). Behavioral Domain analyses found SUD linked to several domains (D), BA only to Reward/Gain (E), and Pooled to Reward/Gain and Reasoning (F). All analyses were significantly associated with Reward/Gain (D, E, and F; z = 4.06, 3.36, and 4.16 respectively). Significance thresholds were z ≥ 3.3 for Paradigm Class and z ≥ 3.0 for Behavioral Domain. ALE, activation/alteration likelihood estimation; BA, behavioral addiction; SUD, substance use disorder.

3.7.2. Behavioral Domain Analyses

Whole‐image Behavioral Domain analyses of ALE output for SUD, BA, and Pooled addiction (SUD + BA) all demonstrated significant associations (z ≥ 3.0) with reward/gain (Figure 7). Reward/gain was the most significant association for both Pooled and BA, and the only significant association for BA. For the SUD ALE, significant Behavioral Domains included: attention (z = 5.3), sexuality (z = 4.9), olfaction (z = 3.9), somesthesis. Pain (z = 3.5), and reasoning (z = 3.4). In the Overlap/Conjunction analysis of the SUD and BA ALEs, the observed cluster (caudate head; Table 6, Figure 5) showed a near‐significant association with reward (z = 2.5). In the Contrast analysis of the SUD and BA ALEs, the observed cluster (thalamus; Table 7, Figure 6) showed no significant functional associations.

4. Discussion

The overall goal of this study was to identify GM alterations associated with SUD and BA, both independently and jointly. This was accomplished by conducting eight transdiagnostic ALE CBMAs. Three primary ALEs computed cross‐study neuroanatomical convergence for: (1) Substance Use Disorders, across five diagnoses; (2) Behavioral Addictions, across two diagnoses; and (3) Pooled Addictions, across seven diagnoses (SUD + BA). These three primary ALEs were supplemented by five ALE‐to‐ALE comparison analyses: interaction (SUD + BA to SUD and BA), overlap (SUD on BA), conjunction (SUD on BA), and contrast (SUD>BA; BA>SUD). Collectively, the findings confirmed all five hypotheses: (1) SUD alone showed significant cross‐study convergences; (2) BA alone showed significant cross‐study convergences; (3) SUD alone and BA alone alteration patterns showed significant overlap and conjunction (minimal); (4) SUD + BA jointly exhibited highly significant interaction effects, with many jointly formed clusters and foci; and (5) both SUD and BA alone and jointly exhibited predominant (but not exclusive) reward‐system involvement. Although independent ALE analyses of SUD and BA only demonstrated overlap in the caudate head, additional anatomical regions (n = 24/34), not previously identified in either independent meta‐analysis, were discovered in the pooled SUD + BA comparison analysis, termed “interactive.”

The altered GM regions identified across all three primary meta‐analyses demonstrated significant associations with brain functional regions, as determined by BrainMap task meta‐data (Fox et al. 2005; Lancaster et al. 2012) For both Paradigm Class and Behavioral Domain analyses, reward showed the most consistent association with all three primary meta‐analysis. Other significant and relevant functional associations included: response inhibition, emotion induction, pain discrimination, and sexual arousal. We interpret this as confirming the expected involvement of brain reward systems in the neuropathology of addiction disorders. Concordant with this interpretation, persons with SUD and BA are known to exhibit impaired performance on cognitive tasks related to reward processing (e.g., delay discounting task) and response inhibition (e.g., Go/No‐Go task) (Dougherty et al. 2014; Acheson et al. 2019). We interpret the functional associations from Paradigm Class and Behavioral Domain analyses in this context: the observed alterations likely reflect localized pathophysiological changes, such as symptom associated overactivity or metabolic stress, which lead over time to structural GM increases or decreases. Observed gray‐matter alterations in numerous neurological and psychiatric disorders have been shown to be strongly network based (Glahn et al. 2008; Crossley et al. 2014; Goodkind et al. 2015; Vanasse et al. 2018, 2021; Chiang et al. 2019; Gray et al. 2023, 2025). Building on this perspective, in the following sections we interpret the anatomical alterations identified in our study in the context of canonical functional connectivity (FC) networks, providing a systems‐level framework for understanding the structural changes associated with SUD and BA (Laird et al. 2011; Buckner and Krienen 2013; Moring et al. 2022; Vanasse et al. 2018, 2021).

4.1. Independent Meta‐Analysis Effects: SUD and BA

Two independent transdiagnostic ALE analyses were conducted based on the two categories of addiction: (1) SUD studies (five disorders: alcohol, nicotine, stimulants, opioids, and cannabis addictions) and (2) BA studies (two disorders: pathological gambling and internet gaming disorder). The convergent regions of SUD‐related GM volume pathology included the right thalamus (medial dorsal nucleus), left caudate head, bilateral MFG (BrA 6, 9, 10 and 11), left cingulate gyrus (BrA 32), left SFG (BrA 6), and bilateral ACC (BrA 32) (Table 3 and Figure 2). Convergent regions of GM volume pathology in BA included the left ACC (BrA 32 and 33), right cingulate gyrus (BrA 24), left caudate body and head, and the left putamen (Table 4 and Figure 3). The cingulate gyrus, including the ACC (BrA 32), was a convergent region in both independent analyses, although their thresholded clusters did not overlap.

Overlap and conjunction analyses of SUD with BA (Table 6 and Figure 5) both detected significant commonality within the caudate nucleus (caudate head). The caudate nucleus is a brain region involved in reward processing, learning, and motor control (Robinson et al. 2012). This was further confirmed in the network analysis, as the caudate head showed a strong association (z = 21.5) with the Fronto‐Striatal (IC‐14) network, which has been shown to be heavily loaded on behaviors such as positive emotions (reward) and motor learning (Vanasse et al. 2021). Moreover, the caudate is intricately linked to pleasure and motivation (Haber and Knutson 2010; Watanabe and Hikosaka 2005), also implicated in our functional analyses. This interpretation aligns with evidence from meta‐analytic connectivity modeling and coactivation‐based parcellation of the caudate, which demonstrated an anteroposterior functional gradient, in which posterior regions were associated with motor functions while anterior regions, implicated here, were more strongly linked to cognitive and affective processes (Robinson et al. 2012). This convergence of neuroanatomical alterations across addiction categories and strong implication of reward‐system association suggest caudate and caudate‐connected network nodes as candidate targets for clinical trials of neuromodulatory therapies, including DBS, TMS, tDCS and FUS.

Significant regional effects were found in SUD but not in BA when examined independently and vice versa. As well, formal contrast analysis found one region (thalamus, medial dorsal nucleus; Figure 6 and Table 7) of significantly greater alteration in SUD than BA. These differences should be interpreted with caution, as they are likely impacted by the unequal sizes of the SUD (83 experiments) and BA (25 experiments) datasets (Laird et al. 2005). Larger and more balanced samples may convert them to commonalities, as in the interaction analysis (below). That said, regions associated with SUD alone (and not BA) included the MFG (BrA 6, 9, 10, and 11), SFG (BrA 6), and the medial dorsal nucleus of the thalamus. These regions are associated with higher cognitive functions, such as decision making, impulsivity, and emotion regulation (Perry et al. 2023; Euston et al. 2012; El‐Baba and Schury 2023). Furthermore, these critical functions are proposed to be central to the development and maintenance of addiction (Koob and Volkow 2010; Volkow and Baler 2014). This association was evident from the network analysis as the MFG (BrA 6, 9, 10, and 11) demonstrated a strong association with the Cognitive Control (IC‐17), Tempero‐Limbic (IC‐18), and Default‐mode (IC‐13) networks, which are involved in memory, cognition, and emotion (Vanasse et al. 2021). Moreover, impaired decision‐making may serve to inhibit abilities of individuals to distinguish between adaptive and maladaptive behavioral options, as well as the ability to consider short‐ and long‐term consequences of actions (Bechara 2005; Bolla et al. 2005; Zhang and Volkow 2019). Simultaneously, changes in dopamine signaling can increase the desire for drugs and contribute to impulsive drug‐seeking behavior (Everitt and Robbins 2016). Emotional dysregulation can further impair decision‐making abilities and increase impulsivity, leading to continued drug use, despite negative consequences (Fox et al. 2008). Understanding these interactions and their underlying neurobiological mechanisms is important for the development of interventions to target the underlying dysregulation related to decision‐making, impulsivity, and emotion regulation.

Regions identified by the BA ALE (but not SUD) included the caudate body and putamen. Results of the whole image BA (Cluster 1) paradigm class analysis indicated that these regions are significantly associated with reward. Additionally, results from the network analysis reinforced this association with reward, as the caudate body and putamen were both associated with the Fronto‐Striatal (IC‐14) network, shown to be strongly associated with reward (Vanasse et al. 2021). These findings indicate that the neurobiological effects related to reward may serve as maintenance factors of BA due to the reinforcing effects of addictive behavior, similar to the underlying effect in SUD (Koob and Volkow 2016). Behavioral approaches for BA, such as differential reinforcement strategies, may be especially warranted for BA.

4.2. Pooled SUD + BA Meta‐Analysis: Interaction Effects

The pooled ALE meta‐analysis (SUD + BA; Figure 4B and Table 5) identified a substantially greater volume of statistically significant gray‐matter alterations (70% greater, Table 2) with more clusters (6 vs. 5) and more local maxima (34 vs. 21) than the sums of the two category‐specific ALEs. The algorithmic interpretation of these emerging effects is that clusters that were sub‐threshold in category‐specific ALEs (SUD and BA) overlapped and achieved significance when pooled. That is, they are SUD‐BA interaction effects, as anticipated by the Syndrome Model of Addiction. (This finding also suggests that additional SUD‐BA commonalities will emerge as the combined literature expands).

These interactive regions consisted of the bilateral insula (BrA 13), bilateral claustrum, left SFG (BrA 10), right inferior frontal gyrus (BrA 13), bilateral lentiform nucleus (lateral globus pallidus), bilateral parahippocampal gyrus (amygdala), right parahippocampal gyrus (hippocampus), left sub‐gyral middle temporal gyrus (BrA 21), and left middle temporal gyrus (BrA 21) (Table 5 and Figure 4B). Non‐interactive regions (also present in the independent SUD/BA analyses) included the bilateral ACC (BrA 32), left cingulate gyrus (BrA 32), right MFG (BrA 6), left MFG (BrA 10), right thalamus (medial dorsal nucleus), and left caudate head. These regions are distributed across four canonical FC networks, including the default mode network (DMN), the Cognitive Control (CC, IC‐17) network (also referred to as the executive control network, ECN), the salience network (SN), and the limbic networks (Laird et al. 2011; Smith et al. 2009; Vanasse et al. 2021).

Significant interactions between SUD and BA regional alterations were found in the ACC (BrA 32), MFG (BrA 9 and 10), MTG (BrA 21), and cingulate gyrus (BrA 32); all of which are associated with the DMN (Camchong et al. 2019; Darnai et al. 2019; Kuo et al. 2019; Li et al. 2015). This association was also confirmed by the network analysis, as Clusters 1 and 5 from the pooled SUD + BA analysis had a strong association with the Default‐mode (IC‐13) network. The DMN is a large‐scale functional brain network that exhibits high activation during rest and less activation during task performance (Laird, Eickhoff, Kurth, et al. 2009; Laird, Eickhoff, Li, et al. 2009). Beyond SUD and BA (Uddin et al. 2009; Wang et al. 2017; Zhang and Volkow 2019), the DMN is implicated in multiple psychiatric disorders, such as depression, anxiety, and posttraumatic stress disorder (Mohan et al. 2016; Zhang and Volkow 2019).

The DMN also strongly interacts with subcortical areas and other large‐scale networks, most prominently with the executive control network (ECN) and the salience network (SN) (Zhang and Volkow 2019). This interaction was evident in the network analysis, as Cluster 1 of the pooled SUD + BA analysis was strongly associated with both the Default‐mode (IC‐13) network and the Cognitive Control (IC‐17) networks. Reduced efficiency of ECN and DMN reflects diminished cognitive control and self‐monitoring (Sridharan et al. 2008; Sutherland et al. 2012). Lack of cognitive control may be indicative of cognitive fusion, in which individuals lack discernment between internal thought processes and behavioral responses (Xiong et al. 2021).

Other interactive effects of SUD and BA were found within the ACC (BrA 32), MFG (BrA 6), SFG (BrA 10), IFG (BrA 13), medial dorsal nucleus of the thalamus, and insula (BrA 13), which are associated with the ECN. This was confirmed by results from the network analysis, as Cluster 1 from the pooled SUD + BA analysis had a strong association with the Cognitive Control (IC‐17) network. The ECN is implicated in addiction and is responsible for various higher cognitive functions (Fox et al. 2006; Jentsch and Taylor 1999; Menon 2011). Overall, these regions are associated with regulating limbic reward regions and higher‐order executive function, as well as working memory, decision‐making, and problem‐solving in goal‐directed behavior (Damoiseaux et al. 2006; Goldstein et al. 2009; Menon 2011). Thus, urges to engage in maladaptive behavior, lack of insight, disregard for long‐term consequences, and other common symptoms could be associated with dysregulation in these regions (Ashare et al. 2013; Cogdell‐Brooke et al. 2020; Le et al. 2021; Loughead et al. 2009; Crockford et al. 2005; Goudriaan et al. 2010).

The SN (including the insula, cingulate gyrus, IFG, thalamus, lateral globus pallidus, claustrum, and ACC) is crucial for segregating the most relevant stimuli to guide behavior, and for mediating dynamic activity between the DMN and ECN to direct attention toward internal or external events (Fedota et al. 2018; Li et al. 2017; Menon 2011; Menon and Uddin 2010; Seeley et al. 2007; Sridharan et al. 2008). Based on the network analysis, the claustrum and insula (BrA 13) do have a very strong association (z = 25.8 and 31.5, respectively) with the Salience (IC‐4) network. In addition, the insula and ACC are functionally connected with other regions, including subregions of the PFC, inferior parietal lobe (Oosterwijk et al. 2012; Peters et al. 2016; Seeley et al. 2007; Williams 2016), subcortical regions of the extended amygdala, and the ventral striatum (Sylvester et al. 2012). Based on this knowledge, the possibility exists that an individuals' inability to regulate attention and emotion, due to dysfunction in the SN, may lead to increased focus on internal events, such as drug cravings, rather than the logical consequences of maladaptive behavior. This hypothesis is consistent with previous studies that suggest the insula and ACC are crucial for psychological well‐being and adaptive functioning (Goodkind et al. 2015; McTeague et al. 2017; Sha et al. 2019; Zhang et al. 2021; Zhao et al. 2021).

Our study identified alterations in the amygdala, lateral globus pallidus, hippocampus, putamen, IFG (BrA 13), and MTG (BrA 21) indicating dysfunction in the limbic and reward networks. The association of these regions with the limbic system was confirmed through the network analysis, as Clusters 2 and 3 from the pooled SUD + BA analysis showed a strong association with the Tempero‐Limbic (IC‐18) network. The limbic system, which includes the amygdala, is responsible for processing emotions, motivation, and memory, all of which play a significant role in addiction (Koob and Volkow 2010). The reward system, which includes the ventral striatum (putamen and caudate head) and the prefrontal cortex (ACC, SFG, MFG, and IFG), is responsible for reinforcing behavior associated with positive outcomes (Koob and Volkow 2010). The amygdala is a part of the reward network through its connections to the frontal cortex, nucleus accumbens, and hippocampus (Lesage and Stein 2016). These networks are highly interconnected and work together to reinforce addiction‐related behavior (Goldstein and Volkow 2011). Alterations in these networks may contribute to the persistent drug‐seeking behavior and difficulty in abstaining from drugs or addictive behaviors seen in individuals with addiction. Morphological changes within these regions may also represent an individual's' increased attention toward the rewarding internal or social effects of the problematic substance or behavior, and a decreased motivation toward other goal‐directed behavior. Cognitive and behavioral therapies may help to clarify values, bring saliency to important and relevant goals, and alter associations with substances or maladaptive behaviors.

Overall, these results suggest shared underlying mechanisms for SUD and BA, emphasizing the role of common neural structural biomarkers across addiction‐related disorders. Pathophysiologically, we interpret these gray‐matter alterations as extensions of the network degeneration hypothesis (NDH; Seeley et al. 2009). That network‐based gray‐matter alterations are observed in psychiatric/behavioral disorders as well as neurological disorders has been demonstrated repeatedly in a now‐large literature spanning a host of diagnoses (Glahn et al. 2008; Crossley et al. 2014; Goodkind et al. 2015; Vanasse et al. 2018, 2021), with hubs of the human connectome being the primary targets of degeneration (Crossley et al. 2014). Recent studies of the NDH suggest that highly active networks undergo early hyperactivation related GM increases that may later progress to atrophy (Gray et al. 2020, 2023, 2025). This framework provides a coherent explanation for the bidirectional GM alterations observed across clusters. Given the inability to systematically account for illness stage at the individual subject level across contributing studies, these opposing volumetric effects likely reflect heterogeneity in disease chronicity, severity, and treatment exposure. Accordingly, regions exhibiting both GM increases and decreases should be interpreted not as inconsistencies but as potential markers of a dynamic, stage‐dependent neurobiological trajectory of addiction. Together, these observations suggest that dynamic, stage‐dependent GM changes may underlie shared neurobiological mechanisms across different forms of addiction, providing a framework to interpret pooled SUD and BA findings. As noted, these findings align with the transdiagnostic SMA, supporting the notion that diverse addictive behaviors may arise from a common underlying mechanism. They also underscore the importance of expanding the BA literature to more precisely characterize both shared and distinct neurobiological features of addiction. Future studies integrating structural and functional modeling may further clarify the roles of these regions and their interactions with other addiction‐related networks.

5. Limitations

Several limitations should be considered when interpreting the present findings. First, the behavioral addiction (BA) literature remains relatively small, with only 25 eligible VBM studies available for inclusion (Table 1). Although the number of experiments exceeded the recommended minimum (23) for VBM ALE analyses (Frahm et al. 2023), smaller sample sizes nevertheless impose statistical limitations. This sample size reduced statistical power and likely prevented detection of additional convergent alterations. As well, the unequal sample sizes of SUD (83 experiments) and BA (25 experiments) can be assumed to have biased the outcomes of overlap, conjunction, and contrast analyses. To mitigate this bias, a lower‐than‐typical cluster‐forming threshold of p < 0.01 (vs. 0.001 or 0.005) was applied, following a Statistical Threshold Determination Analysis (Table 2).

Second, heterogeneity in the inclusion and exclusion criteria across the included studies did not control for demographic variables, psychological comorbidities, or the severity or stage of addiction. Greater standardization of these criteria would improve classification and strengthen future reviews and meta‐analyses.

Third, there was considerable methodological variability across the included studies, including differences in software, statistical approaches, and scanners. However, such variability would be expected to attenuate (underestimate) convergent findings rather than invalidate the results. Finally, false‐positive effects are a known concern in the neuroimaging literature (Eklund et al. 2016; Sprooten et al. 2017). Coordinate‐based meta‐analysis using ALE provides a robust approach for balancing sensitivity with control of false positives (Müller et al. 2018). Nevertheless, correction for multiple comparisons in ALE can be overly conservative (Eickhoff et al. 2016), a limitation that applies not only to primary ALE results but also to ALE contrast and conjunction analyses (Kuppers et al. 2024). While the statistical stringency of FWE correction reduces Type I error, it may also eliminate true positive findings (Button et al. 2013). Importantly, unlike classical effect‐size meta‐analyses, ALE is agnostic to reported test statistics; coordinates are modeled as spatial probability distributions weighted by sample size rather than by the statistical significance of individual findings (Acar et al. 2018).

6. Conclusions

Eight transdiagnostic ALE meta‐analyses of voxel‐based morphometric reports of structural brain alterations in addiction disorders provided strong evidence of neuroanatomical alterations in common for two broad categories of addiction: substance use disorders and behavioral addictions. Regional neuroanatomical alterations across diagnostic categories were most strongly implicated in reward‐system dysregulation. These findings suggest candidate targets for neuromodulation therapeutics, with interactive regions being most likely to have benefit across diagnostic categories. Findings support the transdiagnostic Syndrome Model of Addiction by confirming that SUDs and BAs shared common brain regions and implicate common psychological and neurobiological mechanisms across diagnostic categories. Diagnostic classification systems based on objects of addiction (DSM‐5, ICD‐11, HiTOP) should take note of these observations.

Funding

This work was supported by the National Institutes of Health, GM113896, GM145432, MH074457, U19AG076581, and MH122579; Wellcome Leap as part of the Untangling Addiction program; Saudi Arabian Cultural Mission.

Supporting information

Table S1: MRI instrumentation, acquisition and analysis parameters.

Table S2: Participant demographics.

Data Availability Statement

The data that support the findings of this study are openly available in BrainMap Database at http://brainmap.org.