Introduction

Methamphetamine (METH), the second most commonly used illicit psychostimulant, affect approximately 7.4 million people worldwide, as reported in the 2018 UN Office on Drugs and Crime World Drug Report [1]. METH use is strongly linked to increased mortality [2], higher incidence of HIV and hepatitis infections [3], deteriorated mental health (including psychosis, depression, suicidal tendencies, and violent behavior) [4], and heightened risk of cardiovascular complications [5]. One of the most severe psychiatric consequences of METH use is methamphetamine-induced psychosis (MAP), with a prevalence rate of 43% among individuals diagnosed with METH use disorders (MUD) [6]. MAP is characterized by transient schizophreniform psychotic symptoms resulting from repeated high-dose intoxication [7], and it leads to a range of negative outcomes, including risky behaviors, family breakdown, housing instability, as well as medical and legal challenges [8].

Neuroimaging studies have indicated that chronic METH use results in widespread structural and functional brain abnormalities [9]. For instance, Wang et al. discovered that reduced cortical thickness in the right lateral orbitofrontal cortex, decreased volumes in the left pars triangularis gyrus, and increased volumes in subcortical regions such as the pallidum, hippocampus, and amygdala among METH users compared to healthy controls [10]. Similarly, Luo et al. found that, compared to healthy controls, individuals with MUD showed increased connectivity in the salience network and decreased connectivity in the default mode network [11]. In addition, a review summarized that MAP-related brain alterations were mainly located in the frontal lobe (especially the orbitofrontal cortex), striatum, and limbic systems (including amygdala and hippocampus) [9]. Moreover, large-scale studies by the ENIGMA Addiction Working Group have demonstrated subregional volumetric alterations in the hippocampal and amygdala among individuals with cannabis or alcohol dependence [12, 13], highlighting the importance of limbic substructures in substance use disorders. However, it has remained difficult to delineate the specific regions within these complex structures in detail.

Craving is a core symptom of addiction and a pivotal trigger for relapse [14]. However, craving remains difficult to measure due to its inherent subjectivity and the influence of contextual factors [15]. A better understanding of its neurobiological substrates is crucial to drive forward the development of more effective addiction interventions.

Together, these studies suggest that alterations in cortical and limbic brain structures may contribute to the pathophysiology of MAP. However, the specific cortical regions and their relationship with psychotic symptoms and craving remain poorly delineated. Therefore, the present study aimed to: (1) examine shared and distinct cortical patterns of neurostructural abnormalities among individuals with MAP, those with METH dependence but without psychosis (MA), and healthy controls (HCs); (2) explore the relationship between regional brain measures and clinical features in MUD patients, especially craving intensity.

Methods

Participants

We enrolled patients diagnosed with Methamphetamine Use Disorder (MUD) from the Chengdu Compulsory Detoxification Center between January 2018 and October 2019. All patients were diagnosed by a qualified psychiatrist using the Chinese version of the Structured Clinical Interview for DSM-5 (SCID-5) [16]. The patient cohort was divided a priori into two distinct subgroups, MAP and MA, based on the following diagnostic criteria. MAP was diagnosed based on the DSM-5 criteria for“Substance/Medication-Induced Psychotic Disorder”. Specifically, the presence of hallucinations and/or delusions occurring during or shortly after methamphetamine use was required, which could not be attributed to delirium or better explained by another primary psychotic disorder. Two experienced psychiatrists independently made the diagnosis in a blinded manner, with any discrepancies resolved through a consensus meeting. The MA group included individuals who met the DSM-5 diagnostic criteria for MUD but did not exhibit psychotic symptoms. A control group of healthy volunteers was recruited via advertisements. Controls were carefully matched to the MUD group based on age and ethnicity.

All participants met the following inclusion criteria: (1) aged 18 to 55; (2) male; (3) Han Chinese ethnicity; and (4) right-handed. Exclusion criteria for all participants included: (1) any past or present use of substances other than METH or nicotine; (2) any current or past psychiatric disorders unrelated to substance use; (3) any history of traumatic brain injury or loss of consciousness for 10 min or more; (4) severe general medical illness; and (5) contraindications to magnetic resonance imaging (MRI) or conditions that could potentially affect MRI data quality. Additional exclusion criteria for HCs: (1) current or previous diagnosis of MUD; (2) use of psychotropic medications; and (3) first- or second-degree relatives with a history of psychotic disorders.

The study was approved by Biomedical Research Ethics Committee, West China Hospital, Sichuan University (Date: Sep 1st, 2017/No: 2017 − 449) and it adhered to the tenets of the Declaration of Helsinki. Prior to entering the study, participants provided written informed consent after receiving a comprehensive description of the study.

Study design

All participants underwent the following measurements: an MRI scan and a clinical interview conducted by professional psychiatrists. The interviews provided data regarding age, ethnicity, years of education, smoking status, drinking status, initial age at which METH was used, duration of METH use, METH maximum dose of a single use, frequency of METH use, and duration of METH abstinence. A history of METH use was verified by urine toxicology, which was performed immediately after the participants entered the agencies. The SCID-5 Interview was administered in the healthy control subjects to rule out the presence of any psychiatric disorder.

Nicotine dependence was evaluated using the reliable and validated Fagerström Test for Nicotine Dependence (FTND), which consists of six items, with scores ranging from 0 to 10 [17].

The degree of METH craving was gauged utilizing the visual analog scale (VAS) [18]. Participants were presented with a horizontal line, anchored at 0 (“no craving”) and 10 (“the most intense craving ever experienced”). They were instructed to visualize a personal drug-use scenario and then mark on the line to indicate their current level of craving. The score was determined by measuring the distance from the left end of the line to the participant’s mark, which was subsequently converted to a numerical value ranging from 0 to 10. VAS stands as the predominant tool for evaluating subjective [15], showing robust reliability and validity.

The psychotic symptom severity of MAP was assessed using the Positive and Negative Syndrome Scale (PANSS), a widely adopted and standardized clinical instrument for evaluating schizophrenia and substance-induced psychosis [19, 20]. The PANSS consists of 30 items divided into three subscales: the Positive Subscale (PANSS-P), the Negative Subscale (PANSS-N), and the General Psychopathology Subscale (PANSS-G).

MRI acquisition and preprocess

All MRI scans were conducted using a 3-T scanner (Achieva; Philips, Amsterdam, the Netherlands) equipped with an 8-channel head coil. Participants were instructed to relax, keep their eyes closed, remain still, and stay awake without engaging in specific thoughts during scanning. High-resolution T1-weighted images were obtained using a 3D magnetization-prepared rapid gradient-echo (MPRAGE) sequence with the following parameters: FOV = 240 mm × 240 mm, TR = 8.37 ms, TE = 3.88 ms, flip angle = 7◦, in-plane matrix resolution = 256 × 256, and number of slices = 188. The quality of the brain images was assessed immediately following each scan, and scans were repeated if significant distortions were detected. The standard FreeSurfer “recon-all” pipeline was utilized to generate morphometric data for each participant (https://surfer.nmr.mgh.harvard.edu/fswiki/ReconAllDevTable).

The image processing pipeline incorporated several key steps: (1) motion correction to minimize scan artifacts; (2) intensity normalization for signal uniformity across tissues; (3) spatial transformation to Talairach coordinate space; and (4) skull stripping to remove non-brain tissues. Subsequent processing involved (5) subcortical white matter segmentation, followed by (6) tessellation of the gray matter/white matter (GM/WM) boundary and (7) automated topology correction. Surface reconstruction was achieved through (8) cortical surface deformation and inflation, as detailed in previous methodological studies [21, 22]. Cortical thickness was computed as the distance between the GM/WM boundary and pial surface at each vertex across the cortical mantle, representing local thickness in millimeters [23]. The resulting thickness maps were smoothed using a 10 mm full-width at half-maximum (FWHM) Gaussian kernel to reduce inter-subject variability while preserving anatomical precision. Finally, a 3D surface model of the cortex was generated by integrating intensity gradients and spatial continuity information from the reconstructed boundaries. Quality control was performed at each stage to ensure anatomical accuracy.

Analysis

Demographic characteristics and clinical parameters underwent comparison utilizing analysis of variance (ANOVA), Mann-Whitney nonparametric U-tests, or Chi-Squared tests. In the vertex-wise analysis of cortical measures, including surface area, thickness, and volume, a general linear model was applied at each vertex to compare the three groups using Freesurfer’s command “mri_glmfit”. For surface area and volume analyses, both age and total intracranial volume (TIV) were included as covariates. For cortical thickness (CT), however, only age was included as a covariate, since previous research does not support the adjustment for TIV [24]. The analysis was conducted separately for the left and right hemispheres. To correct for multiple comparisons at the vertex level, we conducted 5,000 permutation-based Monte Carlo null-z simulations using Freesurfer’s “mri_glmfit_sim” command. A cluster-forming threshold of p < 0.001 (sig = 3) was applied, which corresponded to a cluster-wise significance threshold of p < 0.05 in our exploratory study. Multivariate analysis of covariance (MANOVA) was employed assess significant brain regions, with subsequent post hoc analyses conducted. P-values underwent correction for multiple comparisons utilizing the Family-Wise Error (FWE) method. For account for potential confounding effects, all significant clusters were adjusted for relevant demographic and anatomical covariates prior to group comparisons. Specifically, cortical thickness measures were regressed against age and years of education. For volumetric and area measures were adjusted for age, years of education, and TIV to control for differences in overall brain size. The residuals from these regression models, representing variance independent of the specified covariates, were subsequently utilized in the analyses. Group differences in these adjusted neuroimaging measures were examined using one-way analysis of variance (ANOVA). For measures showing main effects of group, post-hoc pairwise comparisons were conducted with Bonferroni correction method to control the family-wise error rate. To further assess robustness, a sensitivity analysis was conducted by additionally adjusting for FTND scores in the regression models when comparing MA and MAP groups for the significant regions.

The correlation between clinical parameters and neurostructural signatures was investigated via correlation analysis with Bonferroni correction. Statistical analyses were conducted using commercially available software SPSS (Inc., Chicago, IL, Ver. 24.0) and R (4.4.2), with significance defined as a two-tailed P-value < 0.05.

Results

Demographics and clinical chrematistics of all subjects

Out of the 171 subjects initially enrolled, 16 individuals were excluded due to poor MRI images, incomplete the scanning procedure, or missing MRI or clinical data. Therefore, a total of 155 male participants were included in the final analysis, comprising 94 patients with MUD (62 MA and 32 MAP) and 61 HCs.

The baseline demographic and clinical characteristics of the participants are summarized in Table 1. The mean age of MA, MAP and HCs were 30.65 (7.31), 26.16 (8.78) and 27.7 (7.40) years, respectively. And the mean education years of MA, MAP and HCs were 9.10 (2.98), 9.84 (3.13) and 15.70 (3.16) years, respectively. Compared with the MA group, the MAP group showed younger age at initial METH use (Z = 2.714, p = 0.006), higher maximum single-use METH dose (Z = 3.442, p < 0.001), higher score on the FTND (Z = 3.978, p < 0.001), and higher VAS scores (Z = 2.639, p = 0.008). No significant differences METH use history (duration, frequency, abstinence) (p > 0.05). The total PANSS score in the MAP group was 61.66 ± 3.39. The subscale scores were as follows: positive subscale (PANSS-P) 25.09 ± 2.08, negative subscale (PANSS-N) 9.47 ± 0.57, and general psychopathology subscale (PANSS-G) 27.09 ± 1.47.

Table 1 Demographic and clinical variables between HCs and MUD group
Full size table

Vertex-wise analysis of MA, MAP and HCs

As depicted in Fig. 1; Table 2 vertex-wise analysis found significant differences in cortical thickness and volume across the three groups (MA, MAP and HCs) after 5,000 permutation-based Monte Carlo null-z simulations and a cluster-forming threshold of p < 0.001 applied, while no differences were observed in surface area.

Fig. 1
Fig. 1
Full size image

Results of vertex-wise analysis of surface-based morphometry between HC, MA, and MAP groups. Notes: cMFG: caudal middle frontal gyrus; HC: healthy controls; MA: methamphetamine dependency without psychosis; MAP: methamphetamine-associated psychosis; rACC: rostral anterior cingulate cortex; SPC: superior parietal cortex; STG: superior temporal gyrus. For the post-hoc analysis, * indicated p < 0.05; ** indicated p < 0.01, *** indicated p < 0.001 (Bonferroni corrected)

Table 2 Results of vertex-wise analysis of surface-based morphometry between HC, MA, and MAP groups
Full size table

As showed in Tables 2 and 3, two clusters revealed significant differences in CT among the three groups after adjusted age and education years, specifically located in the left superior temporal gyrus (STG) (Fig. 1A) and right rostral anterior cingulate cortex (rACC) (Fig. 1B). Post-hoc analyses revealed that both HCs and MA exhibited greater CT than MAP in the left STG (all p < 0.05, Fig. 1A) and the right rACC (all p < 0.05, Fig. 1B). No significant difference was observed between the MA and HCs groups in either region (p > 0.05, Fig. 1A-B).

Table 3 Clusters from vertex-wise analysis in the MA, MAP, and HC groups
Full size table

As showed in Tables 2 and 3, two clusters revealed significant differences in cortical volume among the three groups after adjusted age, education years and TIV, specifically located in the left caudal middle frontal gyrus (cMFG) (Fig. 1C) and right superior parietal cortex (SPC) (Fig. 1D). Post-hoc analyses revealed that the MA group exhibited decreased cortical volume in both cMFG (Fig. 1C) and SPC (Fig. 1D) compared to the HCs group (all p < 0.05), while the MAP group showed no significant difference from the HCs group in either region (all p > 0.05).

In the sensitivity analysis, after additionally adjusting age, education years and total score of FTND, the MA group still demonstrated significantly greater CT than the MAP group in left STG and right rACC (all p < 0.05, Table S1).These findings indicate that the key regional differences remained robust even when accounting for potential confounders.

Association between clinical features and MRI features in MUD

Figure 2 shows the results of the correlation analysis. Left cMFG volume was positively associated with VAS score in MUD (r = 0.341,95% CI = 0.161, 0.530, p < 0.001; Figure S1C), and this association remained significant after applying the Bonferroni correction. In contrast, no significant associations were found between VAS scores and the left STG thickness, right rACC thickness, or right SPC volume (Figure S1A, B and D, all p > 0.05). Furthermore, there were no significant correlations were observed between any cortical measure and other clinical variables, including METH use duration, METH maximum dose per single use, METH use frequency in the past month, or FTND scores (Table S2). In MAP, there is no significant associations between MRI features and total score of PANSS, PANSS-P, PANSS-N and PANSS-G (all p > 0.05).

Fig. 2
Fig. 2
Full size image

Correlation between significant regions and VAS score in MUD. Notes: cMFG: caudal middle frontal gyrus; METH: Methamphetamine; MUD: Methamphetamine use disorders; VAS: Visual analog scale

Discussion

In this study, we conducted surface-based morphometric analyses and identified both shared and distinct cortical abnormalities among MA, MAP, and HC groups. Our findings revealed that MAP exhibited significant cortical thinning in the left STG and right rACC compared to both MA and HCs. In contrast, compared to HCs, MA patients exhibited decreased cortical volume in cMFG and SPC. Additionally, left cMFG volume was positively associated with VAS score, although effect size was modest (r = 0.341), highlighting the need for further exploration.

The observed thinner CT in the left STG among MAP patients aligns with some prior findings, though was not entirely consistent. For instance, previous studies using voxel-based morphometry (VBM) have reported gray matter (GM) reduction in the left STG among METH users and individuals with MAP [25, 26]. And a voxel-wise meta-analysis has also revealed convergent STG alterations across behavioral and addiction-related disorders [27]. STG abnormalities are not specific for substance addiction disorders, but also observed in suicide brain [28], clinical high-risk (CHR) states for psychosis [29], and schizophrenia [30]. The STG is involved in auditory processing [31] and, together with a broader temporal-frontal-parietal network, plays a role in language production, interpretation, and self-monitoring [32, 33]. In light of these studies, our findings show that the MAP group exhibited thinner CT in the STG, suggesting its central role in processes underlying positive symptoms. Furthermore, a recent study using surface-based morphometric analysis revealed cortical discrepancies in the left inferior temporal gyrus, lateral and medial orbitofrontal cortex, inferior parietal lobule, and lateral occipital cortex in MAP [34]. Our results differ, possibly due to the larger sample size, inclusion of only male participants, and lower average age in our study.

Thinner CT was only observed in the rACC among individuals with MAP, which aligned with existing literature that highlights the neurobiological changes associated with MAP and its impact on brain structure [35]. The ACC situated at the interface between the prefrontal cortex and limbic system, is critical for integrating emotional and cognitive processes [36]. The rACC, located around the genu of the corpus callosum, shows structural alterations in various psychiatric conditions such as schizophrenia [37], anxiety disorder [38], mild cognitive impairment [39], alcohol use disorder [40], and MUD [41]. These alterations may be associated with the psychotic symptoms observed in MAP. Previous studies suggest that chronic METH use can lead to significant brain changes, particularly in regions associated with psychosis and cognitive deficits. For instance, dopamine transporter density has been associated with the severity of psychotic symptoms, suggesting that dopaminergic dysfunction as a possible mechanism underlying rACC abnormalities [42]. Interestingly, our results indicated no significant difference in CT between MA and HCs in the rACC, suggesting that these structural changes may be more specifically related to the presence of psychotic symptoms rather than METH use alone. This finding is consistent with the notion that not all METH users develop psychosis, and that factors such as genetic vulnerability and environmental stressors may modulate the risk [43]. Overall, our findings contribute to the growing evidence on neurobiology of MAP, highlighting the role of cortical thickness in the rACC in psychosis. Further research is needed to clarify the long-term impact of these structural changes on cognitive and emotional functions in MAP.

We also observed that MA group exhibited decreased cortical volume in cMFG and SPC, consistent with existing linking METH use to executive dysfunction and attention deficits [29, 44]. The left cMFG is anatomically located within the dorsolateral prefrontal cortex (DLPFC), is essential for decision-making, cognitive control, and emotional regulation [45, 46]. Besides, the SPC has been associated with working memory, attention and visuospatial perception as well as on-line control of visually guided movements [47,48,49]. These findings may reflect shared pathophysiological mechanisms of MA, affecting cognitive functions such as attention, emotion regulation, and decision-making. Notably, MAP patients did not exhibit significant cortical volume reductions in these regions compared to HCs, despite marked reductions in the MA group. This may suggest that psychotic symptoms do not further impact these regions beyond the effects of METH use alone. Alternatively, individual variability, compensatory mechanisms, or the relatively small sample size in the MAP group may have subtle effects. Prior studies indicate that cortical alterations may differ between METH users with and without psychosis, possibly involving distinct neural circuits underlying emotional or psychotic symptoms [35, 41]. In addition, structural MRI studies are known to be sensitive to inter-subject variability and sample size limitations, which can reduce the power to detect group-level differences, especially in cross-sectional studies [50]. These findings highlight the need for cautious interpretation of null results and support future studies with larger, longitudinal samples.

In addition, we examined a positive correlation between the left cMFG volume and craving severity, consistent with prior evidence implicating the prefrontal cortex in craving and relapse risk [51]. Interestingly, although MA patients exhibited a group-level reduction in cMFG volume compared to healthy controls, within the MUD cohort, larger cMFG volume was associated with stronger craving. This apparent paradox may reflect heterogeneity and compensatory or maladaptive mechanisms in frontal regions under addiction. While cMFG volume is generally reduced in MA users, individual variability allows for positive correlations with craving, suggesting that larger volumes may indicate compensatory engagement rather than structural normalization. Prior studies suggest that prefrontal regions involved in cognitive control can be co-opted to sustain craving rather than inhibit it, indicating that cMFG–craving associations may involve complex, non-linear neuroadaptations [52, 53]. As a key component of the prefrontal cortex, the cMFG is associated with functions like cognitive control, decision-making, and emotional regulation, and may play a key role in modulating the severity of craving symptoms [54]. The left cMFG has extensive neural connections with reward-related circuits such as the nucleus accumbens and ventromedial prefrontal cortex. Kober et al. (2015) reported that neural activity in the ventral striatum, amygdala and ventromedial prefrontal cortex/ anterior cingulate cortex (vmPFC/ACC) is associated with cue-induced craving, key regions for reward processing [55]. Increased left cMFG volume may strengthen interactions with the reward system, intensifying cravings by promoting overfocus on craving-related stimuli, despite the cMFG’s role in cognitive control. Further research is needed to fully understand how these neural mechanisms contribute to addiction and craving. Notably, while our study focused on craving-related neural mechanisms, we also explored associations with psychotic features. These features did not show significant correlations with neuroimaging or clinical measures, which may reflect their relatively mild and transient nature in our cohort, as well as the modest sample size. Future studies with larger samples and longitudinal designs will be important to further delineate the neurobiological correlates of substance-induced psychosis.

This study has several limitations. First, one notable limitation of this study is the restricted demographic composition of our sample, which consisted exclusively of male participants of Han Chinese ethnicity. This homogeneity may limit the generalizability of our findings to broader populations, particularly those encompassing diverse genders and ethnic backgrounds. It is well-established that neurobiological mechanisms related to craving can be modulated by sociocultural and biological factors linked to gender and ethnicity [56, 57]. Future studies employing stratified sampling across multiple genders and ethnic groups are warranted to enhance the external validity and confirm the broader applicability of these results. Second, all MUD participants were recruited from a Compulsory Detoxification Center, with an average abstinence duration of over 8 months, which may limit the generalizability of our findings. Third, craving was assessed using a single-item VAS, which lacks the multidimensional assessment provided by validated multi-item craving instruments and may be susceptible to floor effects, as indicated by the generally low craving scores in this cohort. Future studies would benefit from employing comprehensive, psychometrically robust craving questionnaires to enhance measurement sensitivity. Forth, the relatively small sample size, particularly in the MAP group, reflects recruitment challenges and calls for validation in larger cohorts. Lastly, the cross-sectional design prevents causal inferences, and it remains unclear whether the observed cortical changes are pre-existing or drug-induced, highlighting the need for longitudinal studies.

Conclusions

In conclusion, patients with MAP exhibit region-specific alterations in cortical thickness in the superior temporal gyrus and limbic system. These results provide empirical evidence for biological basis of brain structural changes and suggest potential targets for treatment of psychotic symptoms in MAP. More importantly, identified specific atrophy patterns would contribute to our understanding of the pathogenesis of MAP.