The impact of chronic heroin addiction on creative cognition: an EEG study based on divergent thinking Authors: Wenjuan Fu, Yifan Wang, Wanyi Li, Yujia Meng, Jiaqi Dang, Kai Yuan, Haijun Duan Published: 2025 ABSTRACT AbstractChronic heroin addiction induces severe physiological dependence and systematically impairs the neural mechanisms underlying basic cognitive functions. However, the specific effects of opioids on divergent thinking—a higher-order cognitive function—and its neural basis remain underexplored. This study recruited individuals with chronic heroin use disorder (HUD, n = 38) and healthy controls (HC, n = 35) to record neural activity during divergent thinking tasks (Alternative Uses Test, AUT) using electroencephalography (EEG). Source-space analysis quantified power spectral density (PSD) in brain regions, complemented by functional connectivity analysis using weighted phase lag index (wPLI) from selected seed regions. Machine learning assessed the utility of connectivity metrics as cognitive impairment markers. Source-space analysis revealed that compared to the HC group, the HUD group exhibited increased alpha-band power in the left precuneus (L.PCUN) and left superior parietal lobule (L.SPL) and beta-band power in the right superior parietal lobule (R.SPL). Functional connectivity analysis revealed weakened cross-network coupling between the default mode network (DMN) and frontoparietal control network (FPN), with reduced alpha connectivity between the right superior frontal gyrus (R.SFG) and L.PCUN correlating with impaired divergent thinking. Machine learning confirmed this metrics as effective neurobiological markers (AUC = 0.772, accuracy = 0.767) of cognitive impairment in heroin addiction. The findings indicate that abnormal local oscillatory activity and disrupted network integration impair the dynamic “generation-evaluation” loop of divergent thinking, constituting a compensatory pathological mechanism. This study provides neural evidence linking opioid addiction to creativity deficits and supports precision interventions targeting cognitive dysfunction in addiction.Subject terms: Neuroscience, Human behaviour Introduction Opioid abuse has become a global public health crisis, with heroin, a potent μ-opioid receptor agonist, being a leading contributor to drug overdose fatalities [1]. During withdrawal, individuals with drug addiction often encounter stress-inducing events such as interpersonal conflicts or exposure to drug-related cues, necessitating flexible cognitive strategies to suppress cravings. Creative cognition, as defined by Bloom’s taxonomy, represents the highest and most complex level of cognitive abilities. It involves restructuring problems and generating solutions through novel insights [2]. Guilford’s Structure of Intellect model further indicates that divergent thinking is a key component of creative cognition and is widely regarded as an indicator of creative potential [3, 4]. Divergent thinking can help individuals with substance use disorders overcome cognitive rigidity through open-ended thought processes. It emphasizes multi-perspective thinking, thereby promoting emotional regulation, psychological flexibility, and shifts in cognitive strategies [5, 6]. However, while the detrimental effects of heroin addiction on basic cognitive functions have been extensively studied [7, 8], its impact on higher-order cognitive abilities remains underexplored and poorly understood. This gap in understanding limits our ability to address adaptive challenges during addiction withdrawal and hinders improvements in treatment acceptance, efficacy, and sustainability. Substance addiction is a complex and chronic brain disorder that cannot be simply attributed to a lack of willpower; rather, it is closely associated with profound alterations in brain structure and function [9, 10]. Heroin exerts its effects primarily by activating μ-opioid receptors [11] and modulating the midbrain-limbic-prefrontal dopamine system [12], leading to extensive functional inhibition and structural atrophy in the prefrontal cortex [13, 14]. Additionally, chronic heroin dependents exhibit significantly reduced striatal dopamine transporter levels, and this change may result in long-term impairment of dopamine system function [15]. The prefrontal cortex is a core brain region supporting cognitive flexibility and divergent thinking [16], while the dopamine system plays a crucial role in driving motivation and creativity. When dopamine levels decline, individuals tend to enter an “exploitative mode,” relying on existing cognitive pathways, which limits their creativity [17]. These overlapping neural mechanisms suggest that creative cognition may exhibit a higher vulnerability to addiction-related brain function damage. Furthermore, individuals with heroin use disorder (HUD) often exhibit hyperactivity in the reward system and impaired functionality in the executive control system [18], making them more prone to cognitive rigidity and deficits in goal-directed abilities [19, 20]. These impairments hinder the establishment of self-regulation strategies and the recovery of social functioning during the withdrawal process. Although the relationship between drug use and creativity has been discussed in public discourse and some behavioral studies [21–23], existing evidence is largely anecdotal and primarily focuses on the short-term effects of substances. For instance, psilocybin has been shown to enhance spontaneous creative insights temporarily while impairing performance on intentional creative tasks [24]; cocaine enhances divergent thinking in graphical tasks but impairs “flexible divergent thinking” in verbal tasks [25]; and high-potency cannabis disrupts divergent thinking in regular cannabis users [26]. In contrast, chronic heroin addiction has more profound and long-lasting neurophysiological effects, extensively affecting the fundamental cognitive functions underlying divergent thinking [27]. However, there is a notable lack of systematic research into the neurophysiological mechanisms underlying the effects of chronic heroin use on divergent thinking. This research gap not only limits our understanding of how heroin addiction disrupts the processes of divergent thinking but also constrains the scientific development and optimization of neurobiologically informed rehabilitation interventions. Divergent thinking involves rapid, dynamic cognitive processes, and the millisecond-level temporal resolution of EEG enables precise tracking of neurophysiological changes throughout tasks. This capability reveals the temporal dynamics of multiple cognitive functions and inter-regional communication underlying creativity [28]. In addiction research, EEG plays a crucial role in monitoring cognitive function, identifying brain dysfunction, and exploring potential biomarkers [29–32]. Studies have consistently shown significant abnormalities in the EEG power spectrum of individuals with HUD. Compared to healthy controls, they exhibit increased delta, theta, and beta activity, and attenuated or slowed alpha activity in the frontal, central, and temporal regions [33–35]. Power spectral density (PSD) analysis quantitatively assesses abnormal EEG activity in specific frequency bands, reflecting impairments in core cognitive functions supporting divergent thinking. For instance, enhanced alpha waves facilitate internal attention and autonomous imagination [36, 37], moderate reductions in theta power improve cognitive control [38], increased delta waves support sensory suppression and sustained internal attention [39], and elevated beta power promotes divergent thinking [40]. Beyond local neural oscillations, emerging research highlights the disruption of brain network integration caused by opioid dependence. Specifically, the coordination among the default mode network (DMN), frontoparietal network (FPN), and salience network (SN) is often impaired [41–44]. These networks are critical for the flexible switching between internal information generation and external monitoring required for creative thinking [45]. Traditional EEG analysis methods are limited in spatial resolution, as electrode-level analyses hinder precise localization of impaired brain regions [46]. To address this, the present study employed source localization to reconstruct cortical activity by solving the inverse problem, providing a theoretical basis for identifying precise neuromodulation targets [47]. Moreover, conventional group-level comparisons often fail to establish robust links between neural markers and creative performance. Machine learning algorithms, particularly support vector machines (SVMs), offer superior pattern recognition and have shown strong efficacy in biomarker identification within addiction neuroscience [48]. This study employed multimodal EEG analysis to systematically investigate the neural mechanisms underlying divergent thinking impairments in individuals with chronic heroin addiction. Divergent thinking was assessed using the Alternative Uses Task (AUT). EEG scalp signals were source-localized, and PSD was computed to identify abnormalities in key regions such as the prefrontal and parietal cortices across delta, theta, alpha, beta, and gamma frequency bands. Based on the PSD results, regions of interest (ROIs) and frequency bands were selected to compute the weighted phase lag index (wPLI), quantifying functional connectivity patterns between key brain regions across different frequency bands. Additionally, the study integrated a support vector machine (SVM) model to identify neural markers with classification capabilities, providing empirical support for developing neurophysiology-based rehabilitation interventions. It is worth noting that causal inference serves as a key methodological approach in cognitive neuroscience for elucidating the relationship between brain function and cognitive behavior [49]. Given the cross-sectional observational design of the present study, the aim was to identify the neurophysiological associations between divergent thinking and EEG features in HUD [50], thereby laying a theoretical foundation for future intervention research and investigations into causal mechanisms. Methods Participants The experiment included a total of 73 participants: 38 individuals with chronic heroin use disorder (HUD), recruited from multiple drug rehabilitation centers under mandated isolation treatment (age: M = 47.37 years, SD = 8.63; 11 females), and 35 healthy controls (HC; age: M = 46.20 years, SD = 8.15; 18 females). A priori power analysis using G*Power 3.1.9.7 indicated that a total of 72 participants would be required to detect a medium-to-large effect size (Cohen’s d = 0.60) with α = 0.05 and power = 0.80, suggesting our sample size was sufficiently powered. The HUD group met the diagnostic criteria for opioid use disorder (OUD) outlined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). The HC group was locally recruited and excluded individuals with a history of brain injury, epilepsy, alcoholism, drug addiction, or psychiatric disorders. Additionally, individuals with comorbid substance use disorders (e.g., methamphetamine), recent use of psychotropic medications, or a personal or family history of psychiatric disorders were excluded. All participants were right-handed, with normal intellectual function and comparable educational levels. Informed consent was obtained from all participants, and the study received ethical approval from the Ethics Committee of Shaanxi Normal University. To minimize the influence of intergroup differences, statistical analyses were conducted on age, gender, and educational background, revealing no significant differences between the HUD and HC groups (see Supplementary Materials Table S1). In addition, the severity of addiction was assessed using the Opioid Addiction Severity Inventory (OASI) [51]. Experimental task and procedure The AUT assesses divergent thinking by asking participants to generate novel uses for common objects (e.g., brick, flowerpot, ruler)[52, 53]. Fifteen objects were presented in a randomized order. Each trial included a 10-second fixation, a 4-second stimulus display, and a 10-second idea generation phase indicated by a question mark. Participants then verbally reported their ideas within 4 seconds [54] and rated each idea’s originality on a 5-point scale (1 = non-original, 5 = highly original) (Fig. 1). Instructions were delivered visually and aurally using E-Prime 2.0 (Psychology Software Tools, Inc., Sharpsburg, PA, USA). Procedure of the AUT task. Each trial began with a fixation cross presented for 10 s (reference period). Then, a stimulus word (e.g., brick) appeared on the screen for 4 s. During the subsequent 10 s (activation period), while a white question mark was displayed, participants were required to generate useful and original ideas related to the stimulus. Next, the presentation of “response” prompted participants to verbally articulate their most original idea within 4 s. Finally, participants were asked to evaluate their response as either “original” or “not original” by pressing the corresponding key on a keyboard. EEG recording and equipment EEG signals were recorded using the Neuracle system with a 64-channel wireless setup based on the 10-20 system (NeuSen.W64, Neuracle, China), with a sampling rate of 1000 Hz. The Cz electrode served as the reference, and the ground electrode was placed on the forehead. Impedance was maintained below 5 kΩ for each sensor throughout the experiment. Researchers carefully monitored and recorded potential artifacts, such as coughing, sneezing, noticeable muscle movements and external noise. No participants exhibited significant artifacts caused by such disruptions. This setup enabled the accurate and reliable recording of neural activity across multiple brain regions. Data processing All EEG signal processing was conducted in MATLAB R2021b. The EEGLAB toolbox [55] was used for EEG data preprocessing, custom MATLAB scripts were employed for spectral analysis and functional connectivity calculations, and source localization was performed using the Brainstorm toolbox [56]. EEG data preprocessing To ensure data quality, EEG preprocessing was performed using EEGLAB [55]. The data were re-referenced to a common average reference. A 0.1–30 Hz band-pass filter was applied, and data were downsampled to 500 Hz. Artifacts exceeding ±75 µV, including ocular movements [57], were removed, followed by ICA to eliminate residual noise from blinks, muscle activity, and other sources. Task-related EEG segments from 1–9 s during the idea generation phase [58] were extracted for subsequent source reconstruction, spectral analysis, and functional connectivity analysis. This pipeline optimized signal quality for robust investigation of creative neural processes. Source analysis EEG source localization aims to identify the neural sources of brain activity recorded via scalp EEG, allowing for the determination of the specific brain regions that generate the recorded signals [59]. In this study, we performed source localization using the Brainstorm toolbox, with the Desikan-Killiany-Tourville (DKT) atlas as the brain region template. This atlas divides the brain into 62 regions that cover the entire cortical surface. The analysis procedure involved several steps: First, electrode localization was performed to establish the positions of the EEG electrodes. Next, a head model was computed to represent the geometry of the participant’s brain and surrounding tissues. Covariance estimation was then conducted to calculate the data and noise covariance, which is essential for accurate source reconstruction. Finally, source estimation was carried out using the Linearly Constrained Minimum Variance (LCMV) beamformer algorithm, which allowed us to localize the sources of brain activity with high spatial precision [60]. Spectral analysis EEG spectral analysis is a technique used to investigate brain functional states by examining the frequency components of brain electrical activity. This study employed source-based spectral analysis. Specifically, the scalp EEG signals were mapped to regions of interest (ROIs) time series within a standardized brain space, defined by the DKT atlas comprising 62 cortical regions. This mapping is achieved by solving the EEG inverse problem, thereby obtaining source-level activity data (dipole × time) for each virtual dipole. Following the extraction of the time series for each ROI, the Power Spectral Density (PSD) was estimated using Welch’s method. The parameters were set as follows: a sampling rate of 500 Hz, a 1-second Hamming window (500 points), 50% window overlap, and a fast Fourier transform (FFT) length of 256 points. PSD was computed in the linear scale and subsequently used to extract power values within specific frequency bands. Relative power was calculated for five canonical frequency bands: delta (1–5 Hz), theta (5–8 Hz), alpha (8–12 Hz), beta (12–20 Hz), and gamma (20–30 Hz). This source-space approach effectively reduces volume conduction effects [61, 62], enables more precise localization of cortical oscillatory activity, facilitates the identification of abnormal regions associated with cognitive deficits, and provides a potential basis for targeted interventions. Functional connectivity Functional connectivity refers to the interaction patterns between brain regions based on the synchrony of neural activity. In this study, we employed the weighted phase lag index (wPLI) to assess phase synchronization in the source space. wPLI reduces spurious connections caused by volume conduction and provides a more accurate representation of true functional connectivity [63, 64]. Specifically, seed regions were defined based on cortical areas and frequency bands that showed significant differences in power spectral density (PSD) following source analysis. The wPLI was then calculated between the seed regions and all other cortical ROIs to construct functional connectivity maps. Statistical analysis All participants completed the Alternative Uses Task (AUT) during EEG recording, with their responses documented by the experimenter and subsequently rated for originality on a 5-point scale. A 2 (Rating Type: subjective vs. objective) × 2 (Group: HUD vs. HC) repeated-measures analysis of variance (ANOVA) was conducted to examine group differences in divergent thinking performance. Based on signal detection theory [65], sensitivity (d′) and response bias (β) were calculated from the subjective originality ratings using response classification derived from scoring accuracy. Independent samples t-tests were performed to compare the two groups in terms of power spectral density (PSD) and functional connectivity (wPLI). For EEG features that showed significant inter-group differences after multiple comparison correction, we further employed Pearson correlation analysis to examine their relationship with objective originality scores. Additionally, Pearson correlation analysis was also used to compute the relationships between variables such as addiction severity and abstinence duration and the originality scores. All data were tested for normality (Shapiro–Wilk test) and homogeneity of variance (Levene’s test), and the assumptions of the statistical tests were satisfied. All statistical analyses were performed using SPSS version 27.0, with the significance threshold set at p < 0.05. Machine learning classification To classify individuals with high and low divergent thinking based on neural features, we utilized a support vector machine (SVM) model with a radial basis function (RBF) kernel [66]. The focus was placed on classifying divergent thinking levels rather than directly distinguishing individuals with HUD from HC in order to identify fine-grained neurophysiological markers associated with this higher-order cognitive ability. Participants were divided into high and low divergent thinking groups based on the median of their originality scores from the divergent thinking task [67–69]. The SVM was optimized using a box constraint set to 1 and an automatically determined kernel scale. Posterior probability estimates were calculated using Platt scaling. Predictor variables included functional connectivity (wPLI), with feature selection restricted to those showing significant group differences between the HUD and HC groups. All features were standardized using Z-score normalization, and those with a variance below 0.01 were removed. A 10-fold cross-validation strategy was applied, with 90% of the data used for training and 10% for testing in each fold. The evaluation metrics for classification performance included accuracy, precision, recall, F1 score, and the area under the receiver operating characteristic curve (AUC) [70]. Results AUT scores Two independent graduate students in psychology evaluated the originality of each response using a 5-point scale (1–5). The reliability of the inter-rater assessment was confirmed, with the Intraclass Correlation Coefficient (ICC) indicating a strong level of agreement between the two raters (ICC = 0.890). A repeated-measures ANOVA was conducted to examine the differences in rating types (objective vs. subjective) as a within-subject variable, with group (HUD vs. HC) as a between-subject variable. When the assumption of sphericity was violated, a multivariate approach was used for the repeated-measures variable. Bonferroni correction was applied to control for multiple comparisons. Effect sizes (partial η²) were reported to assess the magnitude of differences. The results revealed a significant main effect of rating type (F(1, 71) = 10.651, p = 0.002, partial η² = 0.129), a significant main effect of group (F(1, 71) = 41.603, p < 0.001, partial η² = 0.369), and a significant interaction between group and rating type (F(1, 71) = 12.782, p < 0.001, partial η² = 0.153). Simple effects analysis revealed that: within the HUD group, the subjective ratings were significantly higher than the objective ratings (p < 0.001). Within the HC group, there was no significant difference between the subjective ratings and the objective ratings (p = 0.082). Across both rating conditions (objective and subjective), the ratings in the HUD group were significantly lower than those in the HC group (ps < 0.001) (Fig. 2). The sensitivity (d’) and response bias (β) of creativity evaluation were calculated using response accuracy scores. For the HUD group, the sensitivity (d’) was 0.62, and the response bias (β) was 0.45. In comparison, the HC group demonstrated significantly higher sensitivity (d’ = 2.79) and a stricter response bias (β = 1.48). Comparison of behavioral data in the AUT task between the HUD group and the HC group. *p < 0.05, **p < 0.01, ***p < 0.001. Spectral analysis In the source space analysis, we compared the PSD differences between the two groups in specific brain regions. In the alpha frequency band, the results showed that the HUD group exhibited significantly higher power in the left precuneus (L.PCUN) and the left superior parietal lobule (L.SPL) compared to the HC group (L.PCUN: t(71) = 3.651, p = 0.015, d = 0.855; L.SPL: t(71) = 3.741, p = 0.011, d = 0.876). In the beta frequency band, the power in the right superior parietal lobule (R.SPL) was also significantly higher in the HUD group than in the HC group (t(71) = 3.441, p = 0.020, d = 0.806), all p-values were adjusted using the FDR correction (Fig. 3). Differences in task-related spectral power between the HUD group and HC group. A Brain regions showing significant group differences in spectral analysis were selected as seed regions for subsequent functional connectivity analysis. B Spectral features significantly different between the two groups after FDR correction. C Spectral power curves for the two groups. *p < 0.05, **p < 0.01, ***p < 0.001. Functional connectivity In the alpha frequency band, the HC group exhibited significantly higher functional connectivity than the HUD group. Specifically, the wPLI between the left superior parietal (L.SPL) and the left rostral anterior cingulate (L.rACC) was significantly greater in the HC group (t(71) = − 3.544, p = 0.041, d = −0.830). Similarly, the connectivity between the left superior frontal (L.SFG) and the left precuneus (L.PCUN) (t(71) = − 4.417, p = 0.002, d = −1.035), as well as between the right superior frontal (R.SFG) and the left precuneus (L.PCUN) (t(71) = − 3.544, p = 0.025, d = −0.868), was significantly higher in the HC group. In the beta frequency band, the HC group also demonstrated elevated connectivity, with significantly higher wPLI values observed between the right superior parietal region (R.SPL) and the left cuneus (L.CUN) (t(71) = − 3.926, p = 0.012, d = −0.92); between the right superior parietal (R.SPL) and the left precuneus (L.PCUN) (t(71) = − 3.572, p = 0.038, d = −0.837); between the left superior parietal (L.SPL) and the right superior parietal (R. SPL) (t(71) = − 3.680, p = 0.027, d = −0.862); and between the left supramarginal gyrus (L.SMG) and the right superior parietal (R.SPL) (t(71) = − 3.513, p = 0.046, d = −0.823). All p-values were adjusted using the Bonferroni correction (Fig. 4). Differences in task-related functional connectivity (wPLI) between the HUD group and HC group. A Heatmaps illustrating the magnitude of group differences in the alpha and beta frequency bands. B Brain maps of significant functional connectivity in the alpha and beta frequency bands after Bonferroni correction, with the thickness of the connection lines indicating the magnitude of the difference. C Chord diagrams of significant functional connectivity in the alpha and beta frequency bands after Bonferroni correction. D Bar charts of significant functional connectivity pairs in the alpha and beta frequency bands after Bonferroni correction. *p < 0.05, **p < 0.01, ***p < 0.001. Correlation analysis The correlation analysis revealed significant associations between neural activity in multiple brain regions and originality scores across both frequency bands. In the alpha band, L.PCUN and L.SPL power were negatively correlated with originality scores (r = −0.324, p = 0.005; r = −0.335, p = 0.004). In the beta band, R.SPL power was negatively correlated with originality scores (r = −0.335, p = 0.004). Functional connectivity analysis further indicated that, in the alpha band, connectivity between L.SPL and L.rACC was positively correlated with originality scores (r = 0.260, p = 0.026), as was the connectivity between L.SFG and L.PCUN (r = 0.423, p < 0.001), and between R.SFG and L.PCUN (r = 0.353, p = 0.002). In the beta band, positive correlations with originality scores were observed in connectivity between R.SPL and L.CUN (r = 0.264, p = 0.024), between R.SPL and L.PCUN (r = 0.244, p = 0.037), and between L.SPL and R.SPL (r = 0.244, p = 0.037). Additionally, connectivity between L.SMG and R.SPL demonstrated a significant positive correlation with originality scores (r = 0.264, p = 0.024) (Fig. 5). Moreover, abstinence duration (in days) was positively correlated with originality scores (r = 0.364, p = 0.025). All p-values were adjusted using the FDR correction. Correlation matrix of spectral and functional connectivity features with originality scores. A Correlation between spectral features and originality scores in the two groups. B Correlation between functional connectivity features in the alpha frequency band and originality scores in the two groups. C Correlation between functional connectivity features in the beta frequency band and originality scores in the two groups. Machine learning An SVM classifier was utilized to evaluate the discriminative power of EEG features, specifically functional connectivity indices. Classification performance was assessed using several metrics, including accuracy, precision, recall, F1 score, and the area under the receiver operating characteristic curve (AUC). Seven relevant functional connectivity features were analyzed, and the results are presented in supplementary materials Table S2. Notably, the functional connectivity between R.SFG and L.PCUN demonstrated the highest classification performance (AUC = 0.772, accuracy = 0.767, precision = 0.835, recall = 0.625, F1 score = 0.695) (Fig. 6). Classifying participants’ divergent thinking ability using functional connectivity features with an SVM model. A Receiver operating characteristic (ROC) curve illustrating classification performance based on functional connectivity features. B Classification accuracy based on functional connectivity features. C Evaluation metrics for the functional connectivity feature between R.SFG and L.PCUN. Discussion This study utilized EEG to investigate divergent thinking and its underlying dynamic neural processing mechanisms in individuals with chronic heroin use disorder. Using EEG source imaging, we analyzed the complex interactions among brain regions. By integrating machine learning methods, the study identified neurobiological markers that could serve as indicators of cognitive abnormalities. The findings revealed abnormal neural oscillation patterns and functional network connectivity in HUD. These neural abnormalities may serve as potential circuit-based markers, providing a scientific foundation for developing more targeted, mechanism-based intervention strategies. Originality, as a core metric for evaluating the ability to “break conventional associations and generate novel concepts” [71–73], was found to be significantly impaired in the HUD group, as reflected by their lower objective originality scores. This finding was consistent with previous research on addiction-related cognitive dysfunctions [20, 74, 75] and further supported the notion that heroin addiction detrimentally affects higher-order thinking abilities. Specifically, heroin addiction weakens cognitive control and flexibility [76, 77], resulting in reduced ability to inhibit automated responses and shift conceptual categories. This deficiency obstructs the retrieval of novel information and limits effective engagement in divergent thinking processes. The self-evaluation of the HUD group revealed that their originality scores were significantly higher than the objective evaluation, indicating that the HUD group had difficulty accurately evaluating the actual effectiveness of their divergent thinking processes, thereby overestimating their subjective perception of creativity generation. A similar subjective exaggeration effect has been observed in other substance-abusing populations [78], suggesting that this may represent a common characteristic of impaired cognitive evaluation functions associated with drug addiction. Furthermore, originality scores in the HUD group were positively correlated with duration of abstinence, indicating that creative cognition of individuals with HUD has the potential to recover with sustained abstinence. In source-space analysis, this study identified significantly elevated power in the alpha and beta bands in the HUD group, revealing region-specific neural oscillatory abnormalities linked to impaired divergent thinking. In the alpha band, increased power in the left precuneus (L.PCUN) and left superior parietal lobule (L.SPL) suggested dysfunction in key networks. The L.PCUN, a key hub of the DMN, is involved in self-referential processing, contextual memory retrieval, and semantic associations [79, 80]. Enhanced alpha oscillations in this region often reflect cortical suppression of irrelevant neural representations [81]. In this study, the elevated alpha activity in the L.PCUN may indicate excessive inhibition of DMN-driven spontaneous activity during cognitive tasks, disrupting the integration and retrieval of internal information critical for divergent thinking. The L.SPL, part of the FPN, is primarily responsible for spatial attention allocation and working memory load regulation [82]. Increased alpha power in this region may signify compensatory cognitive resource allocation in the HUD group, reflecting efforts to maintain basic cognitive functioning despite reduced overall efficiency. Beta oscillations are typically associated with cognitive control and goal maintenance [83]. Elevated beta power in the R.SPL may reflect heightened vigilance or hyper-responsiveness to external stimuli during cognitive tasks, exacerbating imbalances in cognitive resource distribution. Collectively, these abnormalities highlight disrupted interactions between internally generative and externally regulated systems, impairing information processing crucial for divergent thinking. Functional connectivity analysis revealed significant reductions in connectivity in the HUD group compared to the HC group. In the alpha band, functional connectivity involving the left precuneus (L.PCUN) and left superior parietal lobule (L.SPL) was markedly diminished, while in the beta band, connectivity involving the right superior parietal lobule (R.SPL) also showed significant reductions. Machine learning results indicated that reduced alpha-band connectivity between the right superior frontal gyrus (R.SFG) and L.PCUN was strongly associated with impaired divergent thinking and could serve as a robust biomarker for classifying divergent thinking abilities. The R.SFG, a critical node in the FPN, is responsible for dynamically allocating goal-directed cognitive resources [84, 85]. Abnormal connectivity between the DMN and FPN has been consistently observed in individuals with substance use disorders [86–89], and the strength of task-based DMN-FPN connectivity has been shown to predict cognitive flexibility [90]. Divergent thinking inherently depends on phase-specific coupling between the DMN and FPN: the DMN generates potential ideas through broad semantic associations, while the FPN evaluates and optimizes candidate solutions [91, 92]. The findings suggest that reduced R.SFG_L.PCUN connectivity in the HUD group may disrupt the dual-pathway to creativity model (DPCM), causing cascading impairments. Specifically, impaired DMN-driven spontaneous associations lead to reduced conceptual expansion (manifested as fewer alternative solutions), while weakened FPN-mediated resource regulation exacerbates deficits in executive monitoring (evidenced by reduced efficiency in suppressing irrelevant information). This dual impairment may lead to a vicious cycle in the context of social adaptation: deficits in cognitive flexibility hinder the HUD group from breaking rigid patterns of drug craving, while impairments in cognitive persistence may restrict their ability to sustain goals, manifesting as attentional lapses and premature abandonment during withdrawal. Notably, this study observed a general elevation in PSD across multiple key brain regions in the HUD group, contrasting with previous findings of decreased alpha activity [35]. However, it aligned with previous reports of reduced functional connectivity in behavioral addiction [93]. Similar network abnormalities have also been observed in other psychiatric disorders, such as schizophrenia, where regional activation was heightened but functional connectivity was diminished during associative tasks [94]. This discrepancy may indicate a compensatory neural mechanism in response to network dysfunction. Specifically, to offset the inefficiency caused by impaired coordination between the DMN and the FPN, the brain appears to amplify local neural oscillations to sustain cognitive operations. However, this inefficient compensation may exacerbate the burden on cross-network integration, which in turn weakens connectivity and prompts additional regions to engage in compensatory efforts, and ultimately reducing the overall resilience of the cognitive system. This cascade further impairs divergent thinking and perpetuates a maladaptive cycle in social functioning. Limitations This study has several limitations. First, the cross-sectional design restricts causal inference. Future longitudinal or prospective studies (tracking high-risk populations) are needed to clarify the causal relationship between neural oscillatory abnormalities and divergent thinking deficits. Second, the study relied solely on EEG as the neurophysiological measure and did not directly assess the impact of dopamine system function on divergent thinking. Future research could combine multimodal imaging techniques (e.g., dopamine transporter PET and resting-state fMRI) to construct a more comprehensive neurobiological marker network. Finally, although this study used machine learning to preliminarily identify potential intervention targets, the plasticity of these targets has yet to be verified. Future research should incorporate neurostimulation techniques to evaluate intervention effects, providing empirical evidence for personalized treatment strategies [95]. Conclusions This study provides a multi-level elucidation of the neural mechanisms underlying divergent thinking impairments in heroin addiction. At the neural oscillation level, findings indicated abnormal enhancements in alpha oscillations in the left precuneus (L.PCUN) and superior parietal lobule (L.SPL), as well as increased beta oscillations in the right superior parietal lobule (R.SPL), reflecting dysregulation between internally generative and externally controlled systems. At the functional connectivity level, significant disruptions in the connectivity between the DMN and the FPN were observed, undermining the dynamic synergy of the “generation-evaluation” loop central to the dual-pathway model of creativity. At the whole-brain network level, the combination of excessive local oscillatory activity and weakened functional connectivity suggests a unique compensatory pathological mechanism. This study shifts the discussion from vague associations to a scientifically grounded understanding of how drug addiction damages divergent thinking, thereby providing a more accurate and evidence-based foundation for public health initiatives and addiction prevention efforts. Importantly, by identifying reliable EEG biomarkers, this research lays the theoretical groundwork for developing precise intervention strategies targeting cognitive dysfunction in individuals with addiction, offering substantial potential for clinical translation. Specifically, interventions targeting alpha-band functional connectivity could be explored as a future direction for neural modulation to further improve higher-order cognitive functions in individuals with substance use disorders. Supplementary information Supplementary Materials Acknowledgements This study was supported by the National Natural Science Foundation of China (32471114, 82371500, U22A20303); the Special Support Program for Innovation Teams Led by High-level SanQin Talents of Shaanxi Province; the Fundamental Research Funds for the Central Universities (GK202502005); the Research Program Fund for the Youth Innovation Team of Shaanxi Universities (2022SXSJZW105); the Chinese National Programmes for Brain Science and Brain-like Intelligence Technology (No.2022ZD0214500); and BRKOT-XDU (20231008Y), Xidian University Specially Funded Project for Interdisciplinary Exploration. Author contributions WF conceived and designed the study, implemented the experiment, collected the data, performed statistical analyses, generated visualizations, and drafted, revised, and refined the manuscript. YW led participant recruitment, contributed to the study design, and revised and refined the manuscript. WL contributed to data collection and revised the manuscript. YM revised and refined the manuscript. JD assisted with data acquisition. HD and KY provided senior supervision, formulated the overarching research framework, coordinated recruitment strategies, and critically revised the manuscript for academic content. Data availability The datasets generated and analyzed during the current study are available from the corresponding authors on reasonable request. Code availability The code used to perform the analysis in the current study is available from the corresponding authors on reasonable request. Competing interests The authors declare no competing interests. Ethics approval The experiment was conducted in strict accordance with national laws and regulations. The experimental methods used were non-invasive to the participants. The entire process was supervised by and approved through the Ethics Committee of the Key Laboratory of Modern Teaching Technology, Ministry of Education, Shaanxi Normal University (Ethics Approval No.: L20240905-01). Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Wenjuan Fu, Yifan Wang. Contributor Information Kai Yuan, Email: kyuan@xidian.edu.cn. Haijun Duan, Email: duanhj@126.com. Supplementary information The online version contains supplementary material available at 10.1038/s41398-025-03783-9. References Associated Data Supplementary Materials Supplementary Materials Data Availability Statement The datasets generated and analyzed during the current study are available from the corresponding authors on reasonable request. The code used to perform the analysis in the current study is available from the corresponding authors on reasonable request.