Substance use disorders (SUDs) are prominent social, legal, and public health challenges. Illicit opioids are among the most-used drugs that cause addiction and are responsible for multiple deaths annually in the United States (Browne et al., 2020). In China, at least 730,000 individuals were exposed to such drugs, typically heroin, in 2020 (Wang et al., 2022). Heroin is an opioid with strong addictive characteristics (Nutt et al., 2007). Currently, heroin addiction (HA) remains a complex societal problem that disrupts health internationally, since heroin users readily develop a high tolerance (Zetland, 2003) that frequently leads to heroin overdose (Latkin et al., 2004).

Transcranial magnetic stimulation (TMS) is a non-invasive brain stimulation tool that has shown promising results in treating different neurological disorders. For instance, TMS improved mood, behavior, and cognition in patients with depression (O'Reardon et al., 2007) and alleviated negative symptoms of schizophrenia (Lorentzen et al., 2022). This stimulation technique was incrementally used as an efficacious treatment for SUDs to potentially reduce craving for different types of drugs, such as cocaine, nicotine, methamphetamine, and heroin, and addiction-related behaviors (Li et al., 2013a; Liu et al., 2017; Ma et al., 2019; Rapinesi et al., 2016; Shen et al., 2016; Terraneo et al., 2016). Moreover, repetitive TMS (rTMS) that induces changes in cortical neuroplasticity can be effective in ameliorating clinical symptoms of drug-addictive behaviors (Song et al., 2021; Tsai et al., 2021; Yuan et al., 2020), especially craving (Kang et al., 2022; Liu et al., 2020; Shen et al., 2016). However, the aforementioned TMS intervention studies for SUDs used craving as a subjective reporting indicator to examine treatment effects, which may impact sensitivity and objectivity.

Electroencephalography (EEG) has been widely used to investigate brain dysfunction in different samples. The temporal dynamics of EEG signals (Apkarian et al., 2005; Pfurtscheller and Da Silva, 1999; Wu et al., 2006) and the location of neural sources (Zhao et al., 2017) can be quantified by employing several EEG analytical methods. Among these, EEG microstate analysis based on topographic clustering of scalp electric fields is more objective and stable. Further, it is simple to perform, and optimizes the spatial information contained in the EEG signal. This analysis enables high temporal resolution to map large-scale cortical activity, spatial organization, and temporal dynamics (Michel and Koenig, 2018). Moreover, the malfunction of the brain network is reflected as changes in EEG microstates (Lin et al., 2022). Thus, EEG microstates analysis is more keen and applicable to special groups with brain abnormalities (Nishida et al., 2013).

The EEG microstate could represent the spatial and temporal characteristics of resting states (Michel and Koenig, 2018), and reflect that the morphological changes of the electric field in the brain were nonlinear and discontinuous (Li et al., 2022). Firstly, the scalp voltage topography of resting-state EEG signals does not change randomly or continuously. Secondly, broad-band spontaneous EEG activity is identified as periods of quasi-stable scalp EEG topography that remain stable for 60–120 ms before rapidly evolving into a new map configuration (Gao et al., 2017). Four typical microstate classes (A, B, C, and D) have been identified using a modified pattern classification algorithm (Michel and Koenig, 2018). These classes could explain 80% of the variance in EEG data and were consistently identified in different studies with significant cross-subject similarity (Andreou et al., 2014; Van de Ville et al., 2010), regardless of participants’ age (Koenig et al., 2002) and sample size (Michel and Koenig, 2018). Moreover, these classes are linked to some large-scale resting-state networks identified using functional magnetic resonance imaging (Britz et al., 2010; Musso et al., 2010; Yuan et al., 2012). Microstate class A correlates with auditory networks associated with speech processing, B with the imagery network responsible for visual processing, C with salience networks responsible for subjective sensory-autonomic processing, and D with attentional networks (Britz et al., 2010). The most widely used microstate time parameters include the following: (1) average duration; (2) frequency of the occurrence; (3) coverage; (4) global explained variance; and (5) transition probability (Khanna et al., 2015; Michel et al., 2009).

EEG microstates could be used to classify patients with psychiatric disorders, including major depressive disorders (Damborská et al., 2019; Lei et al., 2022; Murphy et al., 2020), bipolar disorders (Gschwind et al., 2016), schizophrenia (Kochi et al., 1996; Lin et al., 2022; Murphy et al., 2020), and panic disorders (Wiedemann et al., 1998). Critically, the time series of the microstate of patients with SUDs (e.g., methamphetamine and nicotine), especially those with more frequently used lengthy periods, were disrupted as compared to those of healthy controls (HCs). The mean durations of microstate classes A and B were shorter in patients with methamphetamine misuse (Chen et al., 2020), and the duration of microstate class B negatively correlated with the length of nicotine use (Cheng et al., 2020). Notably, the indicators of the microstate class D (i.e., duration and coverage) and Fagerstrom test of nicotine dependence findings were significantly negatively correlated; additionally, microstate class D could predict the degree of cigarette dependence in adolescent smokers (Li et al., 2022). Moreover, alcohol increased the coverage of the microstate class B but reduced the same for the microstate class C. Further exploratory analyses showed that alcohol also increased the transition from microstate class C to microstate class B, and reduced the bidirectional transition between microstate class C and microstate class D (Schiller et al., 2021). The impairments in brain mechanisms were different for each substance (Verdejo-García and Pérez-García, 2007). Notably, heroin caused the worst detrimental effects on health among the various illicit drugs (Nutt et al., 2007). However, the evidence of microstate classes in patients with HA remains scarce.

Neuroimaging studies demonstrated that HA could lead to abnormal neural activity and functional connectivity (Leshner, 1997). Numerous HA studies investigated structural defects (Li et al., 2013c; Lin et al., 2012; Wang et al., 2012; Yuan et al., 2010a), altered functional connectivity (Ma et al., 2011; Yuan et al., 2010a), and changes in topological properties of brain networks (Liu et al., 2009; Pandria et al., 2018; Yuan et al., 2010a). Reduced prefrontal cortex (PFC) activity is the major cause of decline in executive system and cognitive control in individuals with SUDs (Goldstein and Volkow, 2011). Additionally, loss of PFC gray matter volume has been found in patients with HA (Liu et al., 2009; Qiu et al., 2013). Furthermore, PFC is critical for advanced cognitive functions. The dorsolateral PFC (DLPFC), which is a key brain region for addiction, retains essential roles in working memory and decision-making. Also, the DLPFC is strongly coupled with other brain regions, such as the striatum and cingulate gyrus (Jin et al., 2022). Therefore, many rTMS protocols select DLPFC of the brain, typically the left side, as a stimulation target (Abdelrahman et al., 2021; Liang et al., 2018; Shen et al., 2016; Yuan et al., 2020).

Theta-burst stimulation (TBS) is more widely used as compared to other rTMS treatment because it is more effective with a shorter stimulation time (Huang et al., 2005). TBS is employed to mimic the natural firing patterns of the brain continuously or intermittently, which up- or downregulates the excitability of focal areas on the cortical surface with high precision (Blumberger et al., 2018; Diamond et al., 1988). Notably, intermittent TBS (iTBS) could increase cortical excitability for 60 minutes (Wischnewski and Schutter, 2015), inducing long-term potentiation in neural circuits (Huang et al., 2005). Additionally, rTMS could induce dynamic neuroplasticity in EEG microstate classes. For instance, the mean durations of the four EEG microstate classes were significantly higher after low-frequency rTMS stimulation (Qiu et al., 2020). Moreover, EEG microstate parameters correlated with improvements in symptoms among schizophrenia patients (Sverak et al., 2018). Furthermore, a randomized trial revealed that the duration of microstate class D in the active group was significantly higher than that in the sham group (Pan et al., 2021). Therefore, iTBS may be a more robust and feasible treatment to remediate and boost the microstate classes in patients with HA as compared to common stimulus patterns.

EEG-based HA classification has wide-reaching clinical significance in elucidating the differences in neural mechanisms and outcomes of SUDs in the human brain (Pandria et al., 2018), and thus, is critical for more objective and accurate addiction diagnosis and recovery assessment. However, only one study has shown that a support vector machine (SVM) accurately classifies HA and HC (Wang et al., 2022). The main function of the K-means clustering algorithm is to automatically group similar samples into categories. This algorithm is a partition-based method with wide application because of its simplicity and efficiency (Aggarwal, 2004). Thus, this study aimed to examine the effectiveness of TMS intervention in patients with HA using EEG microstate analysis and verify whether this analysis is appropriate with the K-means clustering algorithm.

To further understand the changes in brain states induced by iTBS over the left DLPFC, measuring and analyzing the dynamic changes in the microstate of patients with HA post-intervention are necessary. We first directly compared the spatial topography and temporal dynamics of classically classified EEG microstate classes and EEG electrode signals (Michel and Koenig, 2018) and further classified healthy individuals and patients with HA who did or did not undergo iTBS using a K-means clustering algorithm. The results provide new perspectives for assessing the treatment effects of TMS among patients with HA.