Various questionnaires have been developed for use before a person becomes addicted to drugs and during abstinence to predict the likelihood of relapse. The Inventory of Drug-Taking Situations is a questionnaire to judge the risk of drug addiction based on everyday situations. Abuse of drugs can be screened for using the assessment tools suggested by the National Institute on Drug Abuse (49). Among the suggested questionnaires, the most popular choice to examine an individual’s involvement with a variety of drugs is the Drug Use Questionnaire or Drug Abuse Screening Test (50, 51). To evaluate drug addiction over time, the Desire for Drug Questionnaire can be used to rate instances of cravings for drugs, whereas the Obsessive Compulsive Drug Use Scale evaluates cravings over a period of time (52). Regarding the effects of the drugs, the Visual Analog Scale (VAS), a visualization scale, is helpful for quantifying levels of craving for drugs. In addition, the Addiction Severity Index is used to evaluate overall issues, from personal to family and society, in the context of drug addiction (53).

As drug addiction affects has both physical and psychosocial effects, scales that characterize anxiety, depression, or impulsivity resulting from drug abuse are often used to obtain a broad view dependence and substance abuse. These scales include the 21-item Beck Anxiety Inventory, 21-item Beck Depression Inventory, and 30-item Barratt Impulsiveness Scale-11 (54). For the assessment of drug abstinence, the Drug Abstinence Self-Efficacy Scale is available (55). During the withdrawal period from drug use, the Subjective Opioid Withdrawal Scale, for example, can be used to check the symptoms of withdrawal (56), and the Methamphetamine Withdrawal Questionnaire was developed to specifically evaluate METH withdrawal-related symptoms (57). The Risk of Relapse Assessment Scale can be used to determine the possibility of relapse (26), while another option is the Stimulant Relapse Risk Scale (58). Furthermore, the Time to Relapse Questionnaire has been proposed to distinguish two types of relapses, with or without forewarning, to enable better treatment (59, 60).

Although questionnaires are the most convenient and broadly accessible approach for a variety of drug addiction-related applications, the results of the scales have limited reliability and accuracy. As the results of these scales are based on the answers of the respondents, the results exhibit individual variations and are influenced by the attitudes and conditions of the participants (61). Moreover, clinicians are often needed to draw conclusions about the severity of addiction based on the results of the scales and consultations. These limitations restrict the scope of application. Therefore, questionnaires often need to be combined with physiological tests to strengthen their conclusions about addiction, withdrawal, abstinence, or relapse.

In addition to questionnaires, saliva, urine, and blood tests are conventional ways to quantitatively detect drug use (22). Nail and hair also contain evidence of drug use with a detection window lasting from weeks to months (62, 63). Moreover, breath analyzers can be used to detect the drug in the breath (64). However, the time windows of the above conventional methods are limited. Moreover, they cannot provide real-time results owing to the time-consuming nature of the required examination and analysis procedures. Emerging wearables are promising options to detect biomarkers in real-time (27, 65), most commonly using sweat to screen for drugs using electrochemical techniques (66). Other body fluids containing drugs are saliva and tears. However, it is sometimes difficult to obtain sufficient body fluids for accurate sensing. In addition to electrochemical sensing of body fluids, wearables can detect other physiological parameters to predict or determine addiction to drugs: these include electrocardiographic (ECG) parameters, heart rate variability (HRV), breath rate, and skin conductance response/galvanic skin response (SCR/GSR). Both heart rate (HR), determined by the R-R interval of ECG signals recorded from a chest band and breath rate increase with increasing dosages of cocaine (67). Other parameters available in ECG waveforms for drug addiction evaluation are the QT, PR, QRS, and QTc intervals and the height of T waves, as summarized in previous work (67). Regarding the morphology of the breathing waveform, the time and depth of inhalation and exhalation, and respiration duration are features used to predict drug-seeking and craving for drugs. Furthermore, a wristband is another option to identify the use of cocaine based on recorded skin temperature, heartbeat, motion, and SCR (28, 68). HRV can be used to evaluate stimulation in individuals with METH addiction via virtual reality (VR) (69) and SCR increases in patients with METH addiction when receiving specific cues (70); importantly, longer use of METH results in a stronger physiological reaction to the cues. Owing to the ease of access and user friendliness of wristband devices for ECG recording, the effects of aerobic exercise on the HRV parameters of patients with METH addiction have been investigated (71). HRV can be further separated into high-frequency HRV and low-frequency HRV. In addition to the frequency domain, parameters can be derived from the recorded time domain of HR, including the standard deviation of normal-to-normal intervals, root mean square difference of the standard deviation, and percentage of beats that change by more than 50 ms compared with the previous beat.

In addition to ECG and HR, drug abuse has effects on pupil size (72). The features of ECG, GSP, and eye tracking that can indicate METH addiction are summarized in Tsai et al. (73). Wearables can be used not only to monitor physiological signals but also to track the psychological impact of drug addiction. For example, the information obtained from accelerometry and GPS location is useful in characterizing the cravings resulting from the addiction (74, 75). Owing to the large variety of the features available to diagnose drug addiction or the resulting psychological changes, machine learning (ML) has been introduced to increase the precision of the analyzed results; more information will be provided in section “6.1.2. Analysis of recorded neural signals.”

The wearable techniques presented in this section are used to measure the physiological reactions of the autonomic nervous system caused by drug usage. However, drug addiction also affects the central nervous system (7). We believe that the development of a closed-loop system for overall management of drug addiction would provide substantial value, regardless of whether patients are still using drugs, in the abstinence stage, or aim to reduce the rate of relapse rate. Application of this closed-loop system would rely on neuroimaging techniques, being the most straightforward approaches for brain studies, to identify and characterize the brain signatures of drug addictions.

Whereas the wearable technologies discussed in the previous section can be used to monitor physiological changes in the body as relates to drug use and withdrawal, clinicians and researchers are also keen to learn how drug use influences the control center of the body, the brain. The process of developing an addiction includes several phases, including drug intoxication, craving, binging, and withdrawal with loss of self-control (76–78). A series of complex changes occur, including modification of brain structures as well as mental and physiological changes, resulting in various symptoms such as depression, impulsiveness, anxiety, aggression, and many other psychological problems (11). Consequently, a tool to monitor the response/effect of a neuromodulation treatment is a need for a closed-loop system. Neuroimaging techniques provide opportunities to monitor the overall process of the functional modifications of the central nerve system in various conditions (79). Therefore, they are suitable to study the mechanisms of neurological disorders (76, 80).

Neuroimaging approaches can be separated into two categories: those that can be used to examine the structural changes at different stages of drug dependence and addiction to understand the physiological mechanisms of addiction (30), and those that monitor the functional brain changes that occur as a consequence of addiction (77). MRI, MRS, and single photon emission computed tomography are used to inspect structures in patients with METH addiction (31, 32, 81). PET is used to study the impact of drug addiction on the brain at a molecular level, in order to find the optimal brain location for treatments (82). fMRI is a powerful tool to evaluate hemodynamic conditions after the neurons have been activated by specific tasks (83, 84). The studies investigating the relationship between METH use and cognitive function using neuroimaging techniques have been summarized elsewhere (32).

Although these neuroimaging techniques provide high spatial resolution, they have limitations. For instance, the devices required are costly and bulky, with low accessibility, and trained operators are needed to conduct the examinations. Moreover, MRI may not be suitable for patients with metallic implants, and the use of radioactive agents in PET limits the frequency of examinations. Furthermore, those with claustrophobia may find it difficult to participate the examinations owing to the spatially confined test environment that is needed. Low temporal resolution also restricts the application of real-time monitoring.

Wearable devices, being compact and easy to access, are appropriate for real-time monitoring of neural activity. Owing to their user-friendly implementation, wearable neuroimaging devices have been widely used to study the outcomes of various treatments, including the effects of exercise on parameters of drug addiction (85). Another example is the evaluation of cravings to predict the risk of relapse during abstinence. Real-time signals recorded in a natural environment provide more reliable information than could be obtained under stressful conditions using bulky equipment.

Electroencephalogram can be used to record brain cortical electrical activity via electrodes attached to the individual’s head. It is popular in research and clinical studies owing to the high temporal resolution of afforded by this approach. This benefit enables EEG to be used to record variations in neural activity when patients receive drug-related stimuli or experience the desire for drug taking.

When neurons change their activity patterns, such as during different phases of addiction, the local hemodynamic conditions in the brain change, resulting in neurovascular coupling (86). Whereas EEG can monitor electrical signals in the brain, fNIRS is a popular wearable device for monitoring the hemodynamic conditions of the brain, enabling these two to be used in combination. For example, the prefrontal, dorsolateral prefrontal, and orbitofrontal cortices are responsible for decision-making (30). When patients with drug addiction use the drug or receive cues related to it, this can change the activity patterns of these brain regions. This results in alternations in oxygenated and deoxygenated hemoglobin concentrations, which can be recorded using fNIRS at the corresponding cortices (87). Studies have shown that changes in these concentrations in patients with drug addictions are different from those in healthy controls (88). In addition, with its advantages of being light, compact, wearable, highly accessible, and user-friendly, fNIRS is becoming a popular tool to study the effects of exercise on drug addictions. For instance, fNIRS has been used to evaluate the effects of spin training and strength training on those with a METH addiction (89). Another study used fNIRS to evaluate the effects of dancing and exercise on aspects of meth addiction (90). As well as its applications in analysis of the hemodynamic variations influenced by exercise, fNIRS has been used to investigate the relative hemodynamic changes at different cortical regions in the brain (91). For example, a classification algorithm based on the fNIRS signals at various brain cortices was used to distinguish addictions to different drugs.

Neurovascular coupling indicates that electrical neuron signals are closely related to hemodynamic conditions. In neurovascular coupling, when neuronal activity is elevated, more oxygen is delivered to the activated brain regions, resulting in a local increase in oxygenated hemoglobin. As EEG measures neural electrical activity whereas fNIRS monitors hemodynamic activity, multimodal EEG-fNIRS recording provides a more holistic measurement, enabling a more comprehensive understanding of the effects of drug use and withdrawal on the brain (86). This dual approach has recently been used in several studies, for example, multimodal EEG-fNIRS has been used to study the brain activity of those with a METH addiction under visual stimulation (92, 93), whereas another study used multimodal EEG-fNIRS for those with an opioid addiction (94).

A further benefit of wearable systems is that they can be applied without the limitations of time and location, with a minimal influence on social activity. Consequently, multimodal EEG-fNIRS systems are suitable for evaluation of the efficacy of treatment and rehabilitation. An important goal during treatment and rehabilitation is the reduction of the incidence of relapse. Craving is a key symptom that promotes relapse to drug use; thus, identifying brain activity patterns during, immediately before, and immediately following cravings could help to optimize treatment and rehabilitation programs. For example, if we could identify the neuronal signatures of craving, we could trigger a closed-loop stimulation protocol to combat these activity patterns, or provide alternative interventions. In addition, brain signal monitoring can provide more information on changes in psychological and physiological conditions than the wearables measuring ECG, HR, and GSR mentioned in section “3.2. Conventional detection techniques and emerging wearable techniques.” Notably, including as many modalities as possible in the wearables would increase their diagnostic precision.

It has been suggested that wearables could be used to track the efficacy of treatment of those with a drug addiction (65). However, the propensity for interventions based on this approach alone is limited. To more effectively reduce rates of drug use and relapse, a closed-loop system to provide real-time treatment that is customizable to each specific case is needed (95). Such a closed-loop system would consist of three parts. The first part is neuroimaging, which can be achieved using multimodal EEG-fNIRS devices. The second involves computational algorithms to identify the neuronal and/or hemodynamic activity biomarkers of addiction. Finally, the loop is closed by neuromodulation approaches that provide stimulation of targeted brain areas to combat pathological activity changes related to addiction.

Approaches to investigating brain signal changes influenced by drug addiction include comparing recorded signals from those suffering from addictions with control populations, or comparing the signals of those suffering from an addiction across the addiction cycle. One approach to minimize the acute and long-lasting effects of drug intake on the participants during attempts to identify brain signal biomarkers, drug-related cues can be applied as alternatives to drugs. This approach is used because the variations in autonomic nervous system and brain signals occur not only after the presentation of drug use but also after cues. To identify useful biomarkers to distinguish addiction and control groups, protocols including various stimulations to provoke cue reactivity have been proposed (96). Cues can be videos, photos, audios, or VR of the drugs or individuals using the drugs. This approach has been done before using MRI and EEG. In these studies, it was found that VR induced more cravings than other stimulation types (97, 98). Natural cues are included as important controls to identify drug-specific cue responses (99). Natural cues can be categorized into those with limited connections to drugs and those that share some features with drug-related cues. The former types may include natural scenes such as trees and flowers, while the latter types could be, for example, a person holding a screwdriver close to the face. This may trigger thoughts of METH use because a screwdriver may appear similar to the use of a long tube typically used for METH consumption. Therefore, placing a different tool of a similar shape close to the face can induce desire for the drug. Another example of the latter type is a light bulb, which may look similar to a METH pipe. It has been demonstrated that these natural cues have common features with the drug-related ones that in turn induce higher levels of craving and desire (100).

As drug-paired cues are highly salient, emotional responses are influenced when an individual receives cues (101). Therefore, when neuroimaging signals are recorded, scales are applied to evaluate changes in emotions, such as valence, arousal, and craving (100). Measuring both the neural signals and the score on emotional scales provide more comprehensive information for over multiple physiological and behavioral dimensions.

Although cues provide an opportunity to distinguish those suffering from an addiction from healthy controls, experiments in which drug-related cues are provided to the participants have potential ethical issues. For example, the risk of relapse may increase after stimulation by drug-related cues. One alternative approach has been recorded, brain signals when participants are involved in cognitive tasks, to differentiate those suffering from an addiction from healthy controls (32); for example, in one study, EEG signals were analyzed when patients with METH addiction performed cognitive tasks, including the N-back task to assess working memory and Stroop task to assess attention (102).

Furthermore, it has been reported that without cue stimulation and cognitive testing, the resting state brain of those suffering from an addiction and healthy individuals exhibit differences (94, 103). Monitoring the brain signals altered by the METH addiction and tracking the recovery process at resting state is an attractive approach because no additional efforts are needed to implement the drug-related cues and synchronize the cues and recorded signals or to arrange a supervisor to conduct cognitive assessments.

Biomarkers of neuroimaging techniques are not only used to distinguish those suffering from drug addiction from healthy controls, but they are also widely used to evaluate the efficacy of abstinence, exercise, and medical interventions. The biomarkers of EEG signals that have been shown to characterize the brain activity of those suffering from METH addiction are listed in Tables 1, 2. The biomarkers were determined by comparing the recorded EEG signals from patients with a METH addiction with those of healthy controls. The brain signals were recorded when participants received various METH-related cues (see section “4.1. Stimulation cues”), after conducting cognitive tasks, or during resting states. The EEG biomarkers that can identify the patients with a METH addiction can be categorized into three types based on the analysis approaches used. First, the time-domain EEG signals can be converted to frequency-domain signals to reveal the spectral information of EEG sub-bands. The sub-bands of each frequency range represent different conditions affected by METH (54, 85, 104–110) of Table 1. The entropy of the EEG signals at a specific frequency range can be derived from the spectrum of that frequency range (104). The second type of biomarkers are based on time-domain signals. As EEG records neural activity on a millisecond timescale, the neural signals triggered by stimuli (visual, audio, etc.) show specific wave forms, namely the event-related potential (ERP) (111, 112).

The third type of EEG biomarker can be visualized by the topography of the data (Table 2). The advantage of plotting brain signals topographically is that signal variations throughout the cortex can be evaluated. Brain activity is typically not localized to one specific region, but rather observed as coordinated activity throughout multiple connected regions. Functional connectivity (FC) in this case refers to the level of connectivity of each channel and the connectivity between cortical regions (54, 113–117). FC can be analyzed by processing an EEG signal using various approaches. Some studies have considered the coherence of the EEG sub-bands of electrode pairs (113, 116). However, coherence neglects the non-linear relationship between the channels. Therefore, another study used visibility graph similarity as a non-linear approach (113). Later, the weighted phase lag index (WPLI), a modification of the phase lag index (PLI), was used for FC analysis. Compared with coherence, PLI (117) and WPLI can better indicate the delays in signals between channels (54, 114, 115). Parameters of such graph theory analyses include node strength, characteristic path length (L), clustering coefficient (CC), and small-world index (SWI), which is derived as the ratio of CC to L. Moreover, the network hub(s) can be identified by the node strength, betweenness centrality, and eigenvalue centrality. These parameters can be used as features of ML models to classify and differentiate those with METH addiction from healthy subjects (115).

Other EEG biomarkers represented by topography include EEG microstates that show the spatial distribution of electrical signals recorded by the electrodes over the scalp (102, 118). The dynamic changes of these electric potential states can be validated by examining a consecutive set of topographies. EEG microstates represent an emerging technique for analysis of EEG signals, especially at resting states. The parameters used to investigate microstates are the mean states duration, total time covered/coverage ratio, global explained variance (GEV), occurrence, and transition probability. Studies to date have been limited because the number and type of microstates are optimized differently, depending on the database used. Fortunately, most microstates derived from the resting state of healthy subjects of different studies comprise four representative states, where each state corresponds to specific neural activity patterns, such as visual or audio responses. However, microstates of people with neurological diseases or actively engaged in task-related states (e.g., receiving cues or being involved in a cognitive task) vary across studies (118). Therefore, interpretation of results is not always straightforward, and comparisons between different studies can be challenging.

In the EEG spectrum analyses summarized in Table 1, one approach is that an individual electrode is inspected only when the number of electrodes is small (108). Alternatively, the spectrum of individual electrodes is calculated, then representative channels are selected for further investigation (111). In studies with larger numbers of electrodes, the average EEG spectrum of all electrodes is often investigated (54, 85, 104–107, 109, 112). Lu et al. and Minnerly et al. studied changes in the EEG spectrum of different brain regions (85, 110). The former separated the brain into four areas, whereas the latter separated the cortices using five different approaches. A lower number of EEG channels reduces the preparation time when the region of interest is well known. However, increasing the number of electrodes allows for studying FC across various brain regions.

Some studies had explored EEG biomarkers when subjects had their eyes closed but were not asleep; this is done to reduce the disturbance due to non-task related visual stimuli (104–106, 108, 110). Other studies report the identification of biomarkers specifically when the subjects had received cues (85, 109, 111, 112). Some studies included cognitive tasks in the experiment protocols. However, most of these studies only analyzed the correlation of the level of cognitive impairment (e.g., the reaction time and the response accuracy) and the EEG spectrum (105, 107). Few studies have monitored the variation in the EEG spectrum during cognitive tasks (108). For future applications in closed-loop neuromodulation systems, the biomarkers found when the participants were simultaneously receiving cues may be more helpful than, e.g., at a resting state, as the use of cues can more accurately simulate the conditions of having a desire for a drug.

In addition to the EEG signal, FC is often studied in the resting state as well. In the task state, the connectivity needs to be analyzed in every pair of channels at every point of interest, resulting in a heavy computational load. This is because the brain is engaged in various tasks at different stages along with the task. Only the data of a selected resting time period is calculated in the resting state. For this reason, in the task state, only the channels of interest are often analyzed to reduce computational load (54). Only one study has evaluated the effects of exercise on those with METH addiction using EEG signals (116). In this study, the EEG signals were not recorded simultaneously during the exercise, but rather before and after the acute and long-term period of exercise.

An fNIRS device is easy to wear without time-consuming preparation such as is required for EEG gel electrodes. Therefore, fNIRS devices have been widely used to study the effects of exercise on patients with METH addiction during abstinence. Optodes of fNIRS are often mounted on the prefrontal cortex and the motor cortex to study the influence of METH on decision-making as well as on cognition and motion abilities. The biomarkers of fNIRS signals are listed in Table 3. These biomarkers include a variation of hemoglobin concentration (34, 87, 89, 90, 119–122) and parameters of functional connectivity (34, 88, 121, 122). The most used hemodynamic parameter is the concentration variation of oxygenated hemoglobin, △[OxyHb]. Deoxygenated hemoglobin concentration variation and regional cerebral oxygen saturation were not found to have been used in METH-related applications. To calculate the FC of channel pairs, Pearson’s correlation is the most frequently used approach (34, 121, 122). Some studies have used the coherence of the channel pairs (88, 89). In this case, the fNIRS signals must first be converted to various frequency ranges. One study calculated the coherence of four frequency ranges (89), whereas another used only one (88). Optimization of the frequency of coherence evaluation needs further study. The fNIRS parameters include global efficiency and local efficiency, in addition to the graph theory parameters of path length and clustering coefficients (122).

The fNIRS studies in individual with METH addiction were all conducted in the past 3–4 years. Interestingly, all these studies were carried out in China, specifically by the groups of Dong (34, 88, 89, 122), Chen, and Zhou (90, 119, 121). This might be because of an increasing focus in mainland China on treating drug addiction via scientific approaches that can quantify the efficacy in treatment outcomes (123).

Most of the experimental groups in the studies summarized in Tables 1–3 consisted of those with a METH addiction before or during abstinence. This was likely because rehabilitation centers are the most convenient places to recruit participants for those studies. However, limited studies have reported how long that those in recovery from METH addiction had been abstinent, and the amount of METH taken has been seldom reported in a systematic fashion. It has been demonstrated that brain activity changes with abstinence duration (54), and that the propensity for relapse varies as a function of abstinence duration. Thus, more detailed research on a wider variety of METH user groups using fNIRS needs to be performed, and long-term follow-up of users after withdrawal may yield important insights toward development more efficacious treatments.

Transcranial direct current stimulation uses a constant low-intensity current that passes through two electrodes attached to the scalp of the participant to modulate neural activity. During tDCS modulation, a current flows between the electrodes and passes through the brain. A positive anodal current is generally considered to depolarize the neurons, thereby increasing cortical excitability and behaviors associated with the cortical region under the electrode. On the other hand, a negative cathodal current hyperpolarizes neurons, thereby inhibiting action potentials and behaviors in the corresponding cortical region.

A standard apparatus for tDCS stimulation, as shown in Figure 2C, includes a target electrode, used to stimulate the region of interest as determined by the modulation task. A reference electrode is commonly placed opposite the target electrode. Modeling studies have shown that the shorter the distance between the two electrodes, the more susceptible the current is to shunting effects. Generally, large distances between the scalp electrodes are expected to increase cortical modulation, allowing the current to be drawn through the cortex rather than shunted across the scalp (126). As the advisable safety threshold for human studies is 2 mA (127), most tDCS stimulation studies use currents between 0.5 and 2 mA with a duration between 5 and 30 min.

The dorsolateral prefrontal cortex (DLPFC) is the area most selected for tDCS stimulation of subjects with METH addiction, as dysfunction in this area has been reported frequently among these individuals. Moreover, DLPFC can also be easily targeted in a non-invasive fashion. Table 4 summarizes recent studies that have used tDCS for the treatment of addiction. Shahbabaie et al. conducted three 20-min sessions with 30 participants in a double-blinded sham-controlled trial (128). They found that the anodal tDCS of the right DLPFC decreased immediate craving at rest; however, cue-induced METH cravings could increase under active online stimulation of the right DLPFC. In another study, tDCS experiments lasting for half a year were conducted. After 20 sessions (1 month), the subject reported a significant reduction in craving and was able to control their cravings. Four booster tDCS sessions were given in the following 5 months as symptom-triggered therapy (129). These booster tDCS treatments were shown to be helpful in controlling psychological stress and drug cravings. Shahbabaie et al. conducted a double-blinded sham-controlled crossover study with 15 males with METH addiction (130). For each anticipant, 20 min sessions of real or sham 2-mA tDCS were applied over the DLPFC on two separate days in a random order. Participants receiving the real tDCS stimulation showed significant decreases in cravings. However, another clinical trial reported that cue-induced craving was reduced significantly but there were no significant effects on spontaneous cravings (131). More recently, a randomized controlled trial investigated the effects of tDCS on cue-induced craving; consistent with the findings of other studies, the results showed that tDCS could significantly reduce cravings (132). Jiang et al. applied tDCS and used two-choice oddball tasks to evaluate behavioral impulsivity prior to and after the treatment. However, they found that their protocol was not optimized to reduce symptoms associated with METH addiction (42). Khajehpour et al. conducted a sham-controlled tDCS stimulation experiment with 42 male participants with METH addiction (112). The results showed that tDCS could mitigate initial attention bias to METH stimuli. Overall, the findings summarized in Table 4 indicate that tDCS stimulation of DLPFC likely can play an active role in suppressing cravings.

The most common electrode locations in these studies are F4 (right DLPFC) for the anodal and F3 (left DLPFC) for the cathode electrodes. In these other studies, only two of the studies listed in Table 4 placed the cathode at other locations. One placed the cathode electrode on the left supraorbital area (128), and the other placed it over the right arm (129). Regarding the current used for stimulation, most studies used 2 mA, while one used 1.5 mA (132). For the evaluation of treatment effects, Shahbabaie et al. used EEG, Khajehpour et al. used fMRI, and the others used VAS or performance on cognitive tasks.

Transcranial magnetic stimulation techniques use a strong electrical current through an electromagnetic coil to generate magnetic pulses (Figure 2B). The magnetic pulse-induced electrical activity in the targeted brain area serves the purpose of neuromodulation. In practice, rTMS is used to elicit neuromodulation and neuroplasticity. Unlike tDCS, where the excitation and inhibition are controlled by anodal and cathodal stimulations, respectively, in rTMS the frequency drives the direction of neuromodulation (133). Frequencies over 5 Hz have been shown to increase cortical excitability, whereas low frequency rTMS, such as 1 Hz, decreases cortical excitability. Still another type of rTMS is the theta burst stimulation (TBS, a variation of high frequency rTMS) has also been applied to evaluate the effects on withdrawal of METH (134–137). TBS can be further separated into continuous TBS (cTBS) and intermittent TBS (iTBS), depending on the existence of intertrain intervals. The TBS consists of 3 pulses at 50 Hz forming 1 burst, and the bursts are repeated at 5 Hz. In cTBS the bursts are applied continuously while in iTBS, the bursts are applied with 2 s on and 8 s off, continuing for a designed number of cycles. TMS treatments have been reported to not only reduce feelings of craving in patients with METH addiction, but also to reduce negative emotions (such as depression and anxiety) and improve cognitive function (138–140). TMS stimulation on DLPFC also an FDA-approved treatment for depression (FDA approval K061053). Therefore, like tDCS, DLPFC is also the most commonly targeted stimulation area to modulate METH addiction (Table 5). A pilot study by Li et al. reported that low-frequency rTMS increased cravings of METH patients receiving drug-related cues (141). Liang et al. developed a protocol with 2 days rest between two 5-day treatments (142). Chen et al. designed a TMS protocol to stimulate not only the DLPFC but also the ventromedial prefrontal cortex (vmPFC) and reported an improved effect of this combined protocol (125). Zhao et al. showed that high-frequency rTMS over the DLPFC and low-frequency rTMS over the right DLPFC could reduce cravings for METH (135). Other studies have shown that TMS of the left DLPFC is effective to reduce METH addiction (142, 143). Wen et al. reported a decrease in theta/beta ratio after participants had received TMS treatment (144). Liu et al. showed that iTBS, which is a more time-efficient protocol, had similar effects to those of rTMS at 10 Hz (145). Besides the studies discussed in Table 5, some ongoing protocols have been proposed to investigate the effects of TMS treatment on METH addiction. One protocol proposes to analyze the power spectrum of EEG sub-bands when individuals receive METH-related VR cues before and after the cTBS treatment (137). Another protocol is designed to examine the effects of various stimulation parameters (146). Protocols are not listed in Table 5 since no experimental data are reported yet.

All studies in Table 5 apply self-rated VAS scores to assess the effects of TMS. Only one study uses EEG signals together with VAS scores to assess the effects (144). Some studies use images (125, 141, 143, 145), some use videos (135) and still others use VR as the cues (144). One study asks the participants to actively interact with the tools used for METH consumption (142). Since the FDA has approved TMS to treat depression, effects on emotions are compared before and after the treatments. The most common assessment is self-rating questionnaires regarding depression, anxiety, and withdrawal symptom. Results have indicated that the VAS, depression, anxiety, and withdrawal symptoms all decrease after TMS treatment (145). iTBS has become a more popular treatment approach because it takes less time to achieve the same therapeutic effect as rTMS (145). However, comparing the long-term effects of various protocols is challenging because the results reported after TMS treatments are not over the same time frame. For example, the VAS scores are measured either immediately or 4 weeks after the TMS (Figure 3). In addition, stimulation intensity is often non-consistent across studies: the published intensity has varied from 70% resting motor threshold (rMT) to 110% rMT. In the future, more data regarding the stimulation intensities and long-term effects will be needed.

Another potential confound of the studies being discussed is that the patients population within a given study is either all male, or all female. This is because of a single-gender policy that applies at the rehabilitation centers. With the wearable closed-loop system we propose in the following section, assessing potential effects of TMS in both males and females would be easier to implement without these limitations.

This system consists of multimodal EEG-fNIRS combined with TMS (Figure 4), a combination that could potentially overcome the limitations of current detection and treatment approaches. The advantages of individual module in the proposed closed-loop system are described in this section. Furthermore, an example of the scenario of applying the closed-loop system is explained in section “6.2. Apply our closed-loop system to treat METH addiction.”

The treatment often has a predefined protocol in a rehabilitation center for drug abstinence. Patients receive a fixed amount of therapy in a fixed period. In addition, the effect of the treatment is difficult to quantify. This condition limits the flexibility of customizing the treatment protocol based on the latest conditions. One scenario of applying the closed-loop system in proposed.