Portable electroencephalography in early detection of depression: Progress and future directions

Abstract

Traditional diagnostic tools for depression, such as the Patient Health Questionnaire-9, are susceptible to subjective bias, increasing the risk of misdiagnosis and emphasizing the critical need for objective biomarkers. This minireview evaluates the emerging role of portable electroencephalography (EEG) as a cost-effective, accessible solution for early depression detection. By synthesizing findings from 45 studies (selected from 764 screened articles), we highlight EEG’s capacity to identify aberrant neural oscillations associated with core depressive symptoms, including anhedonia, excessive guilt, and persistent low mood. Advances in portable systems demonstrate promising classification accuracy when integrated with machine learning algorithms, with long short-term memory models achieving > 90% accuracy in recent trials. However, persistent challenges, such as signal quality variability, motion artifacts, and limited clinical validation, hinder widespread adoption. Further innovation in sensor optimization, multimodal data integration, and real-world clinical trials is essential to translate portable EEG into a reliable diagnostic tool. This minireview underscores the transformative potential of neurotechnology in psychiatry while advocating for rigorous standardization to bridge the gap between research and clinical practice.

Key Words: Depression; Screening; Portable electroencephalography; Electroencephalography biomarkers; Frontal alpha asymmetry; Brain connectivity; Machine learning

Core Tip: The application of conventional electroencephalography (EEG) in depression screening is limited by high equipment cost, operational complexity, and low clinical applicability. This minireview highlights the current research status of portable EEG in depression screening. Systematic analysis of 45 full-text studies shows portable EEG has significant potential in feature extraction (frequency domain, time domain, time-frequency analysis, nonlinear features, and functional connectivity) and prediction capability at the single-subject level when combined with machine learning. Future studies should focus on algorithm optimization, improving data quality, and promoting clinical implementation of portable EEG for depression screening.

INTRODUCTION

Depression, particularly in moderate to severe forms, is a major global health concern, influencing around 280 million people, or 4.4% of the world population, according to a 2021 World Health Organization report[1]. Current screening methods, such as the Patient Health Questionnaire-9 (PHQ-9), rely on patient self-reporting and clinician interpretation, both of which are inherently subjective and may lead to misdiagnosis[2]. This highlights the need for objective biomarker-based screening. Electroencephalography (EEG), a non-invasive neurophysiological monitoring technique, has been widely employed in major depressive disorder research. Numerous studies have demonstrated significant alterations in brain activity between patients with depression and healthy controls[3], which can be mapped to the specific symptom dimensions defined in the Diagnostic and Statistical Manual of Mental Disorders and International Classification of Diseases diagnostic systems. For instance, anhedonia, a core symptom of depression, is associated with impaired reward processing networks in patients, as reflected by reduced P300 amplitude, feedback-related negativity, and decreased functional connectivity in the high-beta band[4,5]. Patients with feelings of worthlessness and excessive guilt exhibit enhanced connectivity between the prefrontal cortex and the cingulate gyrus, reflecting abnormalities in self-referential and negative emotional processing[6,7]. Depressed mood is associated with asymmetry in prefrontal α-wave power, reflecting emotional avoidance and a bias towards negative emotions[8]. These EEG features provide important clues for the objective diagnosis of depression, driving the application of EEG as a potential biomarker in clinical settings. Furthermore, these existing EEG differences offer a theoretical foundation for early depression screening.

However, traditional EEG equipment is cumbersome and requires professional operation, limiting its use in routine clinical environments[9]. The advent of portable EEG devices has addressed these limitations. These wireless, low-cost, and user-friendly systems can be integrated into mobile health platforms, including smartphones and tablets, enabling real-time data collection, processing, and feedback[10]. In the application of EEG technology, portable EEG devices exhibit significant differences in signal quality compared to traditional EEG systems. However, the advantages of portable EEG devices lie in their convenience and flexibility[11]. By collecting real-time EEG data and integrating it with machine learning algorithms, portable EEG can provide more accurate biomarkers for depression. As such, portable EEG devices hold great potential for early depression screening. However, to fully realize this potential, strict regulatory approval and clinical validation are necessary.

In summary, although several studies have explored the use of portable EEG in depression screening, most focused on biomarker research using conventional EEG[3]. To date, a comprehensive and systematic evaluation of the screening efficacy and clinical value of portable EEG for depression remains lacking[12]. In addition, in-depth exploration and synthesis of multimodal data fusion and deep learning applications in this field remain limited[13]. Therefore, this minireview focuses on the application of portable EEG in screening depression, summarizing recent research advances and highlighting the potential of portable EEG in feature extraction, including frequency domain, time domain, time-frequency domain, nonlinear features, and functional connectivity, as well as its integration with machine learning techniques across different types of portable EEG devices.

LITERATURE SCREENING

The literature screening process was conducted in multiple stages. Firstly, a systematic search of PubMed, Web of Science, EMBASE and IEEE Xplore databases was performed to identify studies published between January 1, 2021 and March 31, 2025, using the keywords “portable EEG”, “depression”, and “screening”. This search yielded 764 records. During the title and abstract screening phase, 719 articles were excluded because they addressed only one of the three keywords and therefore did not address their intersection. As a result, 45 articles were selected for full-text review. Among these 45 studies, 10 adopted a retrospective/offline classification design, wherein participants were pre-labeled as either depressed or healthy controls based on clinical interviews or standardized rating scales (e.g., Hamilton Rating Scale for Depression, PHQ-9) prior to model development and testing. The remaining 35 studies concentrated primarily on signal acquisition methods or algorithm development, without performing any prospective, in situ clinical validation. Specifically, none of these 35 studies involved real-time comparisons between portable EEG-derived predictions and clinical diagnoses by psychiatrists in undifferentiated patient populations, highlighting a significant gap between conceptual feasibility and real world clinical deployment. Thus, this review concentrated on the 10 articles with practical applications, assessing their quality and presenting the results in Table 1 to enhance the review’s rigor. Two co-first authors, Wang P and Dai AL, independently assessed the risk of bias using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. Inter-rater agreement was high (kappa = 0.78), and any disagreements were resolved through consensus with a third reviewer, Guo XR. Additionally, G*Power 3.1 was used to evaluate whether each study had sufficient statistical power (α = 0.05, effect size d = 0.5, power = 0.80). Each article was also graded according to the five-level evidence hierarchy of the Oxford Centre for Evidence-Based Medicine 2011 framework. For model validation bias, we examined: (1) Whether cross-validation was explicitly implemented and reported; (2) Whether data were properly split into training and independent test sets with performance reported on the latter; and (3) Whether experiments were repeated at least three times with stability metrics or confidence intervals. Although all 10 studies satisfied criteria (1) and (2), none conducted prospective, in situ concordance testing in which portable EEG predictions were compared with a psychiatrist’s real time diagnosis, underscoring a persistent gap in external validity. These assessments were detailed in Table 1. It is important to note that all 10 of these studies were laboratory-based research.

Table 1 Study quality assessment overview.

Ref.	Bias risk assessment	Sample size assessment	Validation method assessment	Evidence level (OCEBM)	Overall quality score
Khan et al[44]	Low	Reasonable	Unreasonable	5	Medium
Wu et al[34]	Unclear	Unreasonable	Unreasonable	3	Low
Tian et al[45]	Low	Unreasonable	Reasonable	2	Medium
Tian et al[38]	Low	Reasonable	Reasonable	4	High
Sakib et al[43]	Unclear	Reasonable	Reasonable	5	Medium
Sharma et al[41]	Medium	Unreasonable	Unreasonable	4	Low
Lei et al[39]	Medium	Unreasonable	Unreasonable	4	Low
Morita et al[40]	High	Reasonable	Unreasonable	5	Low
Wang et al[46]	Medium	Unreasonable	Reasonable	5	Medium
Saleem et al[42]	Unclear	Reasonable	Unreasonable	5	Medium

OCEBM: Oxford Centre for Evidence-based Medicine.

The following section summarizes the inclusion and exclusion criteria applied across studies to define participant populations, including diagnostic tools, demographic parameters, and conditions for healthy controls. Inclusion criteria including: (1) Depression patients: Most studies screened participants using the PHQ-9, with a score of ≥ 5 indicating eligibility, though some required stricter thresholds (20-27). A subset of studies confirmed diagnoses via the Diagnostic and Statistical Manual of Mental Disorders, 4^th edition. Age restrictions varied, with common ranges spanning 18-25 or 18-60 years; (2) Healthy controls: Participants in this group were matched to depression patients by age, gender, and education level, with PHQ-9 scores limited to 0-4 to ensure minimal depressive symptoms; and (3) Additional requirements: All participants were required to abstain from psychotropic medication for at least two weeks prior to the study and provide informed consent for EEG recordings. Some studies mandated a minimum of 14 measurement sessions to ensure data robustness.

Exclusion criteria including: (1) Depression-related exclusions: Individuals with recent psychotropic medication use (within two weeks), comorbid mental disorders (e.g., anxiety, epilepsy), or brain injuries were excluded to isolate depression-specific effects and minimize EEG interference; (2) Physical health exclusions: Severe cardiovascular, endocrine, or neurological conditions that could alter brain activity were grounds for exclusion; and (3) Other exclusions: Participants exhibiting severe suicidal ideation, invalid EEG data, or failure to meet protocol requirements (e.g., incomplete sessions) were also excluded. Notably, some studies omitted explicit descriptions of exclusion criteria, which may introduce confounding variables and reduce the validity of findings. Future research should prioritize transparent, comprehensive inclusion/exclusion frameworks to enhance reproducibility and clinical relevance.

TECHNICAL CHARACTERISTICS OF PORTABLE EEG

Currently, portable EEG is widely used in brain-computer interface research and mental health monitoring. Compared to traditional EEG systems, portable EEG devices typically exhibit lower signal quality, particularly in capturing high-frequency signals (such as gamma waves) and complex brain activities. However, they perform better in the acquisition of low-frequency signals (such as alpha and beta waves) and have demonstrated high reliability and effectiveness in applications such as meditation, emotional regulation, and attention training[14-17]. At present, four main types of portable EEG devices have been identified (Table 2). Head-mounted EEG, ear-worn EEG, textile-integrated EEG and implantable EEG. Among the existing portable EEG devices, head-worn EEG devices are widely used for depression screening and other cognitive function assessments due to their high usability and accurate EEG data. Next, we will focus on discussing the technical characteristics of head-worn EEG devices, including electrode types, channel count, wireless communication, data management, and intelligent signal processing.

Table 2 Categories of portable electroencephalography.

Type	Technical characteristics	Advantages	Limitations	Ref.
Head-worn EEG	Dry/wet electrodes, rigid/semi-flexible headset, 4-16 channels	Detects depression-related EEG biomarkers in key prefrontal regions, suitable for lab or remote screening	Limited comfort, prone to motion artifacts	[17-19,23,24,26,39]
Ear-worn EEG	Sensors in earplugs or ear-mounted devices, dry electrodes, 1-2 channels	All-day wearability, suitable for daily emotional monitoring, supports auditory stimulus experiments	Limited signal coverage, susceptible to EMG noise	[55-57]
Textile-integrated EEG	Sensors embedded in wearable textiles (e.g., smart caps, headbands), conductive materials	Suitable for long-term monitoring, natural user experience, home or clinical use	Lower signal quality than standard EEG, affected by textile movement	[20,58]
Implantable EEG	High-precision EEG acquisition, ultra-thin flexible electrodes for skin adhesion	Precise monitoring for severe depression, can be combined with tDCS for therapy, high-quality data	Highly invasive, limited to specific medical contexts, patch requires frequent replacement	[59]

EEG: Electroencephalography; EMG: Electromyogram; tDCS: Transcranial direct current stimulation.

Head-mounted EEG devices typically use dry electrodes. Dry electrodes offer the advantage of convenience, which makes them easy to use for long-term monitoring; however, they tend to have higher contact impedance, potentially reducing signal quality and making them more susceptible to motion artifacts, unlike wet electrodes used in traditional EEG systems, which provide superior signal quality but may affect long-term stability and comfort due to the need for electrolyte gel[18,19]. However, advancements in the design of flexible dry electrodes, such as multi-pin designs or microneedle arrays, have shown promise in improving signal quality by reducing impedance and mitigating motion artifacts[20]. Multi-channel EEG provides superior signal quality and spatial resolution, which enhances signal accuracy and sensitivity, offering a more comprehensive brain activity map[21,22]. In contrast, single-channel EEG systems capture data from only one area of the brain, limiting their spatial coverage. However, recent research has shown that despite this limitation, single-channel EEG can still be effective for depression screening. For example, studies have demonstrated that single-channel EEG devices can achieve promising classification accuracy, such as 92% accuracy, in detecting depression, particularly when analyzing metrics like alpha wave asymmetry from a single electrode site[23,24]. In terms of wireless communication, blue tooth is suitable for low power consumption and short-range transmission, but its lower transmission speed and limited range may cause a decline in signal quality, especially when the device is in motion. Wi-Fi offers higher transmission speeds and longer transmission distances, but its higher energy consumption can affect the device’s battery life[25]. High-quality EEG data acquisition requires balancing signal integrity, real-time transmission, and power consumption to ensure that signal transmission does not compromise the signal-to-noise ratio, particularly in mobile or noisy environments, where issues like body movement artifacts and contact impedance are more pronounced[26,27]. Additionally, combining local data storage with cloud computing enables long-term trend analysis and remote screening[28,29]. At the signal processing level, artificial intelligence (AI)-driven signal processing techniques combined with machine learning and deep learning models enable effective detection of depression-related brain activity, making head-mounted EEG a practical tool for real-time, affordable depression screening[23,30,31]. To better illustrate this process, Figure 1 provides a conceptual diagram that outlines the workflow of portable EEG data acquisition, feature extraction, and machine learning integration. Moreover, the integration of multimodal data in portable devices enhances disease screening by combining EEG with other biomarkers, such as heart rate variability (HRV) and electrodermal activity, improving diagnostic accuracy and enabling more reliable, early detection[32-34]. With these advantages, portable EEG shows great promise for depression screening, telemedicine applications, and personalized emotional monitoring[35-37].

Figure 1 The workflow of portable electroencephalography data acquisition, feature extraction, and machine learning integration. CNN: Convolutional neural network; RNN: Recurrent neural network; LSTM: Long short-term memory; GRU: Gated recurrent unit.

ADVANCES IN PORTABLE EEG FOR DEPRESSION SCREENING

This study screened 809 articles from multiple databases, primarily including PubMed, Scopus, and Elsevier, and ultimately included 45 full-text articles for analysis. The research findings from different types of portable EEG devices revealed consistent differences in EEG indicators among patients with depression. In the frequency domain, a reduction in alpha waves, enhancement of beta waves, and abnormalities in delta/theta waves were noteworthy; in the time domain, EEG amplitude was reduced and the P300 wave was diminished; time-frequency features revealed dynamic power changes, and prefrontal alpha power asymmetry was found in resting-state spectral analysis. Nonlinear features showed generally decreased entropy and complexity; and in terms of functional connectivity, reduced coherence between the prefrontal region and between the left and right hemispheres was found. Portable EEG screening studies often utilize frequency-domain, time-domain, time-frequency, nonlinear, and functional connectivity features, combined with machine learning or deep learning algorithms for modeling. Systematic analysis of these indicators has clarified specific EEG features and classification performance that can be used for depression screening: Traditional machine learning algorithms rely on manual feature extraction, achieving classification accuracy of up to 90.7%[38]. However, these results are typically based on small, homogeneous datasets (e.g., mainly single-center studies with sample sizes ranging from 10 to 161, often from single-center studies)[39,40]. Additionally, most studies primarily employ cross-validation techniques (e.g., 10-fold or 5-fold cross-validation) rather than independent or external validation methods. These limitations may affect the generalizability of the models in real-world applications. Deep learning methods, such as convolutional neural network (CNN), recurrent neural network (RNN), and long short-term memory (LSTM), possess greater advantages in terms of automated feature extraction and multimodal fusion, with some models achieving accuracy as high as 99.9%[41]. However, these results are typically derived from cross-validation on small datasets rather than from independent or external validation[42]. Furthermore, deep learning methods require higher computational resources and larger datasets, restricting their practical application. These studies have laid the foundation for the application of portable EEG in depression screening, while requiring further validation and replication in diverse, real-world populations to ensure their reliability and generalizability in clinical and practical settings. To more clearly summarize the various feature analysis methods of portable EEG in depression screening, we have compiled the key content of relevant studies and listed their main research progress and references (Table 3).

Table 3 Portable electroencephalography in depression detection.

Feature type	Models	Key indicators	Main results	Ref.
Frequency-domain (PSD)	Support vector machine, random forest, CNN, RNN; feature combination	Decreased frontal α power; increased β power; abnormal θ/δ power	Frontal EEG + PSD: 90.7% accuracy; combined with nonlinear features improves stability	[34,38,39,43]
Time-domain	LSTM, feature fusion (time + frequency), classifiers	Amplitude, peak, mean value	Used for signal evaluation; combination improves accuracy	[34,41,44,45]
Time-frequency (STFT/WT)	STFT, WT + nonlinear features, classifiers	α/β power; WT adapts to time scales	WT extracts nonlinear features; high accuracy; significant power asymmetry	[38,40,42,45]
Nonlinear features	Nonlinear features + classifiers	SampEn, ApEn, FD, HFD, LZC	Lower complexity in depression; LZC and HFD achieve approximately 90% accuracy	[40,43]
Functional connectivity	Coherence, PLV + classifiers	Decreased frontal coherence, PLV; reduced frequency synchronization	Reduced interhemispheric connectivity; linked to emotion regulation	[34,38]

PSD: Power spectral density; CNN: Convolutional neural network; RNN: Recurrent neural network; EEG: Electroencephalography; LSTM: Long short-term memory; STFT: Short-time Fourier transform; WT: Wavelet transform; SampEn: Sample entropy; ApEn: Approximate entropy; FD: Fractal dimension; HFD: Higuchi fractal dimension; LZC: Lempel-Ziv complexity; PLV: Phase locking value.

Power spectral density

Significant progress has been made in applying frequency-domain features to depression detection via portable EEG. Specifically, studies have found reduced frontal alpha wave activity (8-13 Hz) in depression patients, possibly affecting emotional regulation. At the same time, beta wave activity (13-30 Hz) is enhanced, associated with anxiety and excessive rumination. Depression patients also exhibit generally lower EEG amplitude, particularly in the alpha and beta bands, reflecting reduced neural activity[41,43]. Additionally, abnormalities in theta (4-8 Hz) and delta (0.5-4 Hz) waves may reflect the severity of depression[39]. Based on these frequency-domain features, some studies have combined them with machine learning algorithms for classification, with prefrontal EEG data achieving accuracy up to 90.7%[38]. To enhance detection stability, researchers have also integrated nonlinear features, such as Lempel-Ziv complexity and sample entropy (SampEn), with power spectral density (PSD) features[39]. Studies show that different frequency bands perform differently in classification tasks, and appropriate selection of bands and combinations improves screening effectiveness. Among them, the beta band (12-30 Hz) shows the strongest discriminative power in young depression patients, with models achieving up to 97.22% accuracy[43]. Finally, deep learning methods (e.g., CNNs and RNNs) have been used for automated PSD feature extraction, showing clear advantages in enhancing EEG signal recognition[34]. These developments collectively advance the application of frequency-domain features in portable EEG-based depression screening and support ongoing progress in mental health monitoring.

Time domain features (event-related potential)

Time-domain features play a key role in EEG-based depression analysis. For example, reduced P300 wave amplitude is considered an important neural marker of depression and has been validated in portable EEG detection[44,45]. Based on these findings, machine learning has been applied to improve time-domain feature analysis. For instance, LSTM networks can extract temporal features, achieving classification accuracy of up to 97.98%[41] for depression detection. In addition, advancements in portable EEG devices have made it possible to collect time-domain data with fewer electrodes, providing new technical support for remote depression screening and continuous mental health monitoring[38]. Through the use of portable EEG devices, it is now feasible to collect EEG data from the Fp1 Location on the forehead scalp, and the feasibility of depression prescreening methods in ubiquitous environments is being explored. The result indicates a promising future for developing a ubiquitous application for prescreening depression[33].

Time-frequency features

Time-frequency analysis, using wavelet transform, reveals dynamic changes of non-stationary EEG signals across multiple time scales, providing advantages over single time- or frequency-domain approaches[45,46]. In depression screening, wavelet transform facilitates the extraction of key time-frequency features, including prefrontal alpha power asymmetry. Combined with nonlinear indicators like Lempel-Ziv complexity and SampEn, the proposed system achieves a classification accuracy of 90.70% and an F1-score of 87.33%[38,45]. Moreover, the proposed system achieves a classification accuracy of 95% and maintains a success rate above 95%, enabling effective mental health detection in real-time applications[42]. These findings underscore the role of time-frequency analysis and nonlinear EEG features in advancing portable, objective tools for mental health assessment.

Nonlinear EEG features

In recent years, nonlinear features have been widely used in EEG signal analysis for depression. Studies generally show that the EEG signals of depression patients have lower complexity. SampEn and approximate entropy, which measure signal irregularity, are generally lower in depression patients, indicating that their EEG signals are more predictable. This difference is especially pronounced in the frontal region. Similarly, fractal dimension, another measure of EEG signal complexity, is also lower in depression patients, particularly in the frontal and parietal regions[43]. On this basis, researchers have further explored nonlinear dynamic analysis and multi-scale entropy methods to optimize EEG-based depression classification performance, achieving accuracy up to 90.70% with k-nearest neighbors, and above 88% with random forest and extreme gradient boosting, indicating the effectiveness of nonlinear feature weighting strategies[40]. Additionally, nonlinear features such as Hjorth complexity, Shannon entropy, and Log energy entropy have also been utilized to analyze the complexity and randomness of EEG signals, showing promising classification performance in depression detection[43]. In the future, combining multiple nonlinear features is expected to further improve detection accuracy.

Functional connectivity features: Inter-regional coherence

Studies have shown that EEG connectivity between the left and right hemispheres is reduced in depression patients, with this phenomenon particularly evident in the prefrontal region. Further resting-state EEG analyses found that prefrontal coherence in individuals with depression is significantly decreased, suggesting functional impairment in neural networks related to emotional regulation[34]. Currently, coherence analysis and phase-locking value methods are widely used to detect neural functional abnormalities in depression. In portable EEG device research, scholars have developed three-lead EEG sensors that can effectively capture coherence features in the prefrontal region, further improving depression detection accuracy when combined with machine learning algorithms. Additionally, phase-locking value analysis has revealed that depression patients show lower phase synchronization in low-frequency bands (such as alpha and theta), indicating impaired inter-regional information transmission[38]. These research outcomes have established a foundation for applying portable EEG in depression detection. Yet, to guarantee its reliability and generalizability in clinical and real-world settings, more validation and replication in diverse, real-world populations is needed.

In the analysis of relevant literature, the impact of equipment heterogeneity on the consistency of biomarkers is significant. Single-channel and multi-channel systems often yield contradictory results in connectivity analysis. Single-channel systems may miss key connectivity patterns due to limited information collection, thus affecting the consistency of biomarkers. For instance, the single-channel dry-electrode ear-worn EEG device shows relatively low consistency rates with PHQ-9 and Hamilton Rating Scale for Depression in different studies[39]. In contrast, multi-channel head-mounted EEG devices combined with models like the deep adaptation network achieve better performance. On the multi-modal open dataset, the accuracy reaches 77.0% ± 9.7%, with relatively high sensitivity and specificity[34]. The 14-channel Emotiv Epoc+ wireless EEG device (salt-water-based electrodes) used with the k-nearest neighbor model shows remarkable performance in the alpha and beta frequency band (8-30 Hz). The 5-fold cross-validation accuracy is 98.43% ± 0.15%, and the overall accuracy under a 70/30 training/test data split is 98.10% ± 0.11%. Its sensitivity and specificity are also excellent[43]. These examples highlight the advantages of multi-channel devices in capturing rich EEG information and improving the consistency of biomarkers and model accuracy.

Compared to traditional EEG, portable EEG has several advantages: (1) Portability and ease of use: Portable EEG devices are small and light, easy to carry and operate, available in various environments (like homes and communities), improving accessibility, and enabling early depression detection and large-scale epidemiological surveys; (2) Simplified operation process: With dry electrodes or simplified electrode configurations, portable EEG devices reduce electrode placement time and dependence on professionals, lower skin preparation requirements, and make data collection more efficient; (3) Multi-scenario applicability: Portable EEG devices are applicable in multiple scenarios, including medical, research, and educational settings, offering new tools and technical support for early depression diagnosis, treatment, and prevention[34]; (4) Feature extraction and analysis capability: Despite fewer electrodes, portable EEG devices can fully utilize limited electrode information through information fusion and feature optimization algorithms to extract highly distinctive features. For example, in frequency-domain analysis, portable EEG can detect reduced frontal alpha wave activity and enhanced beta wave activity in depression patients, and combining nonlinear features can further boost classification accuracy[38,39,43]; and (5) Application of deep learning methods: When combined with deep learning methods (like CNN and RNN), portable EEG devices excel in automatically extracting PSD features, with some models achieving 99.9% accuracy[41], indicating their feature extraction and analysis capabilities are on par with or even surpass those of traditional EEG.

CHALLENGE AND FUTURE DIRECTIONS

Despite the growing interest in portable EEG for depression screening, several technical challenges remain. Portable EEG devices are highly susceptible to motion artefacts and environmental noise, which can compromise signal quality. The absence of standardized protocols and incompatible data formats further limit their clinical translation and large-scale application[47]. Additionally, restrictions in battery life and computing power may lead to delays in cloud-based analysis, making real-time screening difficult to achieve[48]. Furthermore, ensuring robust data privacy and protection remains a critical concern for both patients and clinicians[49,50]. To address these challenges and advance the field, future research should focus on several key areas. First, optimization of EEG sensors is essential. Improvements in dry electrode technology to minimize signal drift and enhance wearing comfort are necessary. The use of flexible biocompatible materials will further enhance device adaptability for long-term monitoring. The integration of multimodal data features, including portable EEG and HRV, demonstrates great application potential. Studies have shown that combining portable single-channel EEG signals with HRV data obtained from photoplethysmography sensors can effectively monitor relaxation states and psychological stress levels, and shows significant correlations with self-reported depression, anxiety, and stress scores, thereby validating the feasibility of using portable devices for psychophysiological monitoring in non-laboratory settings[33]. The integration of attention mechanisms may facilitate the extraction of key EEG features while mitigating irrelevant noise[51,52], and multimodal fusion may further reduce misdiagnosis rates. In addition, expanding remote depression screening applications through mobile health platforms will enable patients to perform at-home monitoring, with cloud-based AI analysis and remote consultation from healthcare providers[53]. Combining portable EEG with cognitive behavioral therapy or brain-computer interface technologies could support personalized mental health interventions[37]. Finally, strengthening data security and privacy through blockchain technology for encrypted storage and access control, as well as the use of differential privacy techniques, will enhance patient trust and data integrity[49,50]. Collectively, these advancements in sensor optimization, AI development, multimodal data integration, telemedicine, and data security will pave the way for broader clinical application of portable EEG in depression screening and mental health management.

However, although laboratory results show classification accuracies from 77% to 98%, these figures are mostly derived from limited sample sizes, internal cross-validation, or single-center trials. To translate this technology into clinical practice and secure United States Food and Drug Administration (or equivalent) approval, a rigorous regulatory process is required. First, comprehensive technical validation and risk assessment must be completed to demonstrate the model’s stability across diverse populations and multi-center datasets; second, small-scale clinical trials are needed to confirm safety and feasibility in actual patient groups; and finally, large-scale, randomized, multi-center clinical trials should systematically evaluate sensitivity, specificity, and cost-effectiveness relative to traditional screening methods such as the PHQ-9 and clinical interviews. According to United States Food and Drug Administration guidelines for software as a medical device, this process also requires detailed algorithm validation metrics, data security documentation, and a post-market surveillance plan to ensure long-term reliability and safety. While PHQ-9 is a cost-effective self-report tool widely used in clinical settings, portable EEG offers an objective neurophysiological marker that may enhance diagnostic accuracy, though at a higher initial cost. However, given that our included studies predominantly used PHQ-9 scores to differentiate between depressed and non-depressed individuals, further research is needed to directly compare the cost-effectiveness of portable EEG against PHQ-9 in real-world clinical applications[54].

CONCLUSION

Portable EEG has demonstrated strong performance in depression screening, characterized by reduced alpha power and increased beta power in the frequency domain, decreased P300 amplitude in the time domain, and notable alterations in time-frequency and nonlinear features. Temporal features extracted using a LSTM network have achieved classification accuracies of up to 97.98%. Nevertheless, challenges remain regarding signal quality, the absence of standardized protocols, and limited algorithm generalization. These issues call for continued efforts in sensor optimization, multimodal data fusion, and algorithm refinement to advance the clinical implementation of intelligent screening systems.