Neurofeedback in psychiatry: A decade of clinical and neuroimaging insights – a systematic review

Review type and protocol

This review was conducted in accordance with the PRISMA 2020 guidelines,^[3] and the completed PRISMA checklist is provided in the Supplementary Table S1. The protocol was prospectively registered in PROSPERO (CRD420251090789; https://www.crd.york.ac.uk/PROSPERO/view/CRD420251090789) before study selection and data extraction. The methodology was designed to align with best practices for systematic reviews. Any protocol modifications made during the review process (e.g., broadening inclusion to case series with clinical outcomes) were documented with justification. All data were obtained from published sources, and no ethical approval was required. A medical librarian contributed to the development of the search strategy, and two independent reviewers designed and piloted the protocol. Disagreements were resolved through consultation with a senior reviewer.

Artificial intelligence (AI) was not used for data extraction or analysis. Limited AI support was used only for formatting, language editing, and duplicate removal (through [Rayyan AI-duplicate removal]). All AI outputs were manually verified by the review team.

Inclusion criteria

• Participants: Studies enrolling individuals with a formally diagnosed psychiatric disorder (based on the Diagnostic and Statistical Manual or International Classification of Diseases criteria). Both pediatric and adult populations were eligible, without restrictions on sex, ethnicity, or clinical subtypes (e.g., ADHD, MDD, PTSD, schizophrenia, SUD)

• Interventions: Any form of NF intervention (EEG, fMRI, fNIRS, or quantitative electroencephalography [QEEG]-guided) provided participants received real-time feedback of their own brain activity. Studies of NF-assisted meditation or brain-computer interface training were included if real-time neural feedback was a central component

• Comparators: Sham NF, active comparators (e.g., cognitive training, psychotherapy, medication), waitlist, or pre-post single-group designs without formal comparators

• Outcomes: Clinical outcomes (primary: Symptom severity scales such as ADHD-rating scale [RS], Hamilton Depression (HAM-D), Montgomery-Åsberg Depression (MAD) RS, Clinician-Administered PTSD Scale [CAPS], positive and negative syndrome scale [PANSS]; secondary: Cognitive/behavioral tasks, functioning, quality of life) and/or neurobiological outcomes (e.g., EEG spectral changes, fMRI activation/connectivity measures). Outcomes had to be assessed with standardized methods or validated instruments

• Study designs: Randomized controlled trials (RCTs), non-RCTs, single-arm open-label studies, case series, and case reports

• Publication types: Peer-reviewed full-text articles

• Language: Studies published in English.

Exclusion criteria

• Participants: Studies without a formally diagnosed psychiatric disorder or those focused exclusively on healthy participants

• Interventions: Studies not involving actual NF training (e.g., neuromodulation mislabeled as NF, purely theoretical descriptions, or interventions without real-time neural signal feedback)

• Comparators: Not applicable studies without comparators were allowed, but those without any evaluable outcome measures were excluded

• Outcomes: Studies that did not report clinical or neurophysiological outcomes (e.g., protocol papers, conceptual frameworks)

• Study Designs: Conference abstracts, dissertations, unpublished works, or secondary research (systematic reviews, meta-analyses)

• Publication Types: Gray literature, opinion pieces, or editorials.

Search strategy

An initial scoping search was conducted to map the scope of NF studies in psychiatric populations. The final systematic search covered January 2015 to March 2025 across four databases: PubMed, PsycINFO, Web of Science, and Scopus.

The search strategy was developed in collaboration with a medical librarian and the review team. Core concepts were NF modalities (“neurofeedback,” “EEG biofeedback,” “real-time fMRI,” “rtfMRI,” “fNIRS neurofeedback”), psychiatric conditions (“ADHD,” “attention-deficit,” “depression,” “major depressive,” “anxiety,” “PTSD,” “post-traumatic stress,” “OCD,” “obsessive-compulsive,” “schizophrenia,” “bipolar,” “addiction,” “substance use”), and outcomes (“symptoms,” “clinical trial,” “RCT,” “connectivity,” “EEG,” “fMRI”). Boolean operators combined these domains, and filters were not applied at this stage. Reference lists of relevant primary studies were also hand-searched to identify additional eligible trials. No formal gray literature search was performed, and no unpublished data were sought from study authors. Non-English studies were retrieved during the search but later excluded at the eligibility stage. The full PubMed search string and search history are provided in Supplementary Text 1.

Study screening and selection

All records retrieved were imported into systematic review software (Rayyan QCRI) for screening. Duplicate removal was performed automatically and verified manually by two reviewers (JR and RR). Titles and abstracts were independently screened to exclude clearly irrelevant records (e.g., non-NF interventions, non-psychiatric populations). Full-text articles of potentially eligible studies were obtained through institutional subscriptions and open-access sources. Two reviewers (JR and RR) independently assessed full-text eligibility, with disagreements resolved by a senior reviewer (NG). All eligibility decisions followed the predefined inclusion and exclusion criteria (see “Eligibility Criteria” section).

Data extraction and synthesis

For each included study, data were extracted into pre-piloted forms covering participant characteristics (psychiatric diagnosis, demographics, sample size, and comorbidities), intervention details (NF modality such as EEG, fMRI, fNIRS, or QEEG, targeted frequency band or brain region, feedback type, number and duration of sessions), comparator conditions (sham feedback, active comparators, waitlist, treatment-as-usual [TAU], or none), and outcomes.

Efficacy outcomes included standardized symptom scales (e.g., ADHD-RS, HAM-D, MADRS, CAPS, PANSS), clinician ratings, neuropsychological tests, and responder/remission rates. Neurobiological outcomes included EEG or fMRI target modulation and brain–behavior correlations. Safety outcomes included adverse events and dropout rates. Where available, reported effect sizes (e.g., Cohen’s d, SMD), P-values, and follow-up findings were recorded.

Data extraction was performed independently by two reviewers (JR and RR) and cross-verified by a senior reviewer (NG), with discrepancies resolved by consensus. No data imputation methods were used, and analyses relied solely on published values.

Due to heterogeneity in study designs, comparators, and outcome measures, a new quantitative meta-analysis was not feasible. Instead, we conducted a narrative synthesis organized by disorder category (ADHD, depression, anxiety/OCD, PTSD, schizophrenia, and SUDs), summarizing findings from RCTs and other eligible primary studies. The synthesis highlighted consistencies and discrepancies and described neurophysiological findings such as EEG or fMRI changes. Comparisons were also drawn between EEG- and fMRI-based NF, pediatric versus adult populations, and cross-cutting themes such as durability of effects, placebo influences, and mechanistic insights, in line with PRISMA recommendations for qualitative synthesis.

RoB assessment

Methodological quality was assessed at the individual study level. For RCTs, two independent reviewers (JR and RR) assessed RoB using the revised Cochrane RoB tool (RoB2), covering the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selective reporting. A third senior reviewer (NG) cross-verified assessments and resolved disagreements. Figures illustrating the RoB assessments (traffic light plots and domain summaries) were generated [Supplementary Figure S1]. Most RCTs were rated as having some RoB, most commonly due to challenges in blinding participants and personnel in NF studies, where sham feedback may be distinguishable.^[4]

For non-randomized and uncontrolled studies, no formal RoB scoring tool was applied; their inherent limitations (e.g., absence of controls, expectancy effects, placebo influences) were noted when interpreting results. In addition, all included studies were qualitatively assessed against the CRED-NF checklist [Supplementary Table S2] to evaluate transparency in reporting NF-specific aspects such as blinding, NF learning verification, and adverse event reporting.

Data analysis

The primary outcomes were clinical symptom changes measured with standardized instruments (e.g., ADHD-RS, HAM-D, MADRS, PANSS, CAPS). Secondary outcomes included neurobiological markers (e.g., EEG spectral changes, fMRI activation/connectivity) and tolerability (adverse events, dropout rates). Reported effect sizes (e.g., Cohen’s d, SMD) and P-values were extracted directly from individual studies.

No new quantitative synthesis was conducted due to heterogeneity of interventions, comparators, and outcome measures. Instead, we performed a narrative synthesis organized by disorder category (ADHD, depression, anxiety/OCD, PTSD, schizophrenia, and SUDs), highlighting findings from RCTs and other eligible primary studies. The synthesis emphasized consistencies and discrepancies across studies, described neurophysiological changes associated with NF, and compared results between EEG- and fMRI-based protocols and across pediatric versus adult populations.

Study selection

The systematic search identified 1,220 records. After removal of 200 duplicates, 1020 titles and abstracts were screened. Of these, 900 records were excluded, primarily because they were unrelated to NF or focused on healthy populations. A total of 120 full-text articles were assessed, and 45 primary studies met eligibility criteria [Supplementary Figure S1, PRISMA flow diagram]. These comprised 28 RCTs, 12 controlled or quasi-experimental studies, and 5 open-label studies or case series, collectively involving more than 4,000 participants across psychiatric disorders.

The 75 full-text exclusions were due to (i) lack of a true NF intervention (n = 28), (ii) inadequate outcome reporting (n = 20), (iii) duplicate or overlapping datasets (n = 12), and (iv) secondary research articles such as systematic reviews and meta-analyses (n = 15).

Characteristics of the included studies

Study characteristics are summarized in Table 1. Most studies were conducted in North America, Europe, and

East Asia. ADHD studies predominantly enrolled pediatric samples (mean age ~8–12 years), while depression, PTSD, and schizophrenia studies targeted adults, often treatment-resistant. Across conditions, about 40–55% of participants were female.

NF protocols varied:

• EEG-based NF included TBR, SMR, slow cortical potentials (SCP), FAA, and Alpha–Theta training

• fMRI-based NF targeted the amygdala, prefrontal cortex, anterior cingulate cortex, and insula

• Session dose ranged from 5 to 40 sessions (mean ≈ 30)

• Comparators included sham NF (pre-recorded signals), waitlist, TAU, or active therapies

• Primary outcomes were disorder-specific symptom scales (ADHD-RS, HAM-D, MADRS, CAPS, PANSS, craving scales), with secondary outcomes including neuropsychological tasks, neuroimaging or EEG measures, and tolerability.

RoB assessment

RoB assessment using RoB2 showed that most RCTs were rated as “some concerns”, primarily due to challenges in blinding participants and therapists to active versus sham NF. Only ~20% of RCTs achieved a low RoB across all domains. Sham conditions were often distinguishable, particularly in EEG-NF where feedback was less credible. Non-randomized and open-label studies were judged at high RoB, and results were interpreted with caution. Supplementary Figure S2 presents traffic-light plots of individual studies.

Qualitative synthesis

Depressive disorder

NF applications in depression have expanded but remain less established than in ADHD. Research includes open-label pilots and small RCTs, using both EEG- and fMRI-based protocols. EEG approaches often target the FAA to rebalance reduced left-frontal activity linked to negative affect by increasing left frontal activation (e.g., F3 electrode). Other EEG protocols involve beta up-training for alertness or SMR training addressing anxiety or insomnia. fMRI NF typically targets deep limbic regions such as the amygdala, which shows blunted activation in depression. Young et al. (2017) demonstrated that depressed patients could upregulate left amygdala activity through real-time fMRI NF, resulting in significant symptom improvement compared to sham controls.^[8]

Clinically, NF shows promise as an adjunctive treatment, often yielding moderate reductions in depressive symptoms. Cheon et al. (2016) reported about a 50% decrease in HAM-D scores over eight weeks of EEG-NF with 75% responders despite lacking controls.^[9] The robust double-blind RCT by Young et al. (2017; n = 36) showed a 45% reduction in MADRS scores with amygdala NF versus 15% in controls, with 63% responders and sustained post-training effects.^[8]

Neurobiological findings are mixed. EEG studies variably report FAA shifts post-NF; some show no significant asymmetry changes despite clinical gains,^[9] suggesting alternative mechanisms such as enhanced cognitive control or self-efficacy. fMRI NF studies consistently demonstrate increased activation and connectivity in targeted regions correlating with symptom improvement.^[10] Dual EEG-fMRI NF approaches show promise but remain experimental.^[11]

Given depression’s heterogeneity, personalized NF targeting may enhance outcomes. Combining NF with psychotherapy may yield synergistic effects but requires further study. Engagement challenges among severely depressed patients highlight the need for motivating training paradigms. Overall, NF in depression remains investigational but promising, especially for treatment-resistant cases. Larger controlled trials and individualized protocols are needed to clarify efficacy.

Anxiety disorders and OCD

NF has been applied to various anxiety disorders, including GAD, social anxiety, specific phobias, panic disorder, and OCD, as well as subclinical anxiety and stress. Most studies employ EEG-NF protocols such as alpha enhancement (promoting relaxation), alpha asymmetry training (reducing negative affect bias), beta down-training (reducing hyperarousal), and alpha–theta training to induce a calm, meditative state. In OCD, SMR up-training is used to stabilize arousal and improve impulse control.

In GAD, controlled trials such as Hou et al. (2021) showed that 10 sessions of parietal alpha enhancement significantly reduced trait anxiety and depressive symptoms compared to baseline or controls.^[12] Another small trial comparing alpha– theta and SMR protocols suggested both reduce symptoms, with SMR showing greater benefit for worry and insomnia.^[13]

OCD NF research remains limited; case series have reported symptom reductions and EEG normalization, but robust controlled trials are lacking.^[14] A sham-controlled trial in high-trait anxiety^[15] found no difference between active and sham NF, highlighting strong placebo effects and the need for well-blinded studies.

Neurophysiologically, NF in anxiety typically increases alpha power and reduces beta activity, consistent with lowered cortical arousal. GAD NF has been linked to normalization of frontal TBRs and decreased midline beta associated with worry. Some studies also report improved heart rate variability, reflecting enhanced parasympathetic tone. In OCD, frontal beta reduction correlates with symptom improvement. Limited fMRI evidence suggests that NF may reduce amygdala hyperactivity in response to stress, indicating limbic modulation.

In summary, NF shows potential as an adjunctive treatment for anxiety disorders, with the strongest evidence in GAD and trauma-related anxiety and more modest preliminary support in OCD. Placebo effects complicate interpretation, but NF’s safety and patient preference for self-regulation support continued rigorous trials with credible sham controls and protocol standardization.

PTSD

PTSD has become a major focus of NF research due to high non-response rates to conventional treatments and NF’s brain-based approach targeting fear circuitry. Most studies use EEG-based NF, primarily the alpha–theta protocol, which trains patients to increase theta (4–8 Hz) and alpha (8–12 Hz) waves while suppressing high-beta activity, inducing deep relaxation that facilitates trauma processing. SMR training has also been employed to improve sleep and reduce hyperarousal. More recently, real-time fMRI NF targeting regions such as the amygdala, insula, and posterior cingulate cortex has emerged,^[16,17] alongside hybrid EEG– fMRI methods like amygdala-derived EEG pattern NF.^[18]

Clinically, NF demonstrates strong efficacy in reducing PTSD symptoms. Van der Kolk et al. (2016) conducted a pivotal RCT comparing 30 sessions of alpha–theta NF to group therapy in 52 chronic PTSD patients, with NF yielding significantly greater symptom reductions; 61% achieved >40% improvement and 20% lost their PTSD diagnosis versus none in controls, confirming benefits beyond placebo or therapeutic interaction.^[19] fMRI-NF studies show patients can learn to downregulate amygdala activity during trauma cues, though clinical gains may require more sessions or adjunct therapy.^[16]

Neurophysiological findings include increased alpha/theta power and decreased high-beta activity, consistent with reduced hyperarousal. Neuroimaging confirms reduced amygdala reactivity and enhanced prefrontal control after NF. Hybrid NF methods using “amygdala electric fingerprinting,” which derive individualized EEG signatures correlated with amygdala activity measured through simultaneous fMRI, enable targeted EEG NF to modulate deep limbic circuits, reducing PTSD symptoms and amygdala activation.^[18]

NF alleviates symptoms such as hyperarousal, insomnia, and affect dysregulation, though flashbacks and nightmares may require additional treatment. Patients report deeper relaxation and improved functioning. Despite protocol variability, emerging guidelines like CRED-NF aim to standardize treatment. NF offers a non-trauma-exposure option attractive to patients unable or unwilling to engage in traditional therapy. Growing RCT evidence suggests that NF is a promising adjunctive treatment for PTSD, pending further optimization and long-term follow-up data.

Schizophrenia and psychotic disorders

NF has been explored as an adjunctive treatment for schizophrenia, targeting persistent auditory hallucinations, cognitive impairments, and negative symptoms that are often resistant to medication. Most investigations have employed EEG-based protocols, such as SMR/beta up-training to stabilize brain rhythms, alpha enhancement to regulate arousal, and beta modulation to improve cognitive control. fMRI-based NF remains less common, but early pilot studies suggest potential for modulating fronto-limbic circuits.

Clinical findings highlight additional benefits when NF is combined with antipsychotic treatment. Markiewicz et al. (2024) reported that NF plus medication produced greater reductions in PANSS-negative symptoms and increased serum reelin levels compared to medication alone.^[20] In a randomized sham-controlled trial, Li et al. (2024) demonstrated that NF significantly reduced aggression and total PANSS scores in impulsive inpatients with schizophrenia.^[21] An intensive short-course NF protocol tested by Turiaco et al. (2024) improved cognitive domains such as working memory and executive function, although total PANSS scores remained unchanged.^[22]

Neurophysiological studies show that NF can increase resting alpha activity (associated with calmness), improve sensory gating (P50), and strengthen frontal–temporal connectivity, supporting executive function and auditory processing. While adherence can be challenging due to cognitive deficits and paranoia, interactive and game-like NF paradigms enhance engagement. Across studies, NF has been well tolerated with no serious adverse events.

Overall, NF is a feasible and safe adjunct in schizophrenia, with particular promise for alleviating negative symptoms and enhancing cognition. Larger, standardized RCTs adhering to frameworks such as CRED-NF are needed to refine protocols and establish efficacy.

SUDs

RCTs have explored NF as an adjunctive intervention for alcohol and drug dependence, though evidence remains preliminary. EEG-based NF has primarily used theta/SMR protocols, while real-time fMRI NF has targeted craving-related brain circuits.

In a controlled trial among forensic psychiatric patients with comorbid SUD, Fielenbach et al. (2018) tested 20 sessions of theta/SMR NF plus TAU versus TAU alone. NF increased SMR amplitude significantly but did not produce superior reductions in drug craving, impulsivity, or substance use compared with TAU.^[23] This suggests that while patients can learn to modulate neural activity, clinical benefits may be modest without optimized protocols.

Building on EEG findings, the SyBil-AA trial^[24] is a large ongoing single-blind RCT (n = 100) in alcohol use disorder, testing real-time fMRI NF aimed at modulating ventral striatum and prefrontal control regions during alcohol cue exposure. Patients undergo three NF sessions with relapse prevention follow-up. While outcomes are pending, this trial represents one of the most rigorous attempts to establish NF’s efficacy in alcohol dependence.

The current RCT evidence in SUD indicates that NF can successfully modulate targeted brain rhythms or regions, but robust clinical effects beyond standard treatment remain uncertain. Larger, well-controlled trials such as SyBil-AA are expected to clarify its role in relapse prevention and craving regulation.

Placebo response across disorders

The magnitude of sham response varies by disorder. In pediatric ADHD, Arnold et al. (2021) found children in both NF and sham conditions improved markedly,^[5] with no significant group differences, suggesting expectancy, intensive therapist interaction, and parent/teacher involvement drive substantial placebo gains. In depression, the placebo effect appears smaller; Young et al. (2017) demonstrated that real NF clearly outperformed sham.^[8] PTSD lies between these extremes: sham groups show improvement, but RCTs such as van der Kolk et al. (2016) provide evidence that NF yields disorder-specific benefits beyond placebo.^[19] This variability underscores the importance of disorder-specific rigor when interpreting NF effects.

Elucidating how NF produces clinical change remains a central research goal. The prevailing model suggests that NF induces neuroplasticity through operant conditioning: when patients generate target signals (e.g., increased EEG beta/SMR or fMRI BOLD changes), immediate positive feedback reinforces these patterns through reward circuits. Repeated training strengthens neural networks, yielding durable self-regulation. Empirical evidence shows lasting changes, such as reduced TBR in ADHD or increased frontal alpha power in depression, paralleling symptom improvements. Cognitive– behavioral processes such as attention and relaxation may further enhance benefits. Neuroimaging confirms that NF trainees can volitionally modulate oscillations and deep-brain activity. For example, Young et al. (2017) showed depressed patients upregulated left amygdala activity, improving recall of positive memories and reducing symptoms.^[8] In PTSD and anxiety, alpha-based NF reduces hyperarousal by downregulating fear circuits; in ADHD, beta/SMR training suppresses default-mode overactivity, improving attention. Notably, brain changes often correlate with clinical gains, supporting NF’s mechanistic specificity. Unlike a placebo, NF effects typically persist or strengthen over follow-up, suggesting enduring neuroplastic change.

Despite encouraging findings, methodological issues limit confidence. Many trials remain small, heterogeneous, or uncontrolled, with large effects in open-label studies often not replicated under stricter designs. Blinding is a persistent challenge since participants often detect sham feedback. Arnold et al. (2021) found no NF advantage over sham in ADHD under rigorous blinding,^[5] and Mennella et al. (2017) observed similar outcomes in active and sham NF for depression,^[15] highlighting expectancy effects.

To improve validity, experts recommend incorporating objective outcomes (e.g., cognitive tests and blinded ratings), pre-registering analyses, using credible active controls, and increasing sample sizes. Consensus frameworks such as CRED-NF^[2] aim to standardize design and reporting. Overall, NF shows promise as a neuromodulation therapy, but overcoming methodological weaknesses is essential to confirm its clinical efficacy.