Effectiveness of fMRI-based Neurofeedback Therapy on Depression: A Systematic Review

Abstract

Objective

This systematic review explores the efficacy of fMRI-based neurofeedback as an alternative treatment for individuals with depression who do not respond well to traditional therapies.

Methods

A comprehensive literature search across several databases (PubMed, Web of Science, ProQuest, Cochrane Library, and Google Scholar) identified 649 articles related to neurofeedback and depression, ultimately including 26 studies after eligibility assessments.

Results

The majority of these studies, comprising both open-label and randomized controlled trials, reported significant reductions in depressive symptoms following neurofeedback interventions, which primarily targeted amygdala activity. Eighteen studies showed moderate to large effect sizes (Cohen’s d values from 0.38 to 1.50) on validated depression measures. Notably, neurobiological changes were observed, including increased Blood Oxygenation Level Dependent (BOLD) activity and connectivity in emotion regulation regions such as the amygdala and prefrontal cortex. Some studies indicated sustained symptom improvements, with over 75% of patients applying learned strategies in daily life. However, methodological limitations and a moderate to high risk of bias were noted in several studies.

Conclusion

fMRI-based neurofeedback presents a promising adjunctive therapy for depression, especially for patients who are resistant to standard treatments. Although current evidence shows positive effects on neural function and clinical outcomes, addressing methodological issues and standardizing study designs is essential. Future research should emphasize larger sample sizes and long-term efficacy while enhancing accessibility and cost-effectiveness in diverse clinical environments.

INTRODUCTION

Stressful life events, such as divorce, job loss, or financial issues, can trigger the onset of depression [1]. It has been suggested that these types of important life events should be routinely measured clinically in order to prevent or manage depression [2,3]. Depression is a prevalent mental disorder affecting both physical and mental health, primarily manifesting as major depression and dysthymic disorder [4,5]. Symptoms include loss of interest in activities, sleep disturbances, mood swings, appetite changes, feelings of worthlessness, distorted thoughts, and suicidal ideation, leading to significant social and occupational impairment [6,7]. Physical symptoms such as irritable bowel syndrome, chronic fatigue syndrome, and fibromyalgia can occur in individuals with depression, often without a neurological diagnosis [8,9]. Depression is the second most common disorder in U.S. mental hospitals, with a prevalence in the community estimated to be five times higher than that of schizophrenia [10]. Globally, approximately 280 million people experience depression, and an estimated 5% of adults suffer from depression around world [11]. Depressive disorders are a significant contributor to the global burden of disease. In 2021, globally, 56.33 million disability-adjusted life years (DALYs) were attributable to depressive disorders, with an age-standardized DALY rate of 681.14 per 100,000 people [12]. Depression is more common in women and the risk of cardiovascular diseases is also higher in people with depression [13]. Depression is a common disease that is recognized by the World Health Organization as the largest cause of global disability and affects more than 300 million people worldwide [14]. Although cultural, biological and psychological factors play a role in the prevalence of depression, the root cause of this disorder is still unclear and the most worrying aspect is that teenagers who suffer from depression are 30% more likely to commit suicide than other people [15,16]. It has been observed that about one-third of depressed patients do not respond well to treatments, and this causes a decrease in the level of psycho-social functioning and deterioration of mood [17,18]. Hence, many efforts have been made to develop treatments in pharmaceutical and non-pharmacological fields [19].

A systematic review has revealed that individuals with depression exhibit atypical utilization of the brain network involved in emotion regulation [20]. At the neurobiological level, depression diminishes emotional responses in the frontal, parietal, and subcortical regions [21,22]. In individuals with depression, the volume of gray matter in the ventrolateral prefrontal cortex (VLPFC) is also reduced. Given its structural and functional connections to the frontal, motor, and language areas, the VLPFC contributes to a cognitive control network within the fronto-parietal region, facilitating regulatory processes and playing a crucial role in emotion regulation [23,24]. In addition, VLPFC, dorsolateral prefrontal cortex (dLPFC), and left superior temporal gyrus (STG) are less active in individuals with mood disorders such as depression due to attentional mismanagement and inhibitory regions [25,26]. Accordingly, in recent decades, numerous therapeutic techniques have been developed to target neural circuits involved in sustaining dysfunctional cognitive patterns and modifying their activity in response to treatment. Neuromodulation approaches that focus on emotion regulation networks and emphasize cognitive reappraisal have shown promise in alleviating symptoms of depression. For instance, repetitive transcranial magnetic stimulation (rTMS) is being explored as a potential treatment for resistant depression [27,28]. However, rTMS can cause side effects like headaches and scalp discomfort, and in rare cases, seizures. Its effectiveness varies among individuals, with not all patients responding favorably. There’s also limited research on long-term efficacy, and the treatment requires skilled professionals, which may not be available in all areas. Consequently, rTMS is not a universal solution and may not be suitable for everyone [29].

Another common treatment for depression is neurofeedback. This non-invasive technique provides real-time feedback to help individuals regulate their neural activity. The feedback is focused on areas identified as hypothetical neural substrates associated with specific behaviors or conditions [30]. Neurofeedback training directly targets the brain and plays a role in the proper management of attention and concentration [31]. Neurofeedback helps patients develop personal strategies that are effective in self-regulating brain networks with mental imagery. The fundamental principle of neurofeedback protocols is the supervised training of mental imagery, which aims to treat depression by enhancing cognitive flexibility and the ability to engage in positive mental simulations [32,33]. However, some individuals are resistant to various depression treatments, which is why it’s essential to explore new options. One promising alternative is fMRI-neurofeedback. There is also a new type of real-time fMRI neurofeedback called functional connectivity neurofeedback, in which patients are trained to modulate the functional connections between items they choose [34]. This new approach has also been studied in other psychiatric disorders and has yielded promising outcomes [35]. In recent years, particularly over the last decade, significant efforts have been made to utilize fMRI-neurofeedback to positively impact depression. However, the effectiveness of this new treatment approach for patients with depression is still unknown, and it seems necessary to consolidate the findings in this regard. Although systematic review articles in this field have been published in recent years [36-38], their primary focus has not solely been on fMRI-neurofeedback; they also consider electroencephalography (EEG) alongside fMRI. This mixture of approaches has yielded mixed results, making it difficult to draw a definitive conclusion about the effectiveness of fMRI-neurofeedback in treating depression. Additionally, previous reviews have not considered many trials published in recent years, which can make conclusions in this area challenging. Therefore, in this study, we aimed to conduct a systematic review of the therapeutic applications of fMRI-neurofeedback in patients with depression, addressing existing gaps in the literature. It is hypothesized that fMRI neurofeedback interventions will lead to significant reductions in depressive symptoms. Moreover, it is expected that participants receiving neurofeedback will demonstrate enhanced functional connectivity within neural networks associated with emotional regulation. Lastly, we hypothesize that the magnitude of symptom improvement will correlate positively with increased Blood Oxygenation Level Dependent (BOLD) activity in targeted regions during neurofeedback training, suggesting a dose-response relationship between neural engagement and clinical outcomes.

METHODS

The purpose of this study was to review human studies conducted on depressive patients, regardless of sex, age or trial design, reporting the influences of fMRI-neurofeedback therapy on depression-related signs. Indeed, in light of the increasing research on the use of fMRI-neurofeedback as a therapeutic option for depression in recent years, our objective is to (I) synthesize and contrast existing results, (II) assess the methodological rigor of these investigations, and (III) offer recommendations for future studies that could advance this area of research. The whole process of this study was performed according to PRISMA guideline.

Search Strategy

A systematic search was done on PubMed, Web of Science, ProQuest, Central Cochrane library and Google Scholar up to January 2025 through suitable keywords. The keywords included the intervention of interest (neurofeedback, biofeedback, fMRI-neurofeedback, and fMRI-neurofeedback training) and the diagnosis of interest (depression, depressive patient, and major depressive disorder [MDD]. After removing duplications, two investigators attempted to screen the retrieved articles independently given the titles and abstracts. Relevant English trials (open-label, between- or within-randomized controlled trials [RCTs], and case series) were included in this research, regardless of publication date/source (i.e., books, conference articles or journal articles) and country of origin. Review, editorial and letter publications were excluded, but their references were screened for further relevant papers. Also, EEG-neurofeedback studies were not the focus of our study and were excluded from the results.

Data Extraction

Based on PRISMA processes, two authors conducted data extraction for every eligible publication including publication date, names of authors, trial design, count of participants, sex, age, diagnostic criteria, type of feedback, training protocol, outcome measures, assessment times, and main findings. Conflicts among the authors conducting the data extraction were addressed through discussion.

Quality Assessment

In order to enhance result analysis accuracy and avoid review bias, two separate evaluators individually evaluated the credibility and excellence of chosen publications by applying the Consolidated Standards of Reporting Trials (CONSORT) checklist for RCTs [39] and Joanna Briggs Institute (JBI) checklist for quasi-experimental studies [40]. They scrutinized elements concerning methodology, outcomes, and discussion. Each publication was assigned a cumulative quality rating utilized during data compilation and review findings interpretation, as elaborated in Table 1 (risk of bias column). CONSORT is widely recognized for its rigorous standards in promoting transparency and completeness in reporting randomized controlled trials. Its inclusion ensures that key components of study design, conduct, and analysis are addressed, thereby bolstering the credibility of findings related to the efficacy of fMRI-based neurofeedback therapy. Moreover, JBI, with its comprehensive evidence-based approach, assists in synthesizing qualitative and quantitative data, making it particularly useful for appraising the heterogeneity often found in neurofeedback research. By employing JBI, the review captures a nuanced understanding of varying methodologies and outcomes across studies.

Table 1.

Summary of key characteristics of studies evaluating fMRI-neurofeedback interventions for MDD, including authors, sample size, and outcomes measured

Authors (year; country)	Training protocol (feedback)	Design				Diagnosis (diagnostic criteria)	N Intervention (mean age)	N Control (mean age)	Sex	Risk of bias
		Study	Randomization	Blinding	Control
Zahn et al. (2019; Brazil) [43]	Increasing the connectivity and correlation between the right superior ATL and ACC regions retrieval of guilt-related events (visual)	Between-group RCT	Yes	Double-blind	Sham	MDD (DSM-IV)	14 (45.1 ± 13.7)	14 (45.3 ± 17.5)	Male & female	Low
Mennen et al. (2021; USA) [44]	Closed-loop NFT based on internal attentional states (visual)	Between-group RCT	No	No blinding	Healthy control	MDD (DSM-5)	15 (27.3)	12 (25.4)	Male & female	High
Takamura et al. (2020; Japan) [45]	Real-time fMRI NFT based on DLPFC activity (visual)	Open-label	No	No blinding	No	MDD (DSM-IV)	6 (40.7)	-	Male & female	High
Pereira et al. (2022; Chile) [46]	Matching the neural activity with the happiness emotional brain states of the healthy subjects (visual)	Open-label	No	No blinding	No	Mild to moderate depression (BDI-II)	7 (18−31)	-	Female	High
Young et al. (2017; USA) [47]	Amygdala real-time fMRI neurofeedback (visual)	Between-group RCT	Yes	Double blind	Parietal real-time fMRI neurofeedback	MDD (DSM-IV-TR)	18 (32 ± 12)	16 (31 ± 9)	Male & female	Low
Quevedo et al. (2020; USA) [48]	Upregulating the bilateral amygdala and hippocampus activity via the recall of positive autobiographical memory (visual)	Between-group RCT	No	No blinding	Healthy control	Depression (K-SADS-PL and CDRS)	34 (16.11 ± 1.25)	19 (16.35 ± 1.23)	Male & female	High
Zotev et al. (2016; USA) [49]	Upregulating the left amygdala activity during a happy emotion induction task (visual)	Between-group RCT	Yes	Single blind	Parietal real-time fMRI neurofeedback	MDD (DSM-IV-TR)	13 (41 ± 9)	11 (34 ± 8)	Male & female	Moderate
Taylor et al. (2022; Japan) [50]	Upregulating the connectivity between the left precuneus/PCC and the left DLPFC/middle frontal gyrus (visual)	Open-label	No	No blinding	No	Depression (BDI-II)	10 (23 ± 1.67)	-	Male & female	High
Zotev et al. (2020; USA) [51]	Real-time fMRI-EEG neurofeedback to simultaneously upregulate the left amygdala activity, left rostral ACC activity, and frontal EEG asymmetries in the alpha and high-beta bands (visual)	Between-group RCT	Yes	Single blind	Sham	MDD (DSM-IV)	16 (not specified)	8 (not specified)	Male & female	Moderate
Keller et al. (2021; Germany) [52]	Upregulating either the left or the right VLPFC during cognitive reappraisal of negative images (visual)	Within-group RCT	Yes	Double blind	Healthy control	MDD DSM-IV-TR)	39 (35.2 ± 2.2)	37 (32.3 ± 2.1)	Male & female	Low
Hamilton et al. (2016; USA) [53]	Down-regulating the salience network responses to negative material (visual)	Between-group RCT	Yes	Double blind	Sham	MDD (DSM-IV)	10 (31.2 ± 2.9)	10 (34.5 ± 3.7)	Male & female	Low
Tsuchiyagaito et al. (2021; USA) [54]	Upregulating the left amygdala activity (visual)	Open-label	No	No blinding	No	MDD (DSM-IV)	29 (18−55)	-	Male & female	High
Young et al. (2014; USA) [55]	Upregulating the left amygdala activity (visual)	Between-group RCT	No	Double blind	Parietal real-time fMRI neurofeedback	MDD (DSM-IV-TR)	14 (38 ± 10)	7 (36 ± 9)	Male & female	Moderate
Mehler et al. (2018; UK) [56]	Upregulating the brain’s emotional areas (visual)	Between-group RCT	Yes	Single blind	Upregulation of a control region activated by visual scenes	Moderate to severe depression (MINI)	16 (47.19 ± 12.5)	16 (46.94 ± 12.74)	Male & female	Low
Linden et al. (2012; UK) [57]	Upregulating the brain areas involved in the generation of positive emotions (visual)	Between-group RCT	No	No blinding	Standard treatment	MDD (DSM-IV)	8 (48.38)	8 (48.5)	Male & female	High
Mel’nikov et al. (2023; Russia) [58]	Real-time fMRI neurofeedback targeting the left medial prefrontal cortex (visual)	Between-group RCT	No	No blinding	CBT	Unipolar non-seasonal + depressive disorder without psychotic features (ICD-10)	8 (30.8 ± 10.3)	8 (30 ± 8.5)	Female	High
Yuan et al. (2014; USA) [59]	Upregulating amygdala activity during recall of positive autobiographical memories (visual)	Between-group RCT	No	Double blind	Healthy control	MDD (DSM-IV-TR)	14 (38 ± 10)	13 (35 ± 8)	Male & female	High
Young et al. (2017; USA) [60]	Upregulating the left amygdala activity (visual)	Between-group RCT	Yes	Double blinding	Parietal real-time fMRI neurofeedback	MDD (DSM-IV-TR)	18 (18−55)	15 (18−55)	Male & female	Low
Jaeckle et al. (2023; UK) [61]	Real-time fMRI neurofeedback targeting right superior ATL-posterior subgenual cortex (visual)	Between-group RCT	Yes	Single blind	Self-guided psychological strategies	MDD + anxiety (DSM-IV)	19 (36.74 ± 11.04)	16 (36.73 ± 9.74)	Male & female	Low
Young et al. (2017; USA) [62]	Upregulating amygdala activity (visual)	Between-group RCT	Yes	Double blind	Parietal real-time fMRI neurofeedback	MDD (DSM-IV-TR)	18 (18−55)	16 (18−55)	Male & female	Low
Compère et al. (2023; USA) [63]	Upregulating amygdala reactivity in response to a cognitive task (visual)	Between-group RCT	Yes	Double blind	Parietal real-time fMRI neurofeedback	MDD (DSM-5)	16 (30.66 ± 9.55)	22 (30.66 ± 9.55)	Male & female	Low
Takamura et al. (2023; Japan) [64]	Promoting negative functional connectivity between the DLPFC and the PCC (visual)	Open-label	No	No blinding	No	MDD (DSM-IV)	5 (38.8 ± 6.3)	-	Male & female	High
Tsuchiyagaito et al. (2023; USA) [65]	Upregulating connectivity between the precuneus/PCC and rTPJ (visual)	Between-group RCT	Yes	Double blind	Sham	MDD (DSM-5)	20 (33.6 ± 10.6)	19 (33.42 ± 11.49)	Male & female	Low
Compère et al. (2023; USA) [66]	Upregulating amygdala activity (visual)	Between-group RCT	Yes	Double blind	Parietal real- time fMRI neurofeedback	MDD (DSM-5)	16 (31.31 ± 6.78)	19 (31.47 ± 11.62)	Male & female	Low
Compère et al. (2024; USA) [67]	Upregulating amygdala activity (visual and haptic)	Open-label	No	No blinding	No	MDD (DSM-5)	15 (18−55)	-	Male & female	High
Misaki et al. (2024; USA) [68]	Upregulating functional connectivity between PCC and rTPJ (visual)	Between-group RCT	Yes	Double blind	Sham	MDD (DSM-5)	16 (33.6 ± 10.6)	18 (33.4 ± 11.5)	Male & female	Low

ATL, anterior temporal lobe; ACC, anterior cingulate cortex; NFT, neurofeedback training; DLPFC, dorsolateral prefrontal cortex; RCT, randomized controlled trial; MDD, major depressive disorder; DSM, Diagnostic and Statistical Manual of Mental Disorders; BDI-II, Beck Depression Inventory-II; K-SADS-PL, Schedule for Affective Disorders and Schizophrenia for School-Aged Children, Present and Lifetime Version; CDRS, Children’s Depression Rating Scale; MINI, Mini International Neuropsychiatric Interview; ICD, International Classification of Diseases; VLPFC, ventrolateral prefrontal cortex; PCC, posterior cingulate cortex; rTPJ, right temporoparietal junction; CBT, cognitive behavioral therapy; EEG, electroencephalography; -, not available.

In addition, to improve the standards of reporting in neurofeedback research, the newly released CRED-NF checklist—entitled “Consensus on the Reporting and Experimental Design of Neurofeedback Studies”—proposes both “essential” and “recommended” items related to design and reporting [41]. These include aspects such as pre-registration of experiments, the inclusion of control groups and their respective measures, specifications for feedback, descriptions of outcomes, and protocols for data storage and publication. CRED-NF, tailored specifically for neurofeedback studies, establishes criteria that address the unique challenges associated with this therapeutic modality. Its use allows for a focused evaluation of essential elements such as treatment fidelity and participant engagement, which are critical for accurately assessing outcomes in neurofeedback interventions. By clarifying how these tools influence the overall interpretation of results—such as mitigating biases, enhancing study transparency, and ensuring methodological rigor—the methodology section can provide a more robust framework for understanding the implications of the findings. Two of the authors assessed the studies featured in this review using the 23 criteria outlined in the CRED-NF checklist. In this step, again, any disagreements regarding their assessments were addressed through discussion.

RESULTS

Figure 1 shows the PRISMA flowchart for the adopted process of selecting eligible studies. As shown, the initial systematic search in the five mentioned databases resulted in 649 original articles. After duplication removal, 433 unique publications were screened for checking the eligibility criteria. This process resulted in identifying 26 eligible studies using fMRI-neurofeedback for therapeutic purposes in depressive patients. Six of these studies were open-label trials and the remaining 20 were between- or within-group RCTs. Totally, 414 male/female adults with depression received active fMRI-neurofeedback in these trials with different training protocols. Most studies have used DSM criteria to diagnose depression in patients. Moreover, in most controlled trials, parietal real-time fMRI-neurofeedback have been utilized for sham controlled group. In general, eight studies used passive sham, four used healthy controls for the control group, and eight used active sham. During fMRI neurofeedback, patients might feel claustrophobic because of the confined space created by the machinery [42]. This consideration is especially important for those suffering from MDD, as the symptoms associated with this condition can encompass a lack of interest, increased restlessness, sleep disturbances, and feelings of exhaustion or low energy. Nevertheless, it’s worth mentioning that the rates of individuals discontinuing the process were quite low, and no significant adverse effects have been observed in the original studies.

Fig. 1.

PRISMA flowchart for selecting relevant studies on fMRI-neurofeedback in depressive patients.

In most studies, neurofeedback training protocol has been used to increase amygdala activity with the aim of improving symptoms in depressed patients. Protocols for neurofeedback using fMRI typically involve tasks designed for emotional self-regulation. These protocols train individuals to either increase or decrease the BOLD signal in brain regions associated with emotion processing, including the amygdala, superior temporal gyrus, insular cortex, dorsal anterior cingulate cortex, precentral gyrus, prefrontal cortex, and middle temporal gyrus. In the first proof-of-concept study that was non-randomized and did not employ blinding, Linden et al. (2012) explored the effects of fMRI-neurofeedback in individuals with depression by comparing two groups, each consisting of eight patients with MDD who were undergoing medication. The control group participated in standard mental imagery exercises outside the scanner, while the experimental group underwent four sessions of neurofeedback training. During these sessions, participants applied similar mental techniques to regulate the activity in brain regions that reacted to emotional visual stimuli. The neurofeedback group learned to enhance the BOLD responses in the targeted regions and demonstrated a notably greater improvement in depressive symptoms compared to the control group, which did not show such benefits.

All studies used visual feedback in their training protocol. Most of the studies used a proof-of-concept design and therefore conducted only one treatment session. Most studies have used well-known psychometric tools such as Beck Depression Inventory-II (BDI-II), Montgomery-Åsberg depression Rating Scale (MADRS) and Hamilton Rating Scale for Depression (HRSD) to measure the effect of treatment before and after treatment. Most studies have shown improvement in depression symptoms after treatment and have proven the effectiveness of fMRI-neurofeedback to improve patients. Tables 1 and 2 summarize the details of the included studies [43-68].

Table 2.

A comparison of treatment outcomes between neurofeedback and standard control interventions in individuals with depression, highlighting effect sizes and statistical significance

Authors (year)	Number of training sessions	Neurobiological measures	Behavioral measures	Assessment times	Neurobiological outcomes	Behavioral outcomes	Brain-behavior associations
Zahn et al. (2019) [43]	One	Correlation between ATL and SCC BOLD signal for guilt relative to indignation correlations	BDI, IGQ-Self-hate, Rosensberg Self-esteem, PANAS	Before and after training	Increased levels of ATL-SCC correlations for guilt versus anger (p = 0.03)	Increased self-esteem after active NFT compared to sham (Cohen’s d = 0.98)	In the active group, patients who displayed low connectivity between ATL and SCC when processing feelings of guilt compared to indignation were more likely to enhance this connectivity through NFT training than those who already had high connectivity levels before the training began.
Mennen et al. (2021) [44]	Three	-	MADRS	Before and after training, one-month and three-month after training	-	Depression symptoms were reduced significantly post-training (p = 0.001), and this improvement was maintained one month (p = 0.006) and three months after the training (p = 0.002). Also, this approach led to the improvement of negative attentional biases in patients (p = 0.026)	-
Takamura et al. (2020) [45]	Five	-	BDI-II, HRSD, RRS	Before and after training	-	Significant improvement in RRS-rumination (p = 0.043, Cohen’s d = 1.10), HRSD (p < 0.001, Cohen’s d = 5.63) and BDI-II (p = 0.026, Cohen’s d = 1.41) after NFT	Significant correlation observed between the enhancement of DLPFC regulation and the tendency to ruminate (rho = −0.97, p = 0.006)
Pereira et al. (2022) [46]	Four	-	BDI-II, HRSD	Before and after training, and 10 days after training	-	Clinical improvement after the intervention in both BDI-II (p = 0.031) and HRSD (p = 0.016) scales with no significant improvement 10 days after the intervention	-
Young et al. (2017) [47]	Two	Task-based and resting state amygdala connectivity	MADRS	Before and 5−7 days after the intervention	After the training sessions, connectivity within the amygdala during the recall of positive autobiographical memories showed enhancement across multiple areas of the frontal and limbic networks (p < 0.05). Similarly, during resting states, increased amygdala connectivity was observed within the fronto-temporal-limbic network following the training (p < 0.05).	At the one-week follow-up assessment, the group that underwent amygdala-focused training demonstrated significantly reduced MADRS scores compared to the group that received parietal-targeted real-time fMRI neurofeedback (p < 0.05, Fisher’s value = 3.49)	Significant correlation observed between amygdala-left precuneus activity and clinical change (r = −0.46, p = 0.002).
Quevedo et al. (2020) [48]	One	Task-based amygdala connectivity	BDI-II	Before and after training	Patients exhibited increased connectivity between the right amygdala and right frontocortical areas, alongside reduced connectivity between the left amygdala and right frontocortical regions during neurofeedback sessions (p < 0.01).	Significant improvement in depression symptoms and rumination after NFT (p < 0.001)	The connection between the rightamygdala and areas of the frontocortex showed a positive relationship with changes in rumination and depression (r = 0.498, p < 0.01), while the connection between the left amygdala and frontocortical regions exhibited a negative association with alterations in depression levels (r = −0.287, p < 0.05).
Zotev et al. (2016) [49]	One	Amygdala BOLD activity	POMS, VAS	Before and after training	Increased left amygdala BOLD activity in the intervention group for the happy memories condition (p < 0.01)	Significant improvements in the emotional states of the patients after NFT (p = 0.04)	In the active group, average levels of physical activity showed a strong positive correlation with increases in VAS happiness scores (r = 0.65, p < 0.017) and a significant negative correlation with decreases in POMS tension scores (r = −0.62, p < 0.024).
Taylor et al. (2022) [50]	Four	Functional connectivity	BDI-II, RRS	Before and after training, and one- to two-month after training	Significant increased DLPFC-PCC connectivity (p = 0.03)	Rumination and more general depressive symptoms were significantly reduced after NFT (p < 0.05)	A significant positive correlation observed between changes in DLPFC-PCC connectivity and changes in BDI scores (r = 0.78, p < 0.001)
Zotev et al. (2020) [51]	One	Amygdala BOLD activity	POMS, VAS	Before and after training	Significant enhancement of left amygdala BOLD activity and increased connectivity between left amygdala and left rostral anterior cingulate cortex (p < 0.05)	Significant improvements in the emotional and mood states of the patients after NFT (p < 0.05)	No significant association was observed.
Keller et al. (2021) [52]	One	Brain responses and functional connectivity	BDI-II, HRSD, HFERST	Before and after training, and four-week after training	Increased left VLPFC BOLD activity after NFT (p < 0.05)	Significant improvements in depressive symptoms (p < 0.05). At the 4-week follow-up, 75% of individuals with depression reported effectively applying the strategies they had learned to their daily lives, while 55% experienced a clinically significant reduction in their symptoms.	The engagement in the left target area correlated with a greater application of cognitive reappraisal techniques (r = 0.48, p = 0.002)
Hamilton et al. (2016) [53]	Three	Brain responses and functional connectivity	BDI-II, PANAS	Before and after training	A reduction in brain’s salience network-node response to negative stimuli after NFT (p = 0.06, Cohen’s d = 0.70)	Significant reduction in emotional response to negative scenes and to negative self-descriptive adjectives (p < 0.05, Cohen’s d = 0.78)	No significant association was observed.
Tsuchiyagaito et al. (2021) [54]	Two	Kynurenine pathway metabolism	MADRS	Before and after training, and two-week after training	Significant increase in neuroprotective kynurenines (p = 0.03, Cohen’s d = 0.38)	MADRS score was significantly decreased after NFT (p = 0.009, Cohen’s d = 0.56)	No significant association was observed.
Young et al. (2014) [55]	One	Amygdala BOLD activity	TAS-20, EQ, SHAPS	Before and after training	Patients in the experimental group demonstrated an increase in amygdala activation while recalling positive autobiographical memories (p < 0.05). In the transfer phase, patients in the experimental group exhibited heightened activity in the left superior temporal gyrus, temporal polar cortex, and right thalamus compared to the control group (p < 0.05).	Notablereductions in anxiety scores and increases in happiness scores were observed in the experimental group compared to the control group after NFT (p < 0.05)	-
Mehler et al. (2018) [56]	Five	BOLD activity	HRSD	Before and after training, and 18-week after training	Both groups upregulated target regions (subcortical and limbic regions versus higher visual regions like PPA) to a similar extent.	No significant difference in mean HRSD scores was observed between the groups (p = 0.84). However, depressive symptoms improved by 43% across both groups, and 38% of patients achieved remission. These positive outcomes were sustained through the follow-up period at week 18.	No significant association was observed.
Linden et al. (2012) [57]	Four	BOLD activity	HRSD, PANAS, POMS	Before and after training	Significantly higher activation in the left extrastriate visual cortex and bilateral ventral striatum after NFT (p < 0.001)	Significant clinical improvement on the HRSD score in the NFT group after intervention (p = 0.009, Cohen’s d = 1.5).	Not reported
Mel’nikov et al. (2023) [58]	Eight	BOLD activity	MADRS, BDI-II, ZSRDS	Before and after training	Higher but non-significant activation in the prefrontal cortex after NFT	Non-significant improvement in depressive symptoms	Not reported
Yuan et al. (2014) [59]	One	BOLD activity	HRSD	Before and after training	Active NFT led to enhanced connectivity in the right parahippocampal gyrus, the right superior temporal gyrus, and both sides of the middle frontal gyrus compared to the control group. Conversely, the control group showed greater connectivity between the amygdala and the right lingual gyrus. In relation to the pgACC seed, the NFT group exhibited increased connectivity with the left superior temporal cortex, the left superior frontal gyrus, and the right superior temporal gyrus (p < 0.005).	Although symptoms of depression and anxiety improved in the NFT group (p = 0.01), no significant differences were observed between the groups after NFT (p = 0.86).	In the active group, more pronounced reductions in depression severity are linked to greater increases in amygdala connectivity (p > 0.05). Additionally, there is a significant positive correlation between the changes in connectivity and the number of days following neurofeedback treatment (p < 0.05).
Young et al. (2017) [60]	Two	BOLD activity	HRSD, MADRS, BDI-II	Before and after training	Increased activity in the amygdala in the NFT group compared to control after intervention (p < 0.001)	Significant improvement in the HRSD (p = 0.01, Cohen’s d = 0.92), BDI-II (p = 0.03, Cohen’s d = 0.74) and MADRS (p < 0.001, Cohen’s d = 1.17) scales in the NFT group after intervention	MADRS scores showed a strong association with the residual activity of the amygdala during the final transfer run (b = −15.5, p = 0.004, adjusted R2 = 0.21).
Jaeckle et al. (2023) [61]	Three	BOLD activity and functional connectivity	BDI-II	Before and after training	NFT resulted in a training-induced reduction in right superior ATL-posterior subgenual cortex connectivity for self-blame versus other-blame (p < 0.01)	Both NFT and control groups showed a 46% reduction on the BDI-II after intervention, with no group differences.	Not reported
Young et al. (2017) [62]	One	BOLD activity	MADRS	Before and after training	In the backward masking task, the experimental group exhibited heightened amygdala activity when observing masked happy faces (p = 0.001, Cohen’s d = 1.13), whereas there was a reduction in response to masked sad faces compared to the control group (p = 0.002, Cohen’s d = 0.52) after undergoing NFT.	After undergoing NFT, the experimental group displayed notably reduced MADRS scores in comparison to the control group (p < 0.001, Cohen’s d = 1.17)	The variation in amygdala activity showed a notable correlation with both the changes observed in MADRS scores (r = −0.41, p = 0.002) and the effectiveness of neurofeedback (r = −0.39, p = 0.003).
Compère et al. (2023) [63]	One	BOLD activity	BDI-II	Before and after training, and one week after training	NFT group showed significantly increased amygdala activity compared to the control group after intervention (p < 0.001, Cohen’s d = 0.79)	Although BDI-II scores were reduced in the NFT group (p < 0.001, Cohen’s d = 0.38), no significant differences were observed between the groups after NFT (p = 0.09).	Following the training, the activity of the amygdala during the counting task showed a reduction, which was linked to a decrease in depression scores (r = 0.46, p = 0.01).
Takamura et al. (2023) [64]	Four	Functional connectivity	HRSD, BDI-II, RRS	Before and after training, and four-week after training	There were no significant changes in DLPFC-PCC connectivity after NFT (p = 0.33)	Significant reduction in all scales after intervention and at follow ups (p < 0.05)	No significant association was observed.
Tsuchiyagaito et al. (2023) [65]	One	Functional connectivity	HRSD, MADRS, RRS	Before and one week after training	Significant reduced RSC-rTPJ connectivity in active conditions compared to the sham (p = 0.02, Cohen’s d = 0.87)	Significant lower RRS score after NFT and at follow-up in the active group compared to the sham (p < 0.001, Cohen’s d = 1.03)	A significant positive association between RSC-rTPJ connectivity and RRS score in the active group compared to sham (adjusted R2 = 0.36, b = 1.14, p = 0.04)
Compère et al. (2023) [66]	One	-	BDI-II	Before and after training, six-month and one year after training	-	Significant improvement in depressive symptoms (p = 0.02, Cohen’s d = 0.85) and remission rates at six-month and one year follow ups in the NFT group compared to the control	No significant association was observed.
Compère et al. (2024) [67]	Five	BOLD activity	BDI-II	Before and after training, 1-, 2-, 4-, 6-, 8-, 10-, and 12-weeks follow-up	Significant increased amygdala activity after NFT (p = 0.04)	Significant improvement in depressive symptomatology after intervention and at 12-week follow up (p < 0.003, Cohen’s d = 0.91). Just two NFT sessions are sufficient to help patients increase activity in their amygdala.	Not reported
Misaki et al. (2024) [68]	Three	Functional connectivity	RRS	Before and after training	No significant difference was observed in the PCC-rTPJ connectivity in the active group compared to sham (p = 0.80)	Although RRS score was reduced in the active group (p < 0.001, Cohen’s d = 1.52), no significant differences were observed between the groups after NFT (p = 0.80).	No significant association was observed.

ATL, anterior temporal lobe; SCC, subcallosal cortex; BOLD, Blood Oxygen Level Dependent; BDI, Beck Depression Inventory; IGQ, Internalized Guilt Questionnaire; PANAS, Positive and Negative Affect Schedule; MADRS, Montgomery-Åsberg Depression Rating Scale; HRSD, Hamilton Rating Scale for Depression; RRS, Ruminative Response Scale; VAS, Visual Analog Scale; POMS, Profile of Mood States; VLPFC, ventrolateral prefrontal cortex; PCC, posterior cingulate cortex; RSC, retrosplenial cortex; rTPJ, right temporoparietal junction; SHAPS, Snaith-Hamilton Pleasure Scale; EQ, empathy quotient; ZSRDS, Zung Self-Rating Depression Scale; -, not available.

About half of the studies have moderate to high risk of bias, which indicates methodological problems in conducting these types of trials and improving and confirming their results in future studies. Among the included trials, 12 studies were double-blind, 4 studies were single-blind, and 10 studies were un-blinded. In addition, 14 studies used randomization techniques in conducting their trials and 12 studies had no randomization. As shown in Figure 2, all studies met at least half of the CRED-NF checklist items and scored more than 50 percent. Regarding the items on this checklist, it should be noted that most studies did not pre-register their trials. Upon examining the studies for their reported outcome measures, we discovered that although a number of studies clearly outlined what they considered “control” or “success” measures based on BOLD signal, some studies did not provide such explicit definitions. Additionally, we observed that only half of the studies specified the primary clinical outcome measures. Differentiating between primary and secondary outcome measures is regarded as a crucial quality standard in clinical research, as it plays a key role in assessing the effectiveness of an intervention and managing error rates.

Fig. 2.

CRED-NF scores for all included studies on fMRI-neurofeedback in depressive patients.

CRED-NF, Consensus on the Reporting and Experimental Design of Neurofeedback Studies.

As shown in Figure 3, more than 90% of original researches on the application of fMRI-neurofeedback for depression were administered in developed countries. This reflects the fact that this treatment approach cannot be used as a universal treatment method worldwide due to the costs and availability of fMRI devices. Therefore, one of the important goals of studies in this area should be to optimize the fMRI neurofeedback treatment process to reduce costs and increase benefits so that its applicability becomes more widespread.

Fig. 3.

Number of original studies per country on fMRI-neurofeedback in depressive patients.

Neurobiological Outcomes

The analysis of neurobiological outcomes from the included studies reveals a consistent trend of enhanced brain activation and connectivity in areas related to emotion regulation and processing. Specifically, interventions targeting the amygdala and prefrontal cortex predominantly reported increased BOLD activity, pointing towards more robust neural responses linked to emotional stimuli. For instance, Zahn et al. (2019) [43] demonstrated increased correlations in BOLD signals between the anterior temporal lobe (ATL) and subgenual anterior cingulate cortex (SACC) when processing guilt-related emotions, signifying improved emotional connectivity (p = 0.03). Young et al. (2017) [47] reported that amygdala connectivity during positive recollections was enhanced across multiple networks following neurofeedback (p < 0.05). Furthermore, Zotev et al. (2020) [51] indicated a significant increase in left amygdala activity in patients recalling happy memories (p < 0.05). Overall, these findings suggest that fMRI-based neurofeedback therapy positively influences neurobiological function in patients with depression.

Behavioral Outcomes

Behavioral improvements across studies primarily focus on reductions in depression symptoms and increased emotional regulation. A significant number of studies report favorable outcomes after neurofeedback interventions, with reductions in scores on validated depression measures such as the BDI and MADRS. For example, Mennen et al. (2021) [44] found a substantial decrease in depression symptoms post-training (p = 0.001), and improvements were maintained over time (p = 0.002). Similar results were noted by Keller et al. (2021) [52], where 75% of participants adapted strategies learned during treatment to their daily lives, resulting in clinically significant reductions in symptoms (p < 0.05). The overall evidence suggests that fMRI-based neurofeedback can be an effective adjunctive treatment for managing depressive symptoms.

Brain-behavior Associations

The relationship between neurobiological changes and behavioral outcomes was analyzed across several studies, revealing notable associations that highlight the impact of brain activity on emotional states. In particular, the enhancement of amygdala activity was correlated with reductions in depressive symptoms. For instance, Compère et al. (2023) [63] reported that increased amygdala activity was associated with a significant decrease in BDI-II scores (p < 0.001), indicating that greater engagement of this brain region can lead to improvements in mood. Additionally, Young et al. (2017) [62] found a strong correlation between variations in amygdala activity and changes in MADRS scores, emphasizing how neurofeedback can translate into tangible behavioral modifications (r = −0.41, p = 0.002). Such associations reinforce the premise that neurofeedback interventions not only alter brain function but also lead to meaningful changes in clinical outcomes for patients with depression.

DISCUSSION

In this initial systematic review of neurofeedback research utilizing fMRI in individuals with depression, we discovered that the studies involving fMRI show statistically significant and clinically relevant improvements in clinical measures within the groups. Indeed, the present review reveals promising evidence supporting its therapeutic applications. The studies analyzed consistently demonstrate that this innovative, non-invasive method significantly enhances brain activity related to emotional regulation, particularly within the amygdala and prefrontal cortex. Appreciably, neurofeedback training has shown to improve depressive symptoms across various trials, indicating its potential as an effective adjunct to traditional treatment options for MDD. This aligns with existing literature emphasizing the considerable neurobiological underpinning of depression, where dysregulation in brain circuits, particularly those involved in emotion regulation, plays a critical role in its manifestation [69-71].

It is believed that the overall clinical benefits observed after neurofeedback training can be attributed in part to several non-specific factors, such as the patient’s hopeful expectations, the positive reinforcement from feedback received, and the influence of regression to the mean, all of which likely play a significant role in the improvements seen within groups [72,73]. Neurofeedback training is a multifaceted intervention that allows for various designs of control conditions. These can range from passive control setups (e.g., ongoing standard care as opposed to standard care combined with neurofeedback training), which tend to offer minimal control over non-specific factors, to active control setups (e.g., standard care plus neurofeedback training compared with standard care plus targeted neurofeedback training), which are more effective in controlling for non-specific influences. As discussed extensively by experts within the neurofeedback field, selecting appropriate control conditions remains a challenge in neurofeedback research [74]. Control conditions are vital for assessing non-specific effects and for calculating meaningful effect sizes that facilitate comparisons with other therapeutic methods.

In assessing reporting practices, we found that many of the included studies failed to provide important details that are deemed essential or highly desirable, such as outlining a sampling strategy, specifying the primary outcome measure, reporting remission rates, or detailing feedback controllability. As a result, studies in this field generally carry a significant risk of bias, which limits the ability to make broad generalizations from the findings. Additionally, we observed that several published studies utilized partially overlapping samples, making it challenging to evaluate their quality consistently. We also suggest that the authors may have used flexible sampling termination rules without proper adjustments, raising the risk of Type-I errors. Most of these issues could be mitigated through thorough study preregistrations, which would include clear declarations of the main hypotheses, primary outcome measure, and planned analyses. Moreover, documenting various experimental factors—such as the comfort of the experiment room, staff attentiveness, and participant motivation or feelings of confidence and frustration—could provide insights into differences in self-regulation outcomes and clarify the clinical effects observed [75]. Concepts like self-efficacy, which are linked to depression-related psychopathology, may be influenced through self-regulation training [56]. Additionally, it was noted that only a few studies included in our review have made their imaging or clinical data publicly accessible. While this cautiousness may stem from concerns about data privacy, it is important to highlight that data anonymization technologies are widely accessible. Utilizing these tools would benefit the research community and align with recent initiatives within the neuroimaging and clinical medicine fields to address issues related to reproducibility and replicability [76].

Despite the encouraging findings, the review also brings to light several challenges faced by neurofeedback research in the context of depression treatment. It should be noted that employing suitable and rigorous techniques for data extraction and preprocessing is essential. In the context of fMRI-neurofeedback, controlling for confounding variables—such as real-time corrections for head movement, respiration, and cardiovascular artifacts—is often inadequately documented, despite their potential significant influence on the findings reported in studies [77]. Moreover, the reported clinical benefits of neurofeedback training appear to be considerable; however, the therapeutic impacts that target specific neural pathways may be relatively modest. Therefore, future trials will need to involve larger sample sizes to effectively examine the neurofeedback-specific impacts on depression. Moreover, extended follow-up periods are recommended, as research has shown that the clinical benefits resulting from neurofeedback interventions can persist and even improve for several months after the final session [56].

The identified risk of bias in the reviewed studies warrants careful consideration, as it can significantly influence the interpretation of results regarding the efficacy of fMRI-based neurofeedback therapy for depression. One key area of concern is the statistical methodologies employed in these studies, which can inadvertently lead to inflated or deflated effect sizes. For instance, the use of small sample sizes may result in overestimation of treatment effects due to the stochastic nature of smaller samples, where random variation can create misleading impressions of efficacy. Furthermore, the lack of appropriate control conditions or inadequate randomization techniques may allow for confounding variables to skew results, introducing bias that distorts the true effect of neurofeedback interventions. Additionally, publication bias—where studies with non-significant findings are less likely to be published—may contribute to an inflated perception of effectiveness within the literature. In assessing effect sizes, it is crucial to differentiate between clinically significant outcomes and statistically significant ones, as the latter may not always translate to meaningful improvements in patient well-being. Consequently, a comprehensive understanding of these biases and their potential implications is necessary to accurately evaluate the therapeutic value of fMRI-neurofeedback and inform future research directions that seek to mitigate such risks.

The implementation of fMRI-based neurofeedback therapy in clinical settings raises several ethical considerations that warrant careful attention. First and foremost, informed consent is paramount; participants must be made fully aware of the nature of the intervention, including its potential benefits, limitations, and any associated risks. Given the complexity of fMRI technology and the therapeutic process, it is essential that patients understand how their brain activity will be monitored and influenced. Additionally, confidentiality and data privacy must be rigorously maintained, particularly when handling sensitive neurobiological data that may reveal personal health information. The potential for psychological effects, both positive and negative, necessitates vigilant monitoring by trained professionals to ensure that participants do not experience undue harm or distress during the neurofeedback sessions. Furthermore, equitable access to fMRI neurofeedback treatments should be prioritized to avoid disparities in healthcare, particularly for marginalized populations who may benefit significantly from such innovative therapies. Finally, ongoing ethical reflection is crucial as the field evolves, ensuring that clinical practices adapt to emerging insights and uphold the principles of respect for persons, beneficence, and justice.

Challenges and Future Directions

Promoting the widespread use of fMRI-based neurofeedback therapy necessitates addressing inherent challenges in study design and practical application. Future research should prioritize the refinement of methodologies, including rigorous pre-registration of trials, clear differentiation between outcome measures, and the implementation of robust randomization and blinding techniques to enhance the reliability of findings. Exploring the development of cost-effective models for fMRI-based neurofeedback will also be crucial, especially in lower-income settings where access to such technology is limited. One promising approach is to invest in technological advancements that lower the cost and complexity of fMRI systems. This could include developing portable or more affordable neuroimaging devices that still provide reliable data for neurofeedback training. Additionally, improvements in software that can analyze and present fMRI data in real-time could enhance the user experience and broaden the application of neurofeedback across different settings. Another strategy involves exploring funding opportunities—both public and private—that support research and development in this field. Government grants, philanthropic contributions, and partnerships with industry leaders can provide the necessary financial resources to develop programs that make fMRI-based neurofeedback accessible in underserved communities. Alternative approaches, like integrating neurofeedback protocols into existing therapeutic frameworks, may also facilitate better scalability and acceptance among clinicians. Moreover, longitudinal assessments are necessary to gauge the long-term effects of neurofeedback therapy on depression, informing clinicians about potential sustained benefits or any emerging side effects over time. Expanded research focusing on the neurobiological markers that predict treatment response may enrich our understanding of which patient subgroups could most benefit from fMRI-neurofeedback interventions.

To ensure dependable clinical findings and facilitate comparisons among studies, we suggest that upcoming neurofeedback trials involving patients with depression adopt formal and standardized methods for diagnosing and assessing clinical changes using clinician-rated scales such as MADRS and HRSD as well as self-reported measures like BDI-II. In addition to evaluating overall score changes, we recommend documenting alterations in individual items to better understand shifts in specific symptoms or symptom clusters. It is also important to include the patients’ duration of illness and thorough descriptions of prior antidepressant treatments to evaluate factors like chronicity and treatment resistance, which could influence clinical outcomes [78]. Furthermore, we recognize that the origins of MDD may vary significantly among individuals [79], so a more detailed clinical and phenotypic characterization could aid in identifying patient subgroups that may particularly benefit from neurofeedback interventions. Moreover, we suggest that experiments and result reporting should adopt more stringent methodologies, particularly concerning measures such as respiration and pulse waves. Moreover, it is essential to assess and document the differences in artifacts among various groups, as biased noisy data can result in misleading conclusions.

Conclusion

This systematic review underscores the therapeutic potential of fMRI-based neurofeedback as a viable treatment strategy for individuals with depression. The evidence suggests that enhancements in neural regulation, particularly in the amygdala and prefrontal cortical regions, correspond to observable improvements in depressive symptoms. Neurofeedback is a sophisticated, non-invasive treatment designed to address the cognitive and emotional processes impacted in individuals with depression by utilizing mental imagery-based self-regulation techniques that focus on relevant brain areas or networks. Patients have demonstrated notable clinical improvements along with cognitive and neural modifications after undergoing neurofeedback training using fMRI-based methods. Furthermore, due to its non-invasive nature and relatively low risk of adverse effects, we believe neurofeedback is particularly promising as an additional treatment option for patients who continue to show symptoms despite having received standard care. However, to fully realize its promise, the field must confront existing methodological challenges and disparities in access. As we move forward, fostering an environment that encourages high-quality, expansive research will be vital to establishing fMRI-neurofeedback as a mainstream option in the management of depressive disorders, potentially transforming the landscape of treatment for millions affected by MDD worldwide.