fMRI Neurofeedback

Early real-time fMRI feedback

Real-time functional MRI neurofeedback grew out of brain-computer interface work built on functional MRI in the early 2000s. Nikolaus Weiskopf and colleagues at the University of Tübingen described a system that processed the blood oxygen level-dependent (BOLD) signal on-line and returned it to the participant with a delay of less than two seconds. In a 2003 methods paper, one volunteer tried to raise and lower the signal from the anterior cingulate cortex. The authors reported that the percent signal change increased across feedback sessions, which they read as a learning effect. The paper is a single-volunteer demonstration rather than a controlled study.

A 2004 paper from the same group set out the system's principles. It reported feedback within 1.3 seconds of data acquisition and showed differential feedback and self-regulation for the supplementary motor area and the parahippocampal place area. The authors described the method as a way to study the behavioral effects of local self-regulation in healthy and diseased subjects, which framed it as a research tool from the start.

Development since the early work

A 2012 review by Weiskopf summarizes the development of the method, beginning with the Tübingen work and then surveying later studies. According to the review, training has been used to modulate the local BOLD response in several brain areas, with behavioral effects reported in healthy and patient groups. The review describes the early patient results for pain, tinnitus, depression and perception in schizophrenia as encouraging and says they merit further controlled clinical studies.

A 2014 review by Stoeckel and colleagues describes real-time fMRI neurofeedback as an emerging technology in clinical neurotherapeutics. The authors propose research strategies for developing it as a therapeutic tool. Their paper is a proposal for future development rather than a report of established clinical use.

Neurofeedback as a wider field

The broader neurofeedback field predates the scanner work. In a 2017 review, Sitaram and colleagues describe neurofeedback as a psychophysiological procedure in which online feedback of neural activation is given for the purpose of self-regulation, and they describe it as a progenitor of brain-machine interfaces. The sources differ on when electroencephalography (EEG) neurofeedback began to be used for mental disorders. A 2015 review by Thibault and colleagues places the spread of promoted EEG treatment in the 1970s, while a 2016 review by the same group says the method came on the scene in the 1960s.

Real-time fMRI and related terms

Sources use several overlapping labels. Real-time fMRI (rtfMRI) describes the analysis of functional MRI data as it is acquired. Real-time fMRI neurofeedback describes the use of that analysis to show a person their own BOLD response so they can try to change it. Weiskopf's 2012 review notes that rtfMRI can also serve as a brain-computer interface with whole-brain coverage, and it treats neurofeedback as the application of that interface to the online BOLD response. The shorthand fMRI-NF appears in trial reports, including the 2018 randomized trial by Mehler and colleagues.

Some literature places fMRI neurofeedback within the biofeedback family. The search terms of a 2020 systematic review of neurofeedback in chronic pain included fMRI Biofeedback alongside Neurofeedback and EEG Biofeedback. The label neurofeedback covers both EEG and fMRI methods, so a study title can refer to either one, and each study should be read against the signal it used. The Biofeedback and EEG Biofeedback pages on this site cover those neighboring methods.

Target regions and control arms

Trials name the region that participants are asked to change, such as the amygdala in depression or the rostral anterior cingulate cortex in neuropathic pain. Control arms use a different region. In the amygdala depression trial by Young and colleagues, the control arm trained a parietal region that is not involved in emotional processing. In the postherpetic neuralgia trial by Guan and colleagues, the sham arm received feedback from the posterior cingulate cortex.

Naming the target says little about the mechanism. A 2016 individual-participant meta-analysis of real-time fMRI neurofeedback studies found that the anterior insula and basal ganglia were consistently active during regulation across nine different target regions. The authors argue that such activity may reflect the regulation process itself, not only the region that was targeted, so the term target region should be read with that point in mind.

How the signal is shown

In the 2003 methods paper, the signal from a region of interest was plotted as a time series over color-coded stripes that indicated whether the volunteer was to increase or decrease it. The analysis ran on-line with a delay under two seconds. The 2004 system extended this to feedback from two regions at once. Later trials keep the same basic arrangement: the person is shown a live display of activity from the chosen region and is asked to change it, and in the pain trial the instruction was to use an imagery task.

The trials reviewed differ in format. The 2017 depression trial by Young and colleagues used two sessions. The 2018 trial by Mehler and colleagues used five sessions as augmentation to medication. A 2023 trial by Compère and colleagues gave two fMRI sessions before ten sessions of cognitive behavioral therapy. Each design sets its own number and timing of sessions, so results from one format do not transfer automatically to another.

What the trials describe about participants

Training takes place inside an MRI scanner. The abstracts reviewed report measured brain signal change, symptom scales, memory tests and self-efficacy scores. They do not describe the experience of a session in detail. Mehler and colleagues report that clinical improvement was correlated with an increase in self-efficacy scores, and Young and colleagues report that the percent of positive specific memories recalled increased in the amygdala group.

The people enrolled in the depression trials were adults with depression who were either medicated or unmedicated at the time. Mehler's trial enrolled medicated patients with moderate to severe depression, and Young's trial enrolled unmedicated adults with depression. The trials reviewed were conducted as research studies, and the sources reviewed do not describe use outside research settings.

Conditions studied

Research has concentrated on chronic pain and depression, with earlier work on tinnitus and on perception in schizophrenia. Weiskopf's 2012 review lists pain, tinnitus, depression and perception in schizophrenia among the early patient results, and it does not present them as established uses. A 2020 review of neurofeedback in chronic pain covered EEG and fMRI trials together. A 2023 trial combined fMRI neurofeedback with cognitive behavioral therapy in depression.

Researchers have proposed the approach for depression on the basis of amygdala responses to positive memories. Young and colleagues describe blunted amygdala activity to positive stimuli in depression as the rationale for their trial. Compère and colleagues extend the idea by training amygdala responses before cognitive behavioral therapy. These are hypotheses tested in small trials, and the sources do not describe them as established clinical uses.

Expectations stated by trial authors

The sources reviewed do not record claims by commercial providers of fMRI neurofeedback. Where trial authors state expectations, they are attributed here to those authors. Mehler and colleagues write that the experience of successful self-regulation may be therapeutic. They present this as a suggestion drawn from their data, not as a finding, and they say that if fMRI neurofeedback is effective for depression, self-regulation training of higher visual areas may be an alternative.

The same authors argue that the symptom reductions in their trial exceed the regression to the mean and placebo effects reported for drug trials and other sham-controlled high-technology interventions. They also state that the interpretation of clinical improvements remains limited because the study had no sham-control group. The comparison is the authors' own and does not establish a treatment effect.

Chronic pain

The largest pain review is a 2020 systematic review and meta-analysis of EEG and fMRI neurofeedback in chronic pain. It included 21 studies. Pain reduction was reported by one high-quality randomized trial, by five of six low-quality randomized or non-randomized trials, and by 13 of 14 case series. The pooled effect size across randomized trials was medium (Cohen's d -0.76, 95% confidence interval -1.31 to -0.20), but the studies were highly heterogeneous (I2 73%). Because the review pooled EEG and fMRI studies, its pooled figure is not an estimate for fMRI alone, and the abstract does not separate the two.

A randomized trial in 16 patients with postherpetic neuralgia by Guan and colleagues tested fMRI neurofeedback of the rostral anterior cingulate cortex against sham feedback from the posterior cingulate cortex. Six of eight patients in the experimental group were able to increase and decrease the signal during training, and the control group did not show this modulation. Pain ratings changed differently between the groups, but the authors found no significant correlation between brain signal change and pain change. Two control participants were excluded for large head motion, which leaves a very small analyzed sample.

Depression

A 2017 randomized trial by Young and colleagues assigned 36 unmedicated adults with depression to two sessions of amygdala neurofeedback (19 participants) or to a parietal control region (17 participants). Twelve participants in the amygdala group, compared with two in the control group, had a decrease of more than 50% on the Montgomery-Åsberg Depression Rating Scale. Six participants in the amygdala group, compared with one in the control group, met remission criteria at study end, a number needed to treat of 4. The study was small, with follow-up one week after the last session.

A 2018 randomized trial by Mehler and colleagues compared emotion-region with visual control-region training in 43 medicated patients with moderate to severe depression. Among 32 completers there was no significant group difference at week 12 (B = -0.415, 95% CI -4.847 to 4.016, p = 0.848). Symptoms fell by 43% across both groups, and 38% remitted. The trial had no sham-control group, which the authors identify as limiting interpretation.

A 2023 randomized trial by Compère and colleagues gave 35 adults two fMRI sessions before ten sessions of cognitive behavioral therapy. The amygdala group had lower symptoms and higher remission at 6-month and 1-year follow-up, and the authors state the study was not sufficiently powered to detect small effects. The 2018 trial found no significant group difference, so the depression results are mixed.

Mechanisms and the state of the field

A 2016 individual-participant meta-analysis of 12 real-time fMRI neurofeedback studies, with 175 participants and 899 runs, found consistent activity in the anterior insula and basal ganglia during regulation. Sulzer and colleagues (2013) review progress in the field and set out open questions on study design, learning mechanisms and clinical application. Stoeckel and colleagues (2014) describe the same field as an emerging tool and propose strategies for clinical development.

Critiques and unanswered questions

Critics of the wider neurofeedback field describe the evidence in less favorable terms. Thibault and colleagues (2015) describe substantial methodological and conceptual caveats in empirical reports and call the evidence tenuous. A 2016 review by the same group calls EEG neurofeedback's clinical effectiveness controversial, and it says that imaging and more rigorous protocols may clarify mechanisms and clinical use. Searches of PubMed for real-time fMRI neurofeedback combined with pain, depression, randomized designs and reviews returned the trials and reviews cited here. The sources reviewed do not report costs, cost-effectiveness, or outcomes beyond about one year.

MRI environment and scanner safety

One professional safety guidance for MRI environments is the American College of Radiology's Guidance Document on MR Safe Practices, updated in 2019. Its abstract describes potential risks in the MR environment, including adverse events involving patients, equipment and personnel, and it recommends annual MR safety training for those responsible for MR safety. The abstract reviewed does not list specific contraindications, so screening criteria for implants, pregnancy and claustrophobia should be checked in the full guidance and in each research site's protocol.

The trial abstracts reviewed do not report adverse events in detail, and the sources reviewed do not describe harms from the feedback process itself. Absence from abstracts is not evidence of absence. The full trial reports should be checked for adverse events, dropouts and exclusion criteria before any conclusion is drawn about harms.

Reported effects and expectation

The 2020 pain review reports common side effects of neurofeedback in general as headache, nausea and drowsiness, and it says these generally did not lead to withdrawal from therapy except in one study. The review also reports improvements in depression, anxiety, fatigue and sleep in some studies. Its side-effect data cover EEG and fMRI studies together, and the abstract does not give frequencies or separate the two methods. The authors describe the evidence as largely low quality and call for double-blind sham-controlled trials.

Expectation and placebo responses are a central concern in this literature. Thibault and colleagues describe methodological and conceptual problems in neurofeedback research (2015). Mehler and colleagues note that clinical improvement without a sham-control group cannot be separated from expectation or regression to the mean. The sources reviewed do not show how well a control region protects against expectation effects.

Suitability and care boundaries

The trials reviewed enrolled adults with specific conditions in research imaging settings. Several designs combined training with conventional treatment. Mehler's trial was an augmentation to medication, and Compère's trial preceded cognitive behavioral therapy. The sources do not establish that fMRI neurofeedback is suitable as a stand-alone treatment, and they do not show how it compares with established treatments for the same conditions.

This page does not replace medical assessment or care. A person with depression, chronic pain or another condition should keep the care they are receiving and discuss any research study with the clinician responsible for that care before taking part. The sources reviewed describe fMRI neurofeedback as a research method, and none of them supports using it in place of established treatment.