- Email: [email protected]
Real-time fMRI brain-computer interface: A tool for personalized psychiatry?
Chapter 39
Real-time fMRI brain-computer interface: A tool for personalized psychiatry? David E.J. Linden Faculty of Health, Medicine and Life Sciences, School of Mental Health and Neuroscience, Maastricht University, Maastricht, The Netherlands; Division of Psychological Medicine and Clinical Neurosciences, School of Medicine, Cardiff University, Cardiff, United Kingdom
1 Introduction Since its invention 20 years ago, functional magnetic resonance imaging (fMRI) has become one of the most widely used, and probably the most publicly visible noninvasive technique to measure brain activation. fMRI has played a central role in the development of cognitive neuroscience, and several new fields, including social neuroscience, neuroeconomics, and genetic imaging, which may not have developed had it not been for the unique opportunities afforded by fMRI. The particular strengths of this technique are in its spatial resolution and fidelity, ability to reach deep subcortical structures, and whole-brain coverage, enabling the mapping of functionally connected networks, and the extraction of information from activation patterns that are distributed across different brain areas. In the psychiatric domain, fMRI has made major contributions to the understanding of psychopathology, and the effects of risk genes on cognitive and affective networks (Linden, 2012). In neurology, fMRI has become a central technique for mapping neuroplasticity; for example, in recovery from stroke (Seitz, 2010), and for presurgical mapping. However, fMRI has not yet fulfilled its translational potential, and there is, as of today, no established diagnostic, prognostic, or therapeutic use of this technique for any of the mental disorders. Factors that may have held back the clinical development of fMRI include its lack of molecular resolution (where radionuclide imaging has an advantage), semi-quantitative nature, dependence on control/baseline conditions (with the exception of resting-state fMRI), and difficulty controlling for physiological confounds (cardiovascular, respiratory, and eye and head movements). fMRI-based neurofeedback (fMRI-NF) has the potential to open up radically new paths to translation. During fMRI-NF training, participants receive feedback on their brain activity in real time, and are instructed to change this activation. This change is normally a simple up- or down-regulation, but could also entail changing the activation difference between two areas, their correlation, or the output of a multivariate pattern classification algorithm. The “hemodynamic” delay between neural activity and the vascular response that contributes to the fMRI signal, which is approximately 5 s, does not pose an obstacle when participants are informed of this delay (Weiskopf et al., 2004). The technical requirements include direct network access to the scanner hardware, software that processes the incoming data in real-time and computes incremental online statistics, and software for the conversion of the real-time activation data into sensory stimulation (e.g., visual, tactile) for the participant. The relevant workflows for such a “brain computer interface” (BCI) have been established on systems from all major MRI scanner manufacturers, and several custom-made and commercial software packages have been developed for online preprocessing and statistical analysis of fMRI data. Over the past 10years, fMRI-NF has been used to train healthy participants in the self-regulation of motor, sensory, language, and emotion areas (Weiskopf, 2011). In analogy to the effects of other noninvasive or invasive brain stimulation techniques, one should expect that such self-regulation also has consequences at the behavioral, cognitive, and affective level. However, clinical applications have been sparse, and had initially focused on chronic pain (Decharms et al., 2005). Compared with other neurofeedback techniques (with EEG or MEG), and with noninvasive physical stimulation techniques (tDCS and TMS), fMRI-NF has the advantage of higher spatial resolution and localization accuracy, and better access to deep limbic and subcortical structures. Compared with deep-brain stimulation, fMRI-NF has the advantage of noninvasiveness and spatial flexibility (although it is not intended to replace established DBS indications, e.g., in Parkinson’s disease [PD]). Finally, compared with all external stimulation techniques, neurofeedback has the attractive characteristic of enabling the patients themselves to control their brain activity, and thus contributing to their experience of self-efficacy, which is an important therapeutic factor in many neuropsychiatric disorders (Bandura, 1997). Personalized Psychiatry. https://doi.org/10.1016/B978-0-12-813176-3.00039-0 Copyright © 2020 Elsevier Inc. All rights reserved.
465
466
Personalized psychiatry
fMRI-NF has been boosted by developments in MR physics, allowing for fast acquisition of high quality data sets, and statistics, allowing for online calculation of univariate and multivariate statistics. It is now a mature experimental technique, and the exciting question for the next decade will be whether it can make a true translational contribution in medicine. The central challenges are identifying the symptoms and disorders that will respond to fMRI-NF, adapting the treatment protocols to the neural networks involved in each of them, evaluating the underlying neuroplastic mechanisms, and devising training strategies that enable sustainable long-term effects.
2 Principles and methods Attempts to train humans and animals to regulate their own brain activity, feeding back signals from noninvasive (electroencephalography [EEG]) or invasive recordings, go back to the 1960s (reviewed by Birbaumer & Cohen, 2007). The theoretical principles were largely derived from operant conditioning, whereby the participant learns the optimal strategy through the contingencies between their actions and a reward. In an animal, the reward would have to be a primary reward. For example, the desired neural response would be reinforced by the delivery of food or drink. In humans, the information about the achievement of a particular neural target (e.g., increasing the ratio between theta and alpha power of the EEG by 20%) could itself be the reward. Whether this reward reinforces a particular mental strategy that leads to the desired physiological response, or the physiological response itself is a futile question because the two cannot be separated in any meaningful way under the assumptions of psychophysical unity. It has been reported that humans can achieve reliable selfcontrol over parameters of their EEG, for example, the topography of slow cortical potentials, the alpha-theta ratio, or the ratio between sensorimotor rhythm and theta activity, through such operant conditioning. FMRI-based neurofeedback (fMRI-NF) can be conducted in a similar fashion, where participants are blind to the functional relevance of the targeted activation, and essentially aim to achieve self-regulation by trial and error (Weiskopf et al., 2004). However, fMRI-NF can also harness the considerable knowledge about the neural basis of particular mental and cognitive processes that the past 20 years of functional brain mapping have achieved, and introduce a cognitive component into the learning strategy. For example, in our study on upregulation of the supplementary motor area in Parkinson’s disease (Subramanian et al., 2011), patients were told about the motor planning functions of the target area, and informed that motor imagery might be one viable strategy to up-regulate it. Giving patients initial hints about potential strategies for the upregulation training, which they can then refine through an operant conditioning protocol with fMRI-NF, can considerably reduce scanning time, and provide patients with tangible results within one scanning session (Subramanian et al., 2011).
3 Clinical applications of fMRI-NF Clinical studies of EEG neurofeedback (EEF-NF) training have been conducted in epilepsy, attention deficit/hyperactivity disorder (ADHD), depression and anxiety, as well as other neurological and psychiatric disorders (Birbaumer & Cohen, 2007; Hammond, 2005). Although some of the initial results were promising, the only field where EEG-NF has entered clinical practice is ADHD (using a variety of target parameters, e.g., theta/beta ratio or slow cortical potentials). For ADHD, several published trials found positive effects on symptoms as measured by parent or teacher questionnaires or clinical assessments (Lofthouse et al., 2011). One limitation of most trials has been the lack of blinding of participants to treatment condition, and the general difficulties of setting up of double-blind controlled trials of NF training (Lansbergen et al., 2011), and a recent properly blinded trial did not find any superiority for the active EEG-neurofeedback intervention (Sch€ onenberg et al., 2017). Although these difficulties will apply similarly in fMRI-NF, there are good reasons to explore its use in clinical conditions, including and beyond those that may respond to EEG-NF. Because of the strong cognitive component, fMRINF may achieve its neurophysiological targets faster than EEG-NF, and thus enhance patient motivation and compliance. Furthermore, fMRI-NF can directly target specific brain areas or networks that have been implicated in neuropsychiatric disorders, either because they show a dysfunction that might cause the disorder, or because they may compensate for a primary dysfunction. In the first case, the aim of neurofeedback would be to restore the function to normal levels, and in the second, to promote the recruitment of the compensatory network. Because the plastic effects of neurofeedback training are not confined to the area or network whose activation is explicitly used as the target for training, both processes can very well happen in tandem, making it a versatile tool for redressing imbalances in and between brain networks. Up to now, the clinical potential of fMRI-NF to improve symptoms or change behavior has only been explored in a small number of published studies, all with small patient numbers. FMRI-NF, targeting the anterior cingulate cortex, has shown effects on chronic pain in patients with fibromyalgia (Decharms et al., 2005). Two out of six patients with chronic tinnitus improved after training down-regulation of auditory cortex activity (Haller, Birbaumer, & Veit, 2010). In our study of Parkinson’s disease (PD), we trained patients in the early stages of the disease to up-regulate their supplementary motor area (SMA) (Subramanian et al., 2011). The choice of target area was based on pathophysiological models implicating
Personalized neurofeedback Chapter 39
467
underactivity of the SMA in PD (Nachev, Kennard, & Husain, 2008), and based on our observation that reliable upregulation can be achieved through motor imagery. The five patients who received fMRI-NF, but not five control patients who engaged in motor imagery without feedback, achieved reliable SMA up-regulation, and improved motor fluency and clinical symptoms (Subramanian et al., 2011). In a randomized, controlled trial with the same fMRI-NF design, but addition of actual exercise components for both the NF and a control group, we found a similar size of clinical effects in the active group, but no statistical superiority over the control intervention (Subramanian et al., 2016). In the first psychiatric application, we trained patients with depression to up-regulate brain networks responsive to positive affective stimuli. This paradigm was modeled on our previous work with healthy participants, which had shown that the neurofeedback component is required for reliable control over emotion networks ( Johnston et al., 2011). The eight patients who underwent this fMRI-NF protocol for four sessions improved significantly on the 17-item Hamilton Rating Scale for Depression, and this clinical improvement was not observed in eight control patients who engaged in a protocol of positive emotional imagery (matched to the fMRI-NF protocol for intervention and assessment times and affective stimuli) outside the scanner (Linden et al., 2012). A recent randomized controlled trial pitting up-regulation training of the amygdala against up-regulation of a control area (the intraparietal sulcus region) found clinical improvement in patients with depression that was significantly stronger in the active (amygdala upregulation) than the control intervention (Young et al., 2017). We have recently replicated the clinical improvement after up-regulation training of areas responsive to positive emotions seen in our earlier work (Linden et al., 2012) although similar improvement was seen in a control neurofeedback intervention targeting the parahippocampal place area (Mehler et al., 2018). The feasibility of fMRI-NF has also been demonstrated for schizophrenia (up-regulation of anterior insula) (Ruiz et al., 2011) and stroke (upregulation of ventral premotor cortex) (Sitaram et al., 2012), but data on clinical improvement are not available. FMRI-NF has also been piloted for several substance use disorders (Fovet, Jardri, & Linden, 2015; Hartwell et al., 2013; Karch et al., 2015; Kirsch et al., 2016), and formal clinical trials are under way (Cox et al., 2016; Gerchen et al., 2018).
4 A personalized approach Neurofeedback, like most psychotherapeutic approaches, entails mobilization of personal resources and strategies. For example, each individual will have different strategies for increasing or decreasing activation in a particular brain region. Although the set of potential strategies may be more limited for some areas (e.g., there is only a limited range of familiar upper limb movements that people are likely to try imagining when asked to up-regulate higher motor areas), there would be a very wide range of imageries and thoughts that could be evoked to self-regulate limbic activation, and this is inextricably linked to each individual’s biography and disposition. In this sense, neurofeedback is a highly personalized tool, even if, for its clinical evaluation, standardized criteria (such as attaining a certain level of self-regulation) have to be defined. Beyond this general feature of personalization, neurofeedback protocols can also explicitly incorporate personalized elements at several levels: - Personalized neural targets - Personalized stimulus material - Personalized homework or other adjunct elements
4.1
Personalized neural targets
Defining personalized neural targets comes closest to the “personalized medicine” approach discussed elsewhere in this book. It is predicated on the assumption that pathological activation levels or patterns (too much, too little, or wrongly coupled) can be identified in individual patients (or subgroups of patients), which would be similar to finding a specific molecular marker or highly penetrant genetic risk variant. Identification of such altered brain activation patterns through a diagnostic fMRI scan could then be used to stratify patients for different neurofeedback protocols, targeting the respective dysfunctional network. This approach would require identification of such diagnostic patterns of brain activation in patients (and ideally also the “healthy” target pattern to aim for). An increasing body of work is looking at the reliability of such network markers based on resting-state fMRI, and at some point, it might become possible to identify patients with, say, depression, reliably based on their resting-state fMRI signal, or even subtype them into diagnostic subgroups. However, the relevant features may not be ideal targets for fMRI-NF because of their slow evolution over time. Resting-state fMRI patterns are driven by low-frequency oscillations of the BOLD signal (<0.1 Hz), and thus it would take several tens of seconds for a particular mental strategy to be reflected in a change of resting pattern. However, delayed feedback protocols have been developed, and for some tasks, may actually provide better outcomes than immediate feedback, so the problem of target signals that are slow to change is surmountable.
468
Personalized psychiatry
Most fMRI-NF studies do not define their activation targets through resting-state patterns, though, but use a functional localizer to identify a functional cluster or network that then becomes the target for up- or down-regulation in subsequent neurofeedback training runs. This is another way of personalizing the neural target, although it is not necessarily demonstrated that the patient’s activation would be different to that of a control group. Here the idea is that a functional network needs to be identified at an individual level that should be targeted—because it may play a role in a dysfunction, or may be involved in compensatory processes—although sometimes more standardized constraints are applied. For example, when identifying individual networks responsive to alcohol cues, which can vary greatly from participant to participant (and even within participants from scanning session to scanning session), one could implement anatomical constraints, for example, by only searching within a mask of the basal ganglia, limbic, and paralimbic areas.
4.2
Personalized stimuli and other elements of the neurofeedback therapy
Such functional localizers commonly use material that is relevant to the disease process or the putative functional deficit. Examples include positive affective pictures for localization of emotionally responsive regions in depression (Linden et al., 2012), contamination scenes in contamination anxiety (Scheinost et al., 2013), pictures of spiders in arachnophobia (Zilverstand et al., 2015), pictures of food to train regulation of food craving (Ihssen et al., 2016), or pictures of addictive cues for use in substance (and potentially also behavioral) addictions (Sokunbi et al., 2014). Although the desired activation patterns (and accompanying physiological responses, e.g., arousal) can often be induced with generic stimuli for many of these protocols, a personalized approach is preferable. If the aim of the neurofeedback process is directly to target dysfunctional activation patterns that arise during (and presumably contribute to) clinical symptoms, it will be necessary to induce these symptoms, or at least subclinical surrogates or certain key component processes, during the scanning, of course with the appropriate consent, permissions, and clinical cover in place. Symptom induction is generally most effective with personalized stimuli. For example, patients with obsessive compulsive disorder (OCD) may be invited to take pictures of the scenarios that trigger their symptoms, which can then be used for symptom induction (De Putter, Van Yper, & Koster, 2017). Exposure to triggering events—generally through imagination, but increasingly also through use of virtual reality—is also a key component of psychological interventions for PTSD, and personalized scenarios or scripts can be used to localize disease-relevant areas. Another related application is their presentation during the neurofeedback runs: here, patients will be instructed to train a neural target, for example, down-regulation of amygdala activity during exposure to personalized triggers (trauma words, Nicholson et al., 2017) with the overall therapeutic aim of decreasing their salience. FMRI-NF would thus be incorporated in a wider, personalized extinction therapy. Related concepts have been implemented in fMRI-NF protocols for substance use disorders. In alcohol dependence, for example, both patients’ preferred drinks and their individual motivation goal identified during therapy vary considerably among individuals, and both robust neural activations and meaningful psychological engagement likely depends on the individually tailored selection of pictures, which may require prescanning picture rating sessions (Cox et al., 2016). If these personalized pictures are then also presented during the fMRI-NF training, and indicate training success by varying their size (Sokunbi et al., 2014), the neurofeedback intervention assumes elements of a personalized craving reduction program.
5 Concluding remarks These examples may show how personalized elements that are already routinely used in psychological therapies—in sensu, in vivo, or in virtuo exposure for anxiety disorders and OCD; identification of individualized motivational goals for addiction—can be incorporated in neurofeedback training protocols. Combined with the individual definition of target brain areas/networks and the individualized strategies that patients will use to achieve their targets, the use of customized stimulus makes neurofeedback a highly personalized intervention. This may be attractive for patients who are interested in incorporating neural elements in their symptom reduction training, and more generally want to explore the relationship between their brain activation and mental life, but it also poses challenges for the accurate description of component processes and evaluation of their implementation, for example, through LOGIC models (Quinn et al., 2016). Even with these personalized features, neurofeedback can still be standardized and manualized in a manner sufficient to allow formal testing in randomized controlled trials.
Acknowledgment Some of this material has previously been published in: Goebel R., & Linden D. (2014). Neurofeedback with real-time functional MRI. In C. Mulert & M. Shenton (Eds.), MRI in psychiatry. Springer, Berlin: Heidelberg.
Personalized neurofeedback Chapter 39
469
References Bandura, A. (1997). Self-efficacy: The exercise of control. New York: W.H. Freeman. ix, 604p. ISBN 0716726262 (cased) 0716728508 (pbk). Birbaumer, N., & Cohen, L. (2007). Brain-computer interfaces: Communication and restoration of movement in paralysis. The Journal of Physiology, 579 (Pt. 3), 621–636. ISSN 0022-3751. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation& list_uids¼17234696. Cox, W. M. , et al. (2016). Neurofeedback training for alcohol dependence versus treatment as usual: Study protocol for a randomized controlled trial. Trials, 17(1), 480. ISSN 1745-6215. Available from: https://www.ncbi.nlm.nih.gov/pubmed/27716290. De Putter, L. M., Van Yper, L., & Koster, E. H. (2017). Obsessions and compulsions in the lab: A meta-analysis of procedures to induce symptoms of obsessive-compulsive disorder. Clinical Psychology Review, 52, 137–147. ISSN 1873-7811. Available from: https://www.ncbi.nlm.nih.gov/pubmed/ 28119197. Decharms, R. , et al. (2005). Control over brain activation and pain learned by using real-time functional MRI. Proceedings of the National Academy of Sciences of the United States of America, 102(51), 18626–18631. ISSN 0027-8424. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi? cmd¼Retrieve&db¼PubMed&dopt¼Citation&list_uids¼16352728. Fovet, T., Jardri, R., & Linden, D. (2015). Current issues in the use of fMRI-based neurofeedback to relieve psychiatric symptoms. Current Pharmaceutical Design, 21(23), 3384–3394. ISSN 1873-4286. Available from: http://www.ncbi.nlm.nih.gov/pubmed/26088117. Gerchen, M. F. , et al. (2018). The SyBil-AA real-time fMRI neurofeedback study: Protocol of a single-blind randomized controlled trial in alcohol use disorder. BMC Psychiatry, 18(1), 12. ISSN 1471-244X. Available from: https://www.ncbi.nlm.nih.gov/pubmed/29343230. Haller, S., Birbaumer, N., & Veit, R. (2010). Real-time fMRI feedback training may improve chronic tinnitus. European Radiology, 20(3), 696–703. ISSN 1432-1084. Available from: http://www.ncbi.nlm.nih.gov/pubmed/19760238. Hammond, D. (2005). Neurofeedback with anxiety and affective disorders. Child and Adolescent Psychiatric Clinics of North America, 14(1), 105–123. vii. ISSN 1056-4993. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation&list_ uids¼15564054. Hartwell, K. J. , et al. (2013). Real-time fMRI in the treatment of nicotine dependence: A conceptual review and pilot studies. Psychology of Addictive Behaviors, 27(2), 501–509. ISSN 1939-1501. Available from: https://www.ncbi.nlm.nih.gov/pubmed/22564200. Ihssen, N. , et al. (2016). Neurofeedback of visual food cue reactivity: A potential avenue to alter incentive sensitization and craving. Brain Imaging and Behavior, ISSN 1931-7565. Available from: http://www.ncbi.nlm.nih.gov/pubmed/27233784. Johnston, S. , et al. (2011). Upregulation of emotion areas through neurofeedback with a focus on positive mood. Cognitive, Affective, & Behavioral Neuroscience, 11(1), 44–51. ISSN 1531-135X. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21264651. Karch, S. , et al. (2015). Modulation of craving related brain responses using real-time fMRI in patients with alcohol use disorder. PLoS ONE, 10(7) ISSN 1932-6203. Available from: http://www.ncbi.nlm.nih.gov/pubmed/26204262. Kirsch, M. , et al. (2016). Real-time functional magnetic resonance imaging neurofeedback can reduce striatal cue-reactivity to alcohol stimuli. Addiction Biology, 21(4), 982–992. ISSN 1369-1600. Available from: https://www.ncbi.nlm.nih.gov/pubmed/26096546. Lansbergen, M. M. , et al. (2011). ADHD and EEG-neurofeedback: A double-blind randomized placebo-controlled feasibility study. Journal of Neural Transmission, 118(2), 275–284. ISSN 1435-1463. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21165661. Linden, D. E. (2012). The challenges and promise of neuroimaging in psychiatry. Neuron, 73(1), 8–22. ISSN 1097-4199. Available from: http://www. ncbi.nlm.nih.gov/pubmed/22243743. Linden, D. E. J. , et al. (2012). Real-time self-regulation of emotion networks in patients with depression. PLoS ONE, 7, e38115. Lofthouse, N. , et al. (2011). A review of neurofeedback treatment for pediatric ADHD. Journal of Attention Disorders, ISSN 1557-1246. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22090396. Mehler, D. M. A., Sokunbi, M. O., Habes, I., Barawi, K., Subramanian, L., Range, M., … Linden, D. E. J. (2018). Targeting the affective brain-a randomized controlled trial of real-time fMRI neurofeedback in patients with depression. Neuropsychopharmacology, 43(13), 2578–2585. Nachev, P., Kennard, C., & Husain, M. (2008). Functional role of the supplementary and pre-supplementary motor areas. Nature Reviews Neuroscience, 9 (11), 856–869. ISSN 1471-0048. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation& list_uids¼18843271. Nicholson, A. A. , et al. (2017). The neurobiology of emotion regulation in posttraumatic stress disorder: Amygdala downregulation via real-time fMRI neurofeedback. Human Brain Mapping, 38(1), 541–560. ISSN 1097-0193. Available from: https://www.ncbi.nlm.nih.gov/pubmed/27647695. Quinn, L. , et al. (2016). Development and delivery of a physical activity intervention for people with Huntington disease: Facilitating translation to clinical practice. Journal of Neurologic Physical Therapy, 40(2), 71–80. ISSN 1557-0584. Available from: https://www.ncbi.nlm.nih.gov/pubmed/26863152. Ruiz, S. , et al. (2011). Acquired self-control of insula cortex modulates emotion recognition and brain network connectivity in schizophrenia. Human Brain Mapping, ISSN 1097-0193. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22021045. Scheinost, D. , et al. (2013). Orbitofrontal cortex neurofeedback produces lasting changes in contamination anxiety and resting-state connectivity. Translational Psychiatry, 3, e250. ISSN 2158-3188. Available from: http://www.ncbi.nlm.nih.gov/pubmed/23632454. Sch€ onenberg, M. , et al. (2017). Neurofeedback, sham neurofeedback, and cognitive-behavioural group therapy in adults with attention-deficit hyperactivity disorder: A triple-blind, randomised, controlled trial. Lancet Psychiatry, 4(9), 673–684. ISSN 2215-0374. Available from: https://www.ncbi. nlm.nih.gov/pubmed/28803030. Seitz, R. J. (2010). How imaging will guide rehabilitation. Current Opinion in Neurology, 23(1), 79–86. ISSN 1473-6551. Available from: http://www. ncbi.nlm.nih.gov/pubmed/19926990.
470
Personalized psychiatry
Sitaram, R. , et al. (2012). Acquired control of ventral premotor cortex activity by feedback training: An exploratory real-time FMRI and TMS study. Neurorehabilitation and Neural Repair, 26(3), 256–265. ISSN 1552-6844. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21903976. Sokunbi, M. O. , et al. (2014). Real-time fMRI brain-computer interface: Development of a “motivational feedback” subsystem for the regulation of visual cue reactivity. Frontiers in Behavioral Neuroscience, 8, 392. ISSN 1662-5153. Available from: http://www.ncbi.nlm.nih.gov/pubmed/25505392. Subramanian, L. , et al. (2011). Real-time functional magnetic resonance imaging neurofeedback for treatment of Parkinson’s disease. The Journal of Neuroscience, 31(45), 16309–16317. ISSN 1529-2401. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22072682. Subramanian, L. , et al. (2016). Functional magnetic resonance imaging neurofeedback-guided motor imagery training and motor training for Parkinson’s disease: Randomized trial. Frontiers in Behavioral Neuroscience, 10, 111. Available from: https://www.ncbi.nlm.nih.gov/pubmed/27375451. Weiskopf, N. (2011). Real-time fMRI and its application to neurofeedback. NeuroImage, ISSN 1095-9572. Available from: http://www.ncbi.nlm.nih.gov/ pubmed/22019880. Weiskopf, N. , et al. (2004). Self-regulation of local brain activity using real-time functional magnetic resonance imaging (fMRI). Journal of Physiology, Paris, 98(4–6), 357–373. ISSN 0928-4257. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed& dopt¼Citation&list_uids¼16289548. Young, K. D. , et al. (2017). Randomized clinical trial of real-time fMRI amygdala neurofeedback for major depressive disorder: Effects on symptoms and autobiographical memory recall. The American Journal of Psychiatry, ISSN 1535-7228. Available from: https://www.ncbi.nlm.nih.gov/pubmed/ 28407727. Zilverstand, A. , et al. (2015). fMRI neurofeedback facilitates anxiety regulation in females with spider phobia. Frontiers in Behavioral Neuroscience, 9, 148. ISSN 1662-5153. Available from: http://www.ncbi.nlm.nih.gov/pubmed/26106309.
Real-time fMRI brain-computer interface: A tool for personalized psychiatry? David E.J. Linden Faculty of Health, Medicine and Life Sciences, School of Mental Health and Neuroscience, Maastricht University, Maastricht, The Netherlands; Division of Psychological Medicine and Clinical Neurosciences, School of Medicine, Cardiff University, Cardiff, United Kingdom
1 Introduction Since its invention 20 years ago, functional magnetic resonance imaging (fMRI) has become one of the most widely used, and probably the most publicly visible noninvasive technique to measure brain activation. fMRI has played a central role in the development of cognitive neuroscience, and several new fields, including social neuroscience, neuroeconomics, and genetic imaging, which may not have developed had it not been for the unique opportunities afforded by fMRI. The particular strengths of this technique are in its spatial resolution and fidelity, ability to reach deep subcortical structures, and whole-brain coverage, enabling the mapping of functionally connected networks, and the extraction of information from activation patterns that are distributed across different brain areas. In the psychiatric domain, fMRI has made major contributions to the understanding of psychopathology, and the effects of risk genes on cognitive and affective networks (Linden, 2012). In neurology, fMRI has become a central technique for mapping neuroplasticity; for example, in recovery from stroke (Seitz, 2010), and for presurgical mapping. However, fMRI has not yet fulfilled its translational potential, and there is, as of today, no established diagnostic, prognostic, or therapeutic use of this technique for any of the mental disorders. Factors that may have held back the clinical development of fMRI include its lack of molecular resolution (where radionuclide imaging has an advantage), semi-quantitative nature, dependence on control/baseline conditions (with the exception of resting-state fMRI), and difficulty controlling for physiological confounds (cardiovascular, respiratory, and eye and head movements). fMRI-based neurofeedback (fMRI-NF) has the potential to open up radically new paths to translation. During fMRI-NF training, participants receive feedback on their brain activity in real time, and are instructed to change this activation. This change is normally a simple up- or down-regulation, but could also entail changing the activation difference between two areas, their correlation, or the output of a multivariate pattern classification algorithm. The “hemodynamic” delay between neural activity and the vascular response that contributes to the fMRI signal, which is approximately 5 s, does not pose an obstacle when participants are informed of this delay (Weiskopf et al., 2004). The technical requirements include direct network access to the scanner hardware, software that processes the incoming data in real-time and computes incremental online statistics, and software for the conversion of the real-time activation data into sensory stimulation (e.g., visual, tactile) for the participant. The relevant workflows for such a “brain computer interface” (BCI) have been established on systems from all major MRI scanner manufacturers, and several custom-made and commercial software packages have been developed for online preprocessing and statistical analysis of fMRI data. Over the past 10years, fMRI-NF has been used to train healthy participants in the self-regulation of motor, sensory, language, and emotion areas (Weiskopf, 2011). In analogy to the effects of other noninvasive or invasive brain stimulation techniques, one should expect that such self-regulation also has consequences at the behavioral, cognitive, and affective level. However, clinical applications have been sparse, and had initially focused on chronic pain (Decharms et al., 2005). Compared with other neurofeedback techniques (with EEG or MEG), and with noninvasive physical stimulation techniques (tDCS and TMS), fMRI-NF has the advantage of higher spatial resolution and localization accuracy, and better access to deep limbic and subcortical structures. Compared with deep-brain stimulation, fMRI-NF has the advantage of noninvasiveness and spatial flexibility (although it is not intended to replace established DBS indications, e.g., in Parkinson’s disease [PD]). Finally, compared with all external stimulation techniques, neurofeedback has the attractive characteristic of enabling the patients themselves to control their brain activity, and thus contributing to their experience of self-efficacy, which is an important therapeutic factor in many neuropsychiatric disorders (Bandura, 1997). Personalized Psychiatry. https://doi.org/10.1016/B978-0-12-813176-3.00039-0 Copyright © 2020 Elsevier Inc. All rights reserved.
465
466
Personalized psychiatry
fMRI-NF has been boosted by developments in MR physics, allowing for fast acquisition of high quality data sets, and statistics, allowing for online calculation of univariate and multivariate statistics. It is now a mature experimental technique, and the exciting question for the next decade will be whether it can make a true translational contribution in medicine. The central challenges are identifying the symptoms and disorders that will respond to fMRI-NF, adapting the treatment protocols to the neural networks involved in each of them, evaluating the underlying neuroplastic mechanisms, and devising training strategies that enable sustainable long-term effects.
2 Principles and methods Attempts to train humans and animals to regulate their own brain activity, feeding back signals from noninvasive (electroencephalography [EEG]) or invasive recordings, go back to the 1960s (reviewed by Birbaumer & Cohen, 2007). The theoretical principles were largely derived from operant conditioning, whereby the participant learns the optimal strategy through the contingencies between their actions and a reward. In an animal, the reward would have to be a primary reward. For example, the desired neural response would be reinforced by the delivery of food or drink. In humans, the information about the achievement of a particular neural target (e.g., increasing the ratio between theta and alpha power of the EEG by 20%) could itself be the reward. Whether this reward reinforces a particular mental strategy that leads to the desired physiological response, or the physiological response itself is a futile question because the two cannot be separated in any meaningful way under the assumptions of psychophysical unity. It has been reported that humans can achieve reliable selfcontrol over parameters of their EEG, for example, the topography of slow cortical potentials, the alpha-theta ratio, or the ratio between sensorimotor rhythm and theta activity, through such operant conditioning. FMRI-based neurofeedback (fMRI-NF) can be conducted in a similar fashion, where participants are blind to the functional relevance of the targeted activation, and essentially aim to achieve self-regulation by trial and error (Weiskopf et al., 2004). However, fMRI-NF can also harness the considerable knowledge about the neural basis of particular mental and cognitive processes that the past 20 years of functional brain mapping have achieved, and introduce a cognitive component into the learning strategy. For example, in our study on upregulation of the supplementary motor area in Parkinson’s disease (Subramanian et al., 2011), patients were told about the motor planning functions of the target area, and informed that motor imagery might be one viable strategy to up-regulate it. Giving patients initial hints about potential strategies for the upregulation training, which they can then refine through an operant conditioning protocol with fMRI-NF, can considerably reduce scanning time, and provide patients with tangible results within one scanning session (Subramanian et al., 2011).
3 Clinical applications of fMRI-NF Clinical studies of EEG neurofeedback (EEF-NF) training have been conducted in epilepsy, attention deficit/hyperactivity disorder (ADHD), depression and anxiety, as well as other neurological and psychiatric disorders (Birbaumer & Cohen, 2007; Hammond, 2005). Although some of the initial results were promising, the only field where EEG-NF has entered clinical practice is ADHD (using a variety of target parameters, e.g., theta/beta ratio or slow cortical potentials). For ADHD, several published trials found positive effects on symptoms as measured by parent or teacher questionnaires or clinical assessments (Lofthouse et al., 2011). One limitation of most trials has been the lack of blinding of participants to treatment condition, and the general difficulties of setting up of double-blind controlled trials of NF training (Lansbergen et al., 2011), and a recent properly blinded trial did not find any superiority for the active EEG-neurofeedback intervention (Sch€ onenberg et al., 2017). Although these difficulties will apply similarly in fMRI-NF, there are good reasons to explore its use in clinical conditions, including and beyond those that may respond to EEG-NF. Because of the strong cognitive component, fMRINF may achieve its neurophysiological targets faster than EEG-NF, and thus enhance patient motivation and compliance. Furthermore, fMRI-NF can directly target specific brain areas or networks that have been implicated in neuropsychiatric disorders, either because they show a dysfunction that might cause the disorder, or because they may compensate for a primary dysfunction. In the first case, the aim of neurofeedback would be to restore the function to normal levels, and in the second, to promote the recruitment of the compensatory network. Because the plastic effects of neurofeedback training are not confined to the area or network whose activation is explicitly used as the target for training, both processes can very well happen in tandem, making it a versatile tool for redressing imbalances in and between brain networks. Up to now, the clinical potential of fMRI-NF to improve symptoms or change behavior has only been explored in a small number of published studies, all with small patient numbers. FMRI-NF, targeting the anterior cingulate cortex, has shown effects on chronic pain in patients with fibromyalgia (Decharms et al., 2005). Two out of six patients with chronic tinnitus improved after training down-regulation of auditory cortex activity (Haller, Birbaumer, & Veit, 2010). In our study of Parkinson’s disease (PD), we trained patients in the early stages of the disease to up-regulate their supplementary motor area (SMA) (Subramanian et al., 2011). The choice of target area was based on pathophysiological models implicating
Personalized neurofeedback Chapter 39
467
underactivity of the SMA in PD (Nachev, Kennard, & Husain, 2008), and based on our observation that reliable upregulation can be achieved through motor imagery. The five patients who received fMRI-NF, but not five control patients who engaged in motor imagery without feedback, achieved reliable SMA up-regulation, and improved motor fluency and clinical symptoms (Subramanian et al., 2011). In a randomized, controlled trial with the same fMRI-NF design, but addition of actual exercise components for both the NF and a control group, we found a similar size of clinical effects in the active group, but no statistical superiority over the control intervention (Subramanian et al., 2016). In the first psychiatric application, we trained patients with depression to up-regulate brain networks responsive to positive affective stimuli. This paradigm was modeled on our previous work with healthy participants, which had shown that the neurofeedback component is required for reliable control over emotion networks ( Johnston et al., 2011). The eight patients who underwent this fMRI-NF protocol for four sessions improved significantly on the 17-item Hamilton Rating Scale for Depression, and this clinical improvement was not observed in eight control patients who engaged in a protocol of positive emotional imagery (matched to the fMRI-NF protocol for intervention and assessment times and affective stimuli) outside the scanner (Linden et al., 2012). A recent randomized controlled trial pitting up-regulation training of the amygdala against up-regulation of a control area (the intraparietal sulcus region) found clinical improvement in patients with depression that was significantly stronger in the active (amygdala upregulation) than the control intervention (Young et al., 2017). We have recently replicated the clinical improvement after up-regulation training of areas responsive to positive emotions seen in our earlier work (Linden et al., 2012) although similar improvement was seen in a control neurofeedback intervention targeting the parahippocampal place area (Mehler et al., 2018). The feasibility of fMRI-NF has also been demonstrated for schizophrenia (up-regulation of anterior insula) (Ruiz et al., 2011) and stroke (upregulation of ventral premotor cortex) (Sitaram et al., 2012), but data on clinical improvement are not available. FMRI-NF has also been piloted for several substance use disorders (Fovet, Jardri, & Linden, 2015; Hartwell et al., 2013; Karch et al., 2015; Kirsch et al., 2016), and formal clinical trials are under way (Cox et al., 2016; Gerchen et al., 2018).
4 A personalized approach Neurofeedback, like most psychotherapeutic approaches, entails mobilization of personal resources and strategies. For example, each individual will have different strategies for increasing or decreasing activation in a particular brain region. Although the set of potential strategies may be more limited for some areas (e.g., there is only a limited range of familiar upper limb movements that people are likely to try imagining when asked to up-regulate higher motor areas), there would be a very wide range of imageries and thoughts that could be evoked to self-regulate limbic activation, and this is inextricably linked to each individual’s biography and disposition. In this sense, neurofeedback is a highly personalized tool, even if, for its clinical evaluation, standardized criteria (such as attaining a certain level of self-regulation) have to be defined. Beyond this general feature of personalization, neurofeedback protocols can also explicitly incorporate personalized elements at several levels: - Personalized neural targets - Personalized stimulus material - Personalized homework or other adjunct elements
4.1
Personalized neural targets
Defining personalized neural targets comes closest to the “personalized medicine” approach discussed elsewhere in this book. It is predicated on the assumption that pathological activation levels or patterns (too much, too little, or wrongly coupled) can be identified in individual patients (or subgroups of patients), which would be similar to finding a specific molecular marker or highly penetrant genetic risk variant. Identification of such altered brain activation patterns through a diagnostic fMRI scan could then be used to stratify patients for different neurofeedback protocols, targeting the respective dysfunctional network. This approach would require identification of such diagnostic patterns of brain activation in patients (and ideally also the “healthy” target pattern to aim for). An increasing body of work is looking at the reliability of such network markers based on resting-state fMRI, and at some point, it might become possible to identify patients with, say, depression, reliably based on their resting-state fMRI signal, or even subtype them into diagnostic subgroups. However, the relevant features may not be ideal targets for fMRI-NF because of their slow evolution over time. Resting-state fMRI patterns are driven by low-frequency oscillations of the BOLD signal (<0.1 Hz), and thus it would take several tens of seconds for a particular mental strategy to be reflected in a change of resting pattern. However, delayed feedback protocols have been developed, and for some tasks, may actually provide better outcomes than immediate feedback, so the problem of target signals that are slow to change is surmountable.
468
Personalized psychiatry
Most fMRI-NF studies do not define their activation targets through resting-state patterns, though, but use a functional localizer to identify a functional cluster or network that then becomes the target for up- or down-regulation in subsequent neurofeedback training runs. This is another way of personalizing the neural target, although it is not necessarily demonstrated that the patient’s activation would be different to that of a control group. Here the idea is that a functional network needs to be identified at an individual level that should be targeted—because it may play a role in a dysfunction, or may be involved in compensatory processes—although sometimes more standardized constraints are applied. For example, when identifying individual networks responsive to alcohol cues, which can vary greatly from participant to participant (and even within participants from scanning session to scanning session), one could implement anatomical constraints, for example, by only searching within a mask of the basal ganglia, limbic, and paralimbic areas.
4.2
Personalized stimuli and other elements of the neurofeedback therapy
Such functional localizers commonly use material that is relevant to the disease process or the putative functional deficit. Examples include positive affective pictures for localization of emotionally responsive regions in depression (Linden et al., 2012), contamination scenes in contamination anxiety (Scheinost et al., 2013), pictures of spiders in arachnophobia (Zilverstand et al., 2015), pictures of food to train regulation of food craving (Ihssen et al., 2016), or pictures of addictive cues for use in substance (and potentially also behavioral) addictions (Sokunbi et al., 2014). Although the desired activation patterns (and accompanying physiological responses, e.g., arousal) can often be induced with generic stimuli for many of these protocols, a personalized approach is preferable. If the aim of the neurofeedback process is directly to target dysfunctional activation patterns that arise during (and presumably contribute to) clinical symptoms, it will be necessary to induce these symptoms, or at least subclinical surrogates or certain key component processes, during the scanning, of course with the appropriate consent, permissions, and clinical cover in place. Symptom induction is generally most effective with personalized stimuli. For example, patients with obsessive compulsive disorder (OCD) may be invited to take pictures of the scenarios that trigger their symptoms, which can then be used for symptom induction (De Putter, Van Yper, & Koster, 2017). Exposure to triggering events—generally through imagination, but increasingly also through use of virtual reality—is also a key component of psychological interventions for PTSD, and personalized scenarios or scripts can be used to localize disease-relevant areas. Another related application is their presentation during the neurofeedback runs: here, patients will be instructed to train a neural target, for example, down-regulation of amygdala activity during exposure to personalized triggers (trauma words, Nicholson et al., 2017) with the overall therapeutic aim of decreasing their salience. FMRI-NF would thus be incorporated in a wider, personalized extinction therapy. Related concepts have been implemented in fMRI-NF protocols for substance use disorders. In alcohol dependence, for example, both patients’ preferred drinks and their individual motivation goal identified during therapy vary considerably among individuals, and both robust neural activations and meaningful psychological engagement likely depends on the individually tailored selection of pictures, which may require prescanning picture rating sessions (Cox et al., 2016). If these personalized pictures are then also presented during the fMRI-NF training, and indicate training success by varying their size (Sokunbi et al., 2014), the neurofeedback intervention assumes elements of a personalized craving reduction program.
5 Concluding remarks These examples may show how personalized elements that are already routinely used in psychological therapies—in sensu, in vivo, or in virtuo exposure for anxiety disorders and OCD; identification of individualized motivational goals for addiction—can be incorporated in neurofeedback training protocols. Combined with the individual definition of target brain areas/networks and the individualized strategies that patients will use to achieve their targets, the use of customized stimulus makes neurofeedback a highly personalized intervention. This may be attractive for patients who are interested in incorporating neural elements in their symptom reduction training, and more generally want to explore the relationship between their brain activation and mental life, but it also poses challenges for the accurate description of component processes and evaluation of their implementation, for example, through LOGIC models (Quinn et al., 2016). Even with these personalized features, neurofeedback can still be standardized and manualized in a manner sufficient to allow formal testing in randomized controlled trials.
Acknowledgment Some of this material has previously been published in: Goebel R., & Linden D. (2014). Neurofeedback with real-time functional MRI. In C. Mulert & M. Shenton (Eds.), MRI in psychiatry. Springer, Berlin: Heidelberg.
Personalized neurofeedback Chapter 39
469
References Bandura, A. (1997). Self-efficacy: The exercise of control. New York: W.H. Freeman. ix, 604p. ISBN 0716726262 (cased) 0716728508 (pbk). Birbaumer, N., & Cohen, L. (2007). Brain-computer interfaces: Communication and restoration of movement in paralysis. The Journal of Physiology, 579 (Pt. 3), 621–636. ISSN 0022-3751. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation& list_uids¼17234696. Cox, W. M. , et al. (2016). Neurofeedback training for alcohol dependence versus treatment as usual: Study protocol for a randomized controlled trial. Trials, 17(1), 480. ISSN 1745-6215. Available from: https://www.ncbi.nlm.nih.gov/pubmed/27716290. De Putter, L. M., Van Yper, L., & Koster, E. H. (2017). Obsessions and compulsions in the lab: A meta-analysis of procedures to induce symptoms of obsessive-compulsive disorder. Clinical Psychology Review, 52, 137–147. ISSN 1873-7811. Available from: https://www.ncbi.nlm.nih.gov/pubmed/ 28119197. Decharms, R. , et al. (2005). Control over brain activation and pain learned by using real-time functional MRI. Proceedings of the National Academy of Sciences of the United States of America, 102(51), 18626–18631. ISSN 0027-8424. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi? cmd¼Retrieve&db¼PubMed&dopt¼Citation&list_uids¼16352728. Fovet, T., Jardri, R., & Linden, D. (2015). Current issues in the use of fMRI-based neurofeedback to relieve psychiatric symptoms. Current Pharmaceutical Design, 21(23), 3384–3394. ISSN 1873-4286. Available from: http://www.ncbi.nlm.nih.gov/pubmed/26088117. Gerchen, M. F. , et al. (2018). The SyBil-AA real-time fMRI neurofeedback study: Protocol of a single-blind randomized controlled trial in alcohol use disorder. BMC Psychiatry, 18(1), 12. ISSN 1471-244X. Available from: https://www.ncbi.nlm.nih.gov/pubmed/29343230. Haller, S., Birbaumer, N., & Veit, R. (2010). Real-time fMRI feedback training may improve chronic tinnitus. European Radiology, 20(3), 696–703. ISSN 1432-1084. Available from: http://www.ncbi.nlm.nih.gov/pubmed/19760238. Hammond, D. (2005). Neurofeedback with anxiety and affective disorders. Child and Adolescent Psychiatric Clinics of North America, 14(1), 105–123. vii. ISSN 1056-4993. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation&list_ uids¼15564054. Hartwell, K. J. , et al. (2013). Real-time fMRI in the treatment of nicotine dependence: A conceptual review and pilot studies. Psychology of Addictive Behaviors, 27(2), 501–509. ISSN 1939-1501. Available from: https://www.ncbi.nlm.nih.gov/pubmed/22564200. Ihssen, N. , et al. (2016). Neurofeedback of visual food cue reactivity: A potential avenue to alter incentive sensitization and craving. Brain Imaging and Behavior, ISSN 1931-7565. Available from: http://www.ncbi.nlm.nih.gov/pubmed/27233784. Johnston, S. , et al. (2011). Upregulation of emotion areas through neurofeedback with a focus on positive mood. Cognitive, Affective, & Behavioral Neuroscience, 11(1), 44–51. ISSN 1531-135X. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21264651. Karch, S. , et al. (2015). Modulation of craving related brain responses using real-time fMRI in patients with alcohol use disorder. PLoS ONE, 10(7) ISSN 1932-6203. Available from: http://www.ncbi.nlm.nih.gov/pubmed/26204262. Kirsch, M. , et al. (2016). Real-time functional magnetic resonance imaging neurofeedback can reduce striatal cue-reactivity to alcohol stimuli. Addiction Biology, 21(4), 982–992. ISSN 1369-1600. Available from: https://www.ncbi.nlm.nih.gov/pubmed/26096546. Lansbergen, M. M. , et al. (2011). ADHD and EEG-neurofeedback: A double-blind randomized placebo-controlled feasibility study. Journal of Neural Transmission, 118(2), 275–284. ISSN 1435-1463. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21165661. Linden, D. E. (2012). The challenges and promise of neuroimaging in psychiatry. Neuron, 73(1), 8–22. ISSN 1097-4199. Available from: http://www. ncbi.nlm.nih.gov/pubmed/22243743. Linden, D. E. J. , et al. (2012). Real-time self-regulation of emotion networks in patients with depression. PLoS ONE, 7, e38115. Lofthouse, N. , et al. (2011). A review of neurofeedback treatment for pediatric ADHD. Journal of Attention Disorders, ISSN 1557-1246. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22090396. Mehler, D. M. A., Sokunbi, M. O., Habes, I., Barawi, K., Subramanian, L., Range, M., … Linden, D. E. J. (2018). Targeting the affective brain-a randomized controlled trial of real-time fMRI neurofeedback in patients with depression. Neuropsychopharmacology, 43(13), 2578–2585. Nachev, P., Kennard, C., & Husain, M. (2008). Functional role of the supplementary and pre-supplementary motor areas. Nature Reviews Neuroscience, 9 (11), 856–869. ISSN 1471-0048. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation& list_uids¼18843271. Nicholson, A. A. , et al. (2017). The neurobiology of emotion regulation in posttraumatic stress disorder: Amygdala downregulation via real-time fMRI neurofeedback. Human Brain Mapping, 38(1), 541–560. ISSN 1097-0193. Available from: https://www.ncbi.nlm.nih.gov/pubmed/27647695. Quinn, L. , et al. (2016). Development and delivery of a physical activity intervention for people with Huntington disease: Facilitating translation to clinical practice. Journal of Neurologic Physical Therapy, 40(2), 71–80. ISSN 1557-0584. Available from: https://www.ncbi.nlm.nih.gov/pubmed/26863152. Ruiz, S. , et al. (2011). Acquired self-control of insula cortex modulates emotion recognition and brain network connectivity in schizophrenia. Human Brain Mapping, ISSN 1097-0193. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22021045. Scheinost, D. , et al. (2013). Orbitofrontal cortex neurofeedback produces lasting changes in contamination anxiety and resting-state connectivity. Translational Psychiatry, 3, e250. ISSN 2158-3188. Available from: http://www.ncbi.nlm.nih.gov/pubmed/23632454. Sch€ onenberg, M. , et al. (2017). Neurofeedback, sham neurofeedback, and cognitive-behavioural group therapy in adults with attention-deficit hyperactivity disorder: A triple-blind, randomised, controlled trial. Lancet Psychiatry, 4(9), 673–684. ISSN 2215-0374. Available from: https://www.ncbi. nlm.nih.gov/pubmed/28803030. Seitz, R. J. (2010). How imaging will guide rehabilitation. Current Opinion in Neurology, 23(1), 79–86. ISSN 1473-6551. Available from: http://www. ncbi.nlm.nih.gov/pubmed/19926990.
470
Personalized psychiatry
Sitaram, R. , et al. (2012). Acquired control of ventral premotor cortex activity by feedback training: An exploratory real-time FMRI and TMS study. Neurorehabilitation and Neural Repair, 26(3), 256–265. ISSN 1552-6844. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21903976. Sokunbi, M. O. , et al. (2014). Real-time fMRI brain-computer interface: Development of a “motivational feedback” subsystem for the regulation of visual cue reactivity. Frontiers in Behavioral Neuroscience, 8, 392. ISSN 1662-5153. Available from: http://www.ncbi.nlm.nih.gov/pubmed/25505392. Subramanian, L. , et al. (2011). Real-time functional magnetic resonance imaging neurofeedback for treatment of Parkinson’s disease. The Journal of Neuroscience, 31(45), 16309–16317. ISSN 1529-2401. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22072682. Subramanian, L. , et al. (2016). Functional magnetic resonance imaging neurofeedback-guided motor imagery training and motor training for Parkinson’s disease: Randomized trial. Frontiers in Behavioral Neuroscience, 10, 111. Available from: https://www.ncbi.nlm.nih.gov/pubmed/27375451. Weiskopf, N. (2011). Real-time fMRI and its application to neurofeedback. NeuroImage, ISSN 1095-9572. Available from: http://www.ncbi.nlm.nih.gov/ pubmed/22019880. Weiskopf, N. , et al. (2004). Self-regulation of local brain activity using real-time functional magnetic resonance imaging (fMRI). Journal of Physiology, Paris, 98(4–6), 357–373. ISSN 0928-4257. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed& dopt¼Citation&list_uids¼16289548. Young, K. D. , et al. (2017). Randomized clinical trial of real-time fMRI amygdala neurofeedback for major depressive disorder: Effects on symptoms and autobiographical memory recall. The American Journal of Psychiatry, ISSN 1535-7228. Available from: https://www.ncbi.nlm.nih.gov/pubmed/ 28407727. Zilverstand, A. , et al. (2015). fMRI neurofeedback facilitates anxiety regulation in females with spider phobia. Frontiers in Behavioral Neuroscience, 9, 148. ISSN 1662-5153. Available from: http://www.ncbi.nlm.nih.gov/pubmed/26106309.