The Promise of Neurotechnology in Clinical Translational Science

Principle 1: Verifiable

The most crucial aspect of any neurotechnology system is that its revealed functionality can be evaluated against its planned functionality, which in essence makes it possible to evaluate (as true or false) whether the system has met the user’s requirement. Placed in the broader context of hypothesis-driven science, a “good” neurotechnology system is one whose expected behavior can be tested against a prediction and falsified. As an example, consider the following proposal: “We will produce a system that is user-friendly.” Although this is a laudable goal of any human-computer interaction, it is difficult to verify, and greater specification may be useful; for example, the system can be used by a specific category of users rather than all users in general. Some users may think that a system is user-friendly, whereas other users, for a variety of reasons, may not. This end-user variability introduces noise into a design that will make downstream evaluation of clinical efficacy more difficult. Extending this simplified example to a more complex system allows us to ask (and answer) the question, How will we explicitly know whether the design delivers the expected function? This is a simple but nontrivial issue because the answer to this question allows one to tease apart two critical but completely separate design features: (a) whether the system has done its job (Principle 1) and (b) whether the output of the system produces movement in the clinical construct of interest (corresponding to Principle 5: Mechanism Driven, discussed later).

Our concrete suggestion is that specific steps should be taken to understand the principal output of a system in a process of “build verification” (or build-verification testing, also called smoke testing; Kaner, Bach, & Pettichord, 2001). This is an early but crucial step intended to exercise the complete program to reveal failures severe enough to reject a prospective system design. It is important that build verification endorses a formal comprehensive method of assessing fundamental instability or key failures and, therefore, reveals unanticipated problems more efficiently and earlier than does ad hoc testing, which relies on tester skill and intuition. Build verification, at its essence, also involves the establishment of ground truth or objective data against which subsequent results are compared. For example, a finger-tapping pattern could be used to activate the motor cortex and establish accuracy of a machine-learning classifier based on EEG. We recommend early and, if possible, automated build verification as a cost-effective and timesaving functional test chiefly because it minimizes wasted time in the form of unusable collected data. Build verification therefore has the immediate benefit of finding important “bugs” fast and the long-term benefit of allowing differentiation between verifiability and clinical efficacy.

Principle 2: Useful

The utility principle underscores the notion that if any neurotechnology design is to produce a sustained, positive influence on the field, it will ideally fill a gap for which an established, well-understood, or simpler solution does not already exist or augment (improve on) some established approach. Alternatively, the technology system should offer some other benefit (e.g., more readily available or economical to the end user) than an existing protocol that is already established. Although we argue herein that utility of the system must be considered, we do not argue for its necessity. It is well accepted that treatments are not equivalently useful across all recipients (e.g., Weisz et al., 2013), and end-user preference ought to be considered. Some users (clients) might prefer use of technology, for instance, to a traditional therapeutic approach and, thus, respond better to the technology than they would to a human therapist.

Consider that in the case of most phobias, exposure-based treatment is able to capitalize on habituation and fear-extinction mechanisms as a relatively straightforward and brief method of intervention (cf. Seligman, Swedish, & Ollendick, 2014). There is always the possibility, however, of developing technology-based interventions around these well-understood processes. For some feared stimuli (animals, insects), it is almost certainly simpler to use the actual phobic stimuli in vivo. However, there are problems for which this is not practical, such as scenarios that are impossible to reproduce in the clinic (battlefields, heights, large bodies of water). In these cases, the necessity of a technology-supplemented treatment is evident if one considers that virtual reality (VR) could faithfully recreate certain conditions that are practically impossible to attain in the clinic. We therefore recommend that technology-enhanced treatment systems should be critically evaluated in terms of their actual or incremental value-added necessity and not in terms of their aesthetic appeal or superfluous properties that may serve to only pleasantly surprise the user (the “wow” factor).

Principle 3: Consistent

We operationalize the consistency attribute as the degree of alignment between bench capabilities and bedside requirements. Misalignment (inconsistency) can occur if either (a) the requirements of a bedside application exceed the capabilities of the available technology or (b) the available technology manipulates processes that are not relevant to the clinician or the target clinical problem. Inconsistency may result from lack of clear communication between the two sides of the house, namely, development (engineering) and application (clinicians). We believe that inconsistency, and the wasted effort that would result, can be avoided if there is a clearly established link between what is achievable within the constraints of neurotechnology designs and what is needed to test a hypothesized effect and ultimately resolve a clinical problem.

A mismatch scenario may not be obvious during project development, and for this reason, we specifically note that consistency should be evaluated within the crucible of construct validation. As an example, in the case of cochlear implants, a detailed knowledge of the tonotopic organization of the basilar membrane informed the development of prosthetic designs that stimulate the cochlear nerve according to the same principles, thus bypassing damaged or degraded hair cells completely and allowing the brain to interpret frequency signals as sound. Such a feat could not have occurred unless hardware design was able to deliver functionality that acted directly on the known physiology of the inner ear. In this example, consistency is evident if one considers that the neural-interface design braids into a metamodel that relates the intervention to the problem via logically supportable statements about electrophysiology, electrical engineering, and anatomy. Inconsistencies would be exposed in earnest during this process because estimates of discriminant and convergent validity (as two examples among many) would reveal that a prosthetic acts on a theoretically distinct construct from what was intended.

In clinical psychology, a conceptually similar example is found in the literature on BCI-based neurofeedback for ADHD. In a study by Lim et al. (2012), youth with ADHD completed an attention-training game in which a user controlled an avatar in a “race” using a dry-fit EEG headband. In this game, the higher the concentration level, the greater the speed of the avatar’s movement. This is an excellent example of consistency between the bioengineered device and the clinical application because the construct of attentional focus is both measureable within the constraints of the device and directly relevant to the clinical phenomenon of ADHD.

A corollary here is that fabrication details matter only to the extent that they operate on testable psychological phenomena. The power of a digital intervention is therefore not derived from its basis in a symbol-manipulating processor but, rather, by the fact that it delivers functionality that meshes well with the existing nomological net (by conforming to predictions without contradicting what is already known). In the broadest interpretation, any intervention could be held to this standard; however, we highlight it as particularly important in the domain of technology-enabled intervention or assessment because critical evaluation of consistency is meant to inform the central question: How can we effectively interface what is deliverable from an engineering standpoint with what is known from behavioral and cognitive research? To clarify, this principle refers not to the recommendation that an intervention produces consistent results (see the Principle 4: Reproducible section) but, rather, that it conforms to a nomological network that is defined in the traditional terms: by the relations among the constructs it measures.

Principle 4: Reproducible

A reproducible neurotechnology design is one that distributes the same effect on all, or almost all, cases of a particular user type. On one hand, this is intuitive, mainly because most of us are accustomed to human-computer interfaces that work for all or almost all users (e.g., LCD monitor, two-axis mouse). On the other hand, there are certain characteristics of psychiatric disorders that could conceivably, perhaps dramatically, lower the reproducibility of effects in neurotechnology systems. For instance, heterogeneity in Diagnostic and Statistical Manual of Mental Disorders (5th ed.; American Psychiatric Association, 2013) diagnoses, lack of robust endophenotypes, and documented developmental changes in pathophysiological mechanisms themselves must be taken into account in considerations of the efficacy of a digitally based intervention. Reproducibility could be construed as similar to the traditional notion of test-retest reliability, with the special exception that given the same inputs, deterministic algorithms will always return the same result. Our conceptualization of reproducibility recasts the traditional focus of reliability onto sources of variability that conspire to reduce the similarity of the effect across users.

Charles Babbage (1864), the English philosopher and inventor, noted in his memoirs,

On two occasions I have been asked: “Mr. Babbage, if you put into the machine the wrong figures, will the right answers come out?” I am not rightly able to apprehend the kind of confusion of ideas that could provoke such a question. (p. 67)

This quotation is intimately tied to our notion of reproducibility and highlights the “garbage-in-garbage-out” principle, which proposes that computers will process nonsensical data (garbage in) according to the same logical rules as any other data and produce nonsensical output (garbage out) as a result. Even the ubiquitous QWERTY keyboard could be blamed for irreproducible effects (i.e., high variability in typing speed or spelling errors across heterogeneous user types), even though the input/output and transfer characteristics of the keyboard are unchanged for each user. Conversely, given precisely homogeneous user types (e.g., 10 groups, each populated by users with exactly the same degree of typing skill), the reproducibility of the keyboard’s effect will approach 1.0 within that group because the error rate (garbage in) for the group remains constant.

In keeping with this truism, we challenge the assumption that reproducibility is the sole domain of the engineering design and argue that, in fact, the burden for ensuring reproducibility lies at the opposite end of the translational bridge. Undocumented clinical heterogeneity will introduce noise (garbage in) to an otherwise reproducible effect. Undocumented variation along axes of affect, cognition, and motivation that cut across diagnostic categories (Insel et al., 2010) adds to existing symptom-level variability within diagnostic classes (e.g., two individuals with antisocial-personality disorder may have no overlap in symptoms, given that only three of seven possible symptoms are required for a diagnosis; American Psychiatric Association, 2013), along with developmental principles, such as heterotypic continuity. These sources of variation should be taken into account because they each decrease the reproducibility of a system’s effect. Along these lines, most of these problems can be resolved by precisely defining the individuals for whom an intervention is likely to be successful. When they identify a set of clinical requirements that will eventually be addressed by an engineered-system design, clinicians must be clear on which processes will be manipulated by the intervention. Ultimately, this informs the larger question, For whom will this intervention be effective?

Principle 5: Mechanism driven

This principle highlights the fact that technology-enabled intervention designs should be anchored to a pathophysiological mechanism that can be logically explained and tested. According to this principle, if the clinician or any end user asks, “How does this intervention work?” a response of “I don’t know” indicates a shortcoming in design. To confront this problem, we endorse a policy of “no magic,” which specifies that during the design phase of a project, no unjustifiable decisions should be made (Larman, 2004) and that design should halt precisely before that frontier is crossed. Adherence to this policy ought to lead to a scenario wherein the choice of neurotechnology components and their relationship to a pathophysiological mechanism can be rationally explained and learned. A no-magic approach is not orthogonal to a hypothesis-testing approach. To the contrary, the ability to concretely explain every aspect of the logical flow of ideas, including one’s hypothesis, prevents abstraction of meaning from logical relations that are not (or cannot be) sufficiently established. The alternative is that the design indexes or manipulates some empty conceptual space, which leads to a black-box scenario in which the inner logic of a mechanism can be inferred only by its input and output characteristics. A particularly undesirable characteristic of a black-box scenario is that it explicitly requires a process of reverse engineering or going “backwards” in the development cycle to understand the process by which the intervention achieves its effects.

We consider cognitive-bias-modification (CBM) procedures, an increasingly popular and direct cognitive intervention for various forms of anxiety disorders in children (e.g., Eldar et al., 2012; Shechner et al., 2014) and adults (e.g., Schmidt, Richey, Buckner, & Timpano, 2009), an example of this backward cycling in intervention development. Most forms of CBM (with some exceptions in interpretation, Grey & Mathews, 2000, 2009, and recall training, cf. MacLeod & Mathews, 2004) are premised on the notion that a fast-latency attentional bias toward environmental-threat stimuli plays a causal role in the development and maintenance of the disorders and that by extension, removal of the cause by systematically training attention away from threat should eliminate the symptoms (e.g., MacPherson & Fristad, 2014). Although these general principles have been upheld by several dozen studies (cf. MacLeod & Mathews, 2012, for an integrative review), and three meta-analyses (Beard, Sawyer, & Hofmann, 2012; Hakamata et al., 2010; Hallion & Ruscio, 2011), the neurobiological mechanism by which CBM interventions achieve their effect is still unclear. As a consequence, recent work in this area is increasingly focused on revealing mechanisms that underlie the observed effects (Browning, Holmes, Murphy, Goodwin, & Harmer, 2010; Clarke, Browning, Hammond, Notebaert, & Macleod, 2014), which is a predictable and necessary reaction if the precise mechanism of action cannot yet be completely explained or directly tested.

Principle 6: Complete

An axiom that is frequently attributed to Albert Einstein (probably paraphrased from Einstein, 1934, p. 165) states that “everything should be made as simple as possible, but no simpler.” The completeness of neural-interface design refers to the desirable tension between parsimony and sufficiency in meeting the clinical objective or the reasonable expectation that the presence of the system will meet the user’s requirement. Any intervention or medical device that is necessary but not sufficient to produce change is not maximally useful because it is incomplete. However, there is an opposing force in our conceptualization here, to the extent that parsimony should constrain the development of systems that are unnecessarily complex. Excessively complex system designs may exceed the user’s requirements, and perhaps capabilities, by providing features that are redundant or unnecessary, thus making it difficult to disentangle the precise cause of any observed benefits or lack thereof. In the establishment of some initial common vision for a project, the tension between sufficiency and parsimony should be explicitly considered in light of the proposition that a system that has complex design features will, by definition, obscure the true source of its observed effect. Ultimately, this means that a neural-interface model should not even be created unless it is deemed likely that the precise scope (no more, no less) will believably produce the expected change without generating alternative explanations.

Principle 7: Deployable

Deployability, as a general property, refers to the ability of the intervention concept to accommodate needs of users without compromising functionality. In an idealized sense, this also means that the intervention concept should be able to tolerate distribution on a large scale with perfect fidelity. The long-term impact of neurotechnology is closely related to the ability to scale up, whereas neurotechnologies that produce an effect only if a set of idealized conditions is met are unlikely to be beneficial in real-world settings. However, deployability, like the other principles, is an aspirational goal. An intervention is not to be construed as inherently inferior or fatally flawed if distribution is not immediately (or perhaps even ever) possible. For example, there may be instances in which ease of deployability of a system may be necessarily low, such as in the case of gamma knife capsulotomy for obsessive-compulsive disorder (e.g., Kondziolka, Flickinger, & Hudak, 2011; Sheehan, Patterson, Schlesinger, & Xu, 2013), in which an extremely unique confluence of hardware, software, and intellectual expertise is required to deploy a targeted treatment. In these cases, the utility (Principle 2) of the intervention should be high enough to offset concerns about deployability (e.g., because no other viable alternative is available). Thus, there may be cases in which hardware and software designs are unique; however, this ostensible weakness is offset by the impact of the intervention. All else being equal, it is desirable for the intervention concept to survive distribution. If only one working group in the world has the required expertise to make a system work, all other principles that we have outlined here may have been met, but the engineering design may not confer the desired clinical or societal benefit because it cannot be distributed with any effect.

Summary of principles

In the foregoing sections, we presented a set of standards that can be used in both the development and the critical evaluation of neural-interface designs. The logical and perhaps inevitable courtship between clinical psychology and biomedical engineering has proceeded quickly enough that its union has outpaced the available guidelines for development, and we consider these principles to be timely design considerations that, when fulfilled, are likely to lead to neurotechnologies that will produce a lasting benefit. Although interrelated, these principles can be thought of somewhat sequentially hierarchically (e.g., falsifiability is critical in early development phases, but deployability can be a later consideration). In Figure 1, we apply these principles to a project our group is undertaking to develop and evaluate the feasibility of a real-time neurofeedback system to remediate deficits in facial-emotion recognition, using synchronizing fMRI and EEG, with adolescents and adults who have ASD. As depicted, for each principle, we propose specific benchmarks or goals related to this system.

OPEN IN VIEWER

Fig. 1.

Model showing application of the seven principles to an example of neurotechnology development. FER = facial-emotion recognition; rt-fMRI = real-time functional MRI; EEG = electroencephalography.

It is unlikely that any system will meet all of these principles, and we do not contend that an intervention system is inherently “poor” if it fails to meet all seven. Rather, these principles are meant to compose an aspirational heuristic for understanding the variables in the problem space that exists between clinical psychology and neural-interface designs and, in particular, for critical evaluation of such designs in the context of peer review. Because we conceptualize these principles as benchmarks toward an ideal course of the development of neural interfaces for clinical psychology, caution should be taken to not be too rigid in evaluation of new neurotechnology, lest we risk impeding progress. Virtually all technology starts out as expensive, complicated, and clunky to use. However, given sufficient demand coupled with innovativeness of the core idea, the ultimate application can exceed expectations (e.g., consider the early start of the personal computer).

Emotion recognition, expression, and regulation: Transdiagnostic processes

At the most basic level, socioemotional processes involve mental routines and their biological substrates that collectively support normative social interaction. Socioemotional competence can therefore be regarded as the capacity to correctly identify the subjective, internal states of other people as well as intentionally monitor and regulate one’s own behavior and affect in real time. Problems with emotion recognition and emotion regulation have been implicated in many forms of psychopathology (Gross & Jazaieri, 2014) and can be expressed in various ways, including a lack of awareness or appreciation of other people and their feelings, not responding appropriately to others’ emotional needs, not changing behavior to fit a social situation, and letting one’s own emotions negatively affect a social interaction. In particular, recognition and expression of emotions are interrelated processes that influence probability of successful social interactions (Nuske, Vivanti, & Dissanayake, 2013).

The ability to discriminate certain expressions develops early on in childhood. Babies as young as 3 months of age show the ability to distinguish happy, sad, and surprised emotions from static cues (Young-Browne, Rosenfeld, & Horowitz, 1977), and by 7 months of age, they are able to discriminate dynamic happy and angry faces (Soken & Pick, 1999). By 4 years of age, typically developing children can accurately verbally label most basic, prototypical emotions (Taylor, Bagby, & Parker, 1999). As previously mentioned, these processes are sometimes impaired, which makes social interactions and self-regulation difficult for affected individuals. Emotion regulation refers to an ability to modify one’s emotions to respond to situational demands and to meet one’s goals (Thompson, 1994). Problems with emotion regulation are expressed in many ways, including low frustration tolerance and reactive aggression.

Impaired emotion recognition and theory-of-mind abilities (i.e., inferring others’ thoughts and intentions) has been associated with childhood-onset conduct disorder (Donno, Parker, Gilmour, & Skuse, 2010; Fairchild, Van Goozen, Calder, Stollery, & Goodyer, 2009) and is present among children with high levels of psychopathic traits (Dadds, Masry, Wimalaweera, & Guastella, 2008). Developmentally, socioemotional competence has been shown to moderate the relationship between oppositionality in early childhood and later delinquency, such that young children with poorer socioemotional competence are more likely to progress to conduct disorder (Mandy, Skuse, Steer, St. Purcain, & Oliver, 2013). Impaired recognition of others’ facial emotions, particularly the expression of fear, has been shown to be associated with criminal behaviors among people with schizophrenia (Weiss et al., 2006) and bipolar disorder (Demirel et al., 2014), and there is a fairly robust association between inability to perceive fear via facial cues and antisocial behavior (Marsh & Blair, 2008). For many people with ADHD, poor social perspective taking, insensitive interpersonal approaches, emotion-regulation difficulties, and difficulty forming stable social relationships are life-course persistent problems (Kristensen et al., 2014; Marton, Wiener, Rogers, Moore, & Tannock, 2009). Impaired emotion recognition, especially in response to more subtle or fleeting cues (Rump, Giovannelli, Minshew, & Strauss, 2009; Tell, Davidson, & Camras, 2014; Uljarevic & Hamilton, 2013) and atypical cognitive processing of emotional stimuli (Eack, Mazefsky, & Minshew, 2014) has been shown among children and adults with ASD. Impaired access to, and outward expression of, one’s own emotions (e.g., Shalom et al., 2006) and impaired emotion-regulatory abilities (e.g., Mazefsky et al., 2013; White et al., 2014) have also been reported in ASD. It is clear that deficits in these domains are pervasive and transdiagnostic. They are also private, often not outwardly observable, processes not readily assessed with traditional behavioral approaches. For this reason, neurotechnologies might be especially useful for both assessment and rehabilitation of deficits in these socioemotional processes. In the following sections, we describe some of the research in these areas, although we note that the uncontrolled, correlational nature of the extant research prevents us from drawing inferences about causality or mechanisms of action. Nevertheless, such studies can offer indication as to viable courses to pursue in rigorous research to firmly establish directionality. After summarizing the relevant research, we address the potential utility of such systems for the study of problems in socioemotional functioning, specifically its relevance for better understanding of the mechanisms underlying such problems.

Assessment

There have been recent efforts to apply neurotechnology for assessment of serious mental illness. Traditional functional assessment, in a behavior-analytic sense (i.e., structured observation of antecedents, behaviors, and consequences), is often conducted to understand the factors maintaining a problem. Functional assessment offers a direct, unfiltered view of the client and the behavior or behaviors of interest, and it has been used successfully in many clinical contexts, such as understanding the causes of problematic behavior in individuals with limited speech or developmental disabilities (Jacobson & Holburn, 2004) and evaluation, and modification, of environmental factors that prompt and maintain occurrence of self-injurious behavior (Carr, 1977; Durand, 1987). However, structured functional assessment can be complicated for a variety of reasons, including ethical considerations (e.g., privacy of others in the setting), degree of stimulus control required (e.g., delivery of antecedent and consequent events needs to be precise), behavioral responsivity (e.g., the target behavior is exacerbated or squelched temporarily as a result of observation), and the impact of the clinician’s presence on test validity (e.g., the client alters behavior intentionally because he or she is being observed; e.g., Emerson, 1992; Iwata, Kahng, Wallace, & Lindberg, 2000). Perhaps most germane to use of neurotechnology in this regard, functional assessment may not be viable for phenomena that are not readily observable (e.g., actively interpreting a perceived emotion), given that these behaviors do not easily lend themselves to structured behavioral observation (Groth-Marnat, 2009; Iwata et al., 2000). In addition, use of technology may limit time burden on human assessors.

A technology-based functional assessment for socioemotional processes has the potential to overcome these limitations (Principle 2: Useful). For instance, virtual environments that closely simulate real-world situations (e.g., a classroom, a store) can be designed and used to determine what a person would do in the actual situation. Parsons, Bowerly, Buckwalter, and Rizzo (2007) found that children with ADHD demonstrated more commission and omission errors and more superfluous body movements (hyperkinesis), relative to children without ADHD, in a VR classroom. They also found group differences in behavior when distracters were introduced into the virtual environment. The investigators concluded that the virtual environment was able to effectively differentiate between participants with and without ADHD, and behavioral results from the VR classroom correlated with widely used ADHD assessment tools, including parent/clinician-administered tools and parent-completed questionnaires. These results provide preliminary support for application of neurotechnology in the assessment of socioemotional processes involved in psychopathology (e.g., emotional reactivity) and suggest that interactive, VR technology might be applied to functional assessment. Further evaluation is warranted to determine incremental validity (Principle 2: Useful), stability of results (Principle 4: Reproducible), and generalizability and ecological soundness (Principle 7: Deployable).

Assessments employing VR might be especially useful in assessing for emotion-regulation deficits, given that tools for direct measurement in clinical populations are lacking. Self-report of emotion regulation is likely insufficient on its own, given that it requires insight into one’s own emotions and the ability to label those emotions—skills that are often deficient in individuals struggling with effective emotional regulation. Observational approaches (e.g., frustration-tolerance tasks) are often used to supplement self-reports. Accurate assessment of emotion regulation should include assessment of the clients’ attention to their emotional state, their ability to describe the emotion, and their ability to differentiate the different emotions (Mazefsky & White, 2013). Neurotechnology may prove useful in allowing for these observations in a more controlled environment that targets the underlying mechanisms (Principle 5: Mechanism Driven). In summary, neurotechnology might provide a useful and hypothesis-driven approach to the assessment of emotional states, especially if the individual is not able to accurately self-report.

Intervention

With respect to rehabilitation of faulty socioemotional processes, there has been momentum in the development of computerized and psychosocial interventions to target facial-identity/emotion recognition (e.g., Faja et al., 2012; Tanaka et al., 2010). These programs often assess change in emotion recognition ability alongside more traditional “end-goal” behavioral outcomes (e.g., Dadds, Cauchi, Wimalaweera, Hawes, & Brennan, 2012; Lopata et al., 2010). There are also programs that target emotion-regulation development in children and adults (e.g., Goldstein et al., 2014; Linehan, 1993; Webster-Stratton & Reid, 2004). More general (i.e., not focused on emotion regulation per se) neurofeedback-based approaches for treatment of psychopathy or antisocial disorder have shown good outcomes as well (e.g., van Outsem, 2011). Park et al. (2011) found a VR-based treatment to develop social skills in adults with schizophrenia to be promising. These programs have, by and large, yielded promising findings with respect to clinical and behavioral outcomes.

Building on these targeted-treatment approaches, we believe the next challenge is to develop and apply client-interactive technologies to understand and remediate the socioemotional processes that underlie clinical symptoms or maladaptive behaviors (Principle 1: Verifiable and Principle 5: Mechanism Driven). Most such systems to date have been mouse operated rather than fully interactive directly with the user’s nervous system. For example, the user might indicate his or her response by clicking among a set of response options. With BCI, the user controls and interacts with the virtual environment neurally, without need for manual response, which provides several advantages. There is emerging literature on the application of real-time fMRI for neurofeedback in clinical disorders (Sulzer et al., 2013; Weiskopf, 2012). For example, neurofeedback in eating disorders may target regions identified in transcranial magnetic stimulation (Broadbent et al., 2011), such as the dorsal circuits that contribute to reward processing (Bartholdy, Musiat, Campbell, & Schmidt, 2013). Additional work has demonstrated that real-time EEG and fMRI neurofeedback can be helpful in the treatment of depression by harnessing power asymmetry in the EEG and activation in the amygdala when individuals report emotional memories (Zotev, Phillips, Yuan, Misaki, & Bodurka, 2014).

The potential of neurotechnology

There are several potential benefits of applying neurotechnology to the study and treatment of socioemotional processes involved in psychopathology. First, this approach allows the user to engage in a virtual social environment in which interactions and practice with emotion expression, recognition, and self-regulation can happen spontaneously and in which feedback occurs in real time. As previously noted, neurotechnology provides an opportunity to design assessment and intervention tools that do not depend on a person’s ability to respond to stimuli verbally or physically (Principle 7: Deployable). This is important when one works with young children but also with clients who are unwilling or unable to report on their feelings and experiences (e.g., those with intellectual or verbal impairments). A bottom-up approach (Principle 5: Mechanism Driven) to development of assessment and treatment approaches, in which the underlying mechanism or mechanisms are directly targeted rather than the manifest pathology, may be both parsimonious and clinically effective (e.g., Lerner, White, & McPartland, 2012). Neurotechnology has promise in helping us achieve this critical translation. By intervening at the deeper, underlying process, it may be possible to effect change at the manifest level, transdiagnostically. Neurotechnologies are promising in this regard, given that they allow tighter experimental control in efforts to intervene at the process (mediator) level of the problem.

Compared with many other forms of treatment (e.g., medication), it is conceivable that untoward side effects are minimal or maybe even nonexistent. From a usability perspective, neurotechnologies are promising for many populations (e.g., children) because they are portable and fun to use. In addition, the technology may be used to supplant or, more likely, augment more traditional in-person therapies, thereby reducing cost and promoting accessibility. They also allow a “safe” training environment. Unlike practice in a real social situation, the client/user can be freed of fear of negative evaluation by others (e.g., rejection because of poor facial recognition). The user can be completely independent and not reliant on a therapist or “coach” in the interactions. These qualities make it likely that clinically effective technologies will be highly translational and well disseminated. Finally, although perhaps most critical, such technology permits direct connection between the interface and central nervous system functioning. Owing to the tremendous advances made in clinical neuroscience during the past decade, we have the benefit of knowledge of several identified brain regions involved in socioemotional processes in the context of different forms of psychopathology.

Elucidating mechanisms of psychopathology

Neural mediators of emotion processing at different stages (e.g., detection of, response to, and interpretation of emotion) potentially include interactions between the amygdala, striatum, and the prefrontal cortex (see Monk, 2008, for review). These regions show structural and functional alterations in clinical populations with noted difficulties with emotion behaviors. For example, structural brain differences in amygdala volume are present in youth with an anxiety disorder (De Bellis et al., 2000; Milham et al., 2005), adults with depression (Sheline, Gado, & Price, 1998; Siegle, Konecky, Thase, & Carter, 2003), and individuals with ASD (Monk, 2008; Schuman et al., 2004). Similarly, although results from studies using different tasks demonstrate differential activation, studies often identify differences in amygdala, striatum, and prefrontal cortex activation in these populations compared with healthy control individuals. Several other brain structures, including subgenual prefrontal cortex and the orbitofrontal cortex, are also reduced in size in adults with depression (Öngür, Drevets, & Price, 1998). Functionally, several studies have shown greater amygdala activation in individuals with anxiety and depression relative to healthy control individuals (Monk, 2008).

These are just a few examples of how brain systems may function atypically in clinical populations; these examples illustrate that many of the same brain regions subserve socioemotional deficits across distinct clinical disorders. Thus, it is important to investigate the mechanisms involved in socioemotional deficits across disorders and domains. This transdiagnostic approach will allow us to better understand the mechanism or mechanisms behind socioemotional difficulties in clinical populations irrespective of a specific disorder. By having a clear understanding of how neural circuitry that subserves emotion is altered in clinical populations, interventions can be more readily available to aid these individuals in rehabilitation and perhaps recovery. Neurotechnologies may offer the potential to shape more adaptive brain functioning in terms of socioemotional deficits in psychopathology.