1.1.

Methodology and Searching Strategy in PubMed

We established a search strategy and include summary tables (see Tables S1–S3 in the Supplementary Materials) to provide a global idea of the state of fNIRS research on the three selected topics. To cover different approaches of NIRS, we used the following search terms for the methodology: [(near infrared) or (NIRS) or (optical imaging) or (diffuse optical tomography) or (EROS) or (fast optical signal)]. this was linked to the search terms for the targeted pediatric population [(neonates) or (infants) or (children) or (adolescents)]. these terms were linked to the search terms for each clinical condition (as specified in fig. 1 and in each section): (methodology) and (population) and (clinical condition). The search strategy and selection process for the three sections are illustrated in Fig. 1.

Fig. 1

Flowchart illustrating the selection of articles in the scoping review process. Search terms for the methodology and population ($n=12,933$ hits) were combined with search terms for the three pediatric conditions. The abstracts/titles were then scanned to exclude false and/or redundant hits. The remaining studies formed the basis for review, Secs. 2–4. They are listed in Tables S1–S3 in the Supplementary Materials.

2.1.

What has been Studied?

The PubMed review search yielded 175 hits. The search terms for this section were “epilepsy” or “interictal” or “epileptic spikes” or “spasm” or “seizure” or “refractory epilepsy.” After screening the abstracts to remove false and or redundant hits, 39 articles remained. Half were published before 2013. All articles are listed in Table S1 in the Supplementary Material. They can be grouped into three classes: (1) NIRS and interictal manifestations (epileptic spikes and epileptic discharges), (2) NIRS and epileptic seizures, and (3) NIRS and the lateralization/localization of language function in potential surgical candidates with refractory epilepsy. In the next sections, we will present an overview of the main results extracted from these papers.

2.2.

What are the Challenges?

Investigating epilepsy using simultaneous EEG/ECoG and fNIRS includes the challenges of multimodal and multidimensional data acquisition and analysis. From a technical point of view, it is necessary to further develop tools that allow for the simultaneous recording of the two modalities with complementary spatial (fNIRS) and temporal (EEG) resolution without electrical crosstalk between the two modalities (notably electrical artifacts on EEG arising from NIRS optodes). This requires the development of medical devices and caps that support both electrodes and optodes adapted to age and recording conditions, in intensive care for example, with the capacity of dealing with high-density configurations in both modalities. Although multidimensional (neuronal and vascular) data are challenging, combining EEG/ECoG and fNIRS makes it possible to address the dysfunction/function of several compartments: neuronal (EEG), vascular (fNIRS), and intercellular (FOS). fNIRS investigations in the context of epilepsy should always include a simultaneous EEG or ECoG recording and video monitoring to identify epileptogenic activity and for off-line artifact rejection (e.g., body movements, sucking, etc.). The challenge consists of the integration and simultaneous analysis of multidimensional information, which can be useful for a better understanding of the dynamics of the different sectors in the emergence of interictal spikes and seizures at the group level. The challenge for clinical applications is to also supply information at the individual level to improve individual care. This may necessitate high temporal and spatial resolution for simultaneous EEG/ECoG recordings. Another challenge is the assessment of neuronal networks that include nonsuperficial hubs. EEG/ECoG and fNIRS suffer from poor spatial resolution in deep structures, in particular, in children. Therefore, because epilepsy is a network pathology, the lack of information about dysfunction in deep structures is an important limitation. In addition, the lack of observed neurovascular coupling in interictal epileptic spikes (IES) or epileptic seizures may be related to the poor sensitivity of fNIRS or an insufficiently high density of optodes to extract the hemodynamic information. Nevertheless, EEG and NIRS provide information from two different compartments, neuronal and vascular. The relationships between these compartments have been largely debated (see Ref. 38 in pediatric epilepsy). Changes in neuronal synchronization (hypersynchronization in epilepsy) do not always imply an increase in metabolism, whereas an increase in the firing rate of a large population of neurons is likely to modify the energy needs in the region of epileptic discharge. Finally, the change in HbO within 5 s reflects an average of what occurs at the neuronal level within at least 1 s, knowing that the duration of hypersynchronization of the IES is $\sim 500\mathrm{ms}$, which is flanked by decreases in synchronization.³⁹ The resulting hemodynamic effect may be negligible.

2.3.

What are Potential Perspectives?

The main potential of fNIRS coupled with EEG lies in its ability to monitor the neuronal, vascular, and intercellular compartments. Technically, this requires medical devices that combine the two modalities at the sensor scale (electroptodes^® TM;⁴⁰). This also implies developing diffuse optical tomography strategies³⁷ that allow spatial resolution at the surface of the cortex comparable to that of 3T fMRI when using diffuse optical tomography with a high density of optodes.^37,41 Epilepsy is certainly a pathology linked to the endogenous hypersynchronization of populations of neurons that are propelled into a pathological spiral. However, this occurs in a particular hemodynamic and metabolic environment that should be monitored for both interictal epileptic spikes and seizures. Moreover, while EEG and ECoG only monitor the synchronous activation of well orientated pyramidal neurons, fNIRS makes it possible to evaluate the vascular dynamics resulting from neuronal activation of not only pyramidal cells or interneurons but also of glial cells. This does not only pertain to the activation of hypersynchronous neurons but also unsynchronized neuronal activation, provided that this is coupled to increased oxygen supply.⁴⁰ The concordance of information between the different compartments may converge at the individual level. For example, neuronal desynchronization and vascular deactivation in BECTS in the frontal regions could be used to evaluate the impact of epileptic spikes on transient cognitive deficits. By carefully monitoring such neurophysiological signatures, the impact of therapeutic management could be observed at a distance from the epileptogenic zone. Moreover, FOS makes it possible to monitor the dynamics of the extracellular space and therefore the availability of neurotransmitters at different levels of pathological neuronal networks. Improving our knowledge about such fine-tuning between different compartments could pave the way for novel therapeutic management and the development of new treatments. At the individual level, particularly in the management of drug-resistant epilepsy in children, this would allow for a better definition of the epileptogenic zone to be resected, as well as of the eloquent zones that need to be protected. These measurements can be easily performed repeatedly at the patient’s bedside through the scalp and in the operating room directly from the cortex. Thus, despite the relatively small number of publications in the field of childhood epilepsy, fNIRS coupled with EEG is still an interesting approach that could lead to clinical developments, including high-density strategies and real-time signal processing tools.

3.1.

What has been Studied?

The scoping review search yielded 167 hits. The search terms used in this section were: ((speech language impairment) or (stutter) or (dyslexia) or (dysgraphia) or (reading difficulties) or (speech impairment) or (hearing impairment) or (deafness) or (cochlear implant) or (sign language)). After screening the abstracts to remove false and or redundant hits, 40 articles remained, which are listed in Table S2 in the Supplementary Material.

Since auditory input is provided from birth, and even intrauterine auditory signals may shape precursors of language development,⁵³ severely impaired hearing will strongly affect later language development. It is thus notable that the largest number of pediatric fNIRS studies have focused on the effect of cochlear implants (CIs) on hearing and the perception of spoken language. The first study⁵⁴ demonstrated the high detectability of a bilateral fNIRS-response to segments of auditorily presented stories. Although largely a feasibility-study, detection of an fNIRS response in 82% of normal-hearing children and 100% of hearing adults is a good starting point to investigate alterations of the fNIRS response in CI-users. Further advantages of the study’s design for potential clinical use were an undemanding sparse-channel approach and a state-of-the-art analysis technique allowing for robust detection of cortical activation. Other studies have confirmed the feasibility of this approach and have addressed cross-modal plasticity as a factor in success of CI users to adapt to the device. As no physical or artifact-interference occurs with fNIRS in CI-users, its suitability for the study of this population has been highlighted in five review papers. However, the only study truly addressing potential clinical relevance is that of Wang et al.⁵⁵ Using a longitudinal design with two sessions within 6 months after unilateral CI-implantation, the study showed no difference between the implant side in terms of the fNIRS response to linguistic input, but an advantage for left-CI-implantation for the perception of nonlinguistic material. This is consistent with established models that show that speech comprehension depends on bilateral perisylvian cortices.⁵⁶ The authors concluded that the generally recommended right-CI-implantation may not be justified if only unilateral CI-implantation is possible due to limited resources or medical reasons. Such a direct clinical recommendation is remarkable in the field, as most studies have described activation patterns without a direct consequence on clinical decisions.

Concerning language perception and comprehension, a number of studies have confirmed the potential of fNIRS to detect a response in infants with developmental deficits and in prematurely born children. An impressive example of the use of fNIRS in very early language development is provided by a study showing that even in children prematurely born three months before term, a linguistic (phonemes /ba/ versus /ga/) and voice contrast (male versus female) are differentiated by the premature infant’s brain.⁵⁷ In an interesting follow-up study, the group was able to show that the EEG response to the contrast could be elicited in preterm infants who suffered from an intraventricular hemorrhage, whereas a hemodynamic response was not detected, showing deviations from typical neurovascular coupling in this pathology.⁵⁸ This and other studies underline the high versatility of fNIRS in research on early language development. However, more clinically oriented applications targeting later language development are scarce. This is partially due to the wide range of tasks that have been used, targeting the many facets of emerging linguistic competence. It may be necessary, however, to establish “standard” stimulation procedures to assess how the individual child’s brain processes language. This may allow for the use of neurophysiological markers, including fNIRS measurements, in clinical research on the diagnosis, prognosis, and/or therapy-related changes of (imminent) speech-language impairment. Established and simple linguistic contrasts should be used⁵⁷ and future research may profit from findings using event-related potentials (ERPs) to describe milestones in developmental neurolinguistics.⁴⁸ A number of authors have highlighted this limitation in the literature reviewed here. Most prominently, a study assessing the validity of typical baseline conditions for auditory language tasks used in fMRI and EEG (time-reversed speech, TR, or noise correlated to the speech signal, SCN) showed that they may not be suited to fNIRS experiments.⁵⁹ The authors tested 25 children of $\sim 9$ years of age and compared fNIRS responses to spoken sentences to responses to TR or nerve conduction velocity. Although a response over auditory areas was successfully measured in most children, no lateralization or a statistically significant difference between speech and either of the nonspeech baselines was found. Although other researchers successfully used these baselines (e.g., for TR⁶⁰) the study highlights potential limitations when stimulation paradigms from other imaging modalities are transferred to fNIRS research.

For speech “production,” several studies have addressed alterations in the cortical activation pattern in individuals who stutter. During development, 5% to 8% of preschool children develop such speech dysfluencies, but stuttering persists in only $\sim 10\%$ of these children. Therefore, a major challenge in this field is to predict whether a child who stutters will or will not spontaneously recover. While most fNIRS studies have described the altered cerebral activation patterns in children who stutter relative to their nonaffected peers, Hosseini et al.⁶¹ showed that such a challenge can also be addressed by studying neurophysiological signals supplied by fNIRS. Using data from a previous study,⁶² the group developed a classifier that detected children who stutter in 87% of cases when based on fNIRS data. The classifier was trained on the data of 46 children, including children who stutter, children who showed no speech disfluencies, and a group of children who had recovered from stuttering. Eighteen channels covering the bilateral superior temporal and inferior frontal areas were used for all children. Of interest in the context of the current paper, application of the classifier to children who recovered from stuttering classified them as “no-stutter.” Although this may simply reflect a state difference at the time of the fNIRS measurement, it is a first promising step toward applying such a classifier to children who stutter and correlating it with a later outcome (recovery versus persistence of stuttering).

A small number of studies have addressed “dyslexia.” Although one of the most common developmental disturbances, the published papers using fNIRS are methodologically weak (see Table S2 in the Supplementary Material for details of the studies). For example, one study selectively measured oxygenation changes over a left prefrontal area to claim that the differential oxygenation response in dyslexic children relative to a control group indicates deficient working memory.⁶³ Working memory depends on a large network and changes in the fNIRS response cannot be a surrogate marker for the assessment of such a basic cognitive function. Similarly, the claim that the difference in oxygenation response “causes” the working memory deficit is not justified. Studies targeting network functions, such as working memory, need to measure more than one of the “cortical hubs” and should be very careful to not confuse correlational and causal relationships when interpreting the data.

In summary, fNIRS has been successfully used to demonstrate its applicability in various developmental disorders/disturbances of language and communication disorders. Recent studies mostly targeting feasibility and/or questions concerning the underlying pathophysiology have attempted to explore the prognostic options of fNIRS measurements in children with CIs and those who stutter. In the next section, we will highlight the respective challenges for such applications.

3.2.

What are the Challenges?

Targeting diagnostic or prognostic functional imaging markers of the developing brain is like walking on shifting sand. This is true for the assessment of all cognitive functions and clinical populations, including children with language and communication disorders and those with attentional disorders, as discussed in the following section on ADHD (Sec. 4). Due to the constitutive high plasticity in this age range, a specific contrast may only be indicative of later development within a very narrow age window. Similarly, the concept of “critical periods” during language and cognitive development assumes that infants may easily acquire specific skills in a specific age window and will have great difficulties to do so at later stages of development. Concerning language, an impressive, albeit debated, example is the “vocabulary spurt”^64,65 referring to the exponential growth of the word knowledge of children starting at around 20 months of age. Although such “windows” may exist, it is also important to acknowledge the large interindividual variance. In summary, longitudinal assessment may be necessary. The ERP literature has demonstrated that electrophysiological “milestones” of language development can be defined.⁴⁸

One of the challenges of using fNIRS to study language and communication is the movement artifacts induced when the child is performing an expressive language task. The choice of method for the identification and correction (or rejection) of movement artifacts is important and may differ between the type of artifact and populations. Careful signal inspection is, therefore, important and necessary.

A study also investigating speech perception highlighted the caveat concerning the use of fNIRS without adequate control for extracerebral signal contribution (see Ref. 66 for a demonstration of the stimulus-locked autonomic response in adults). In this study,⁶⁷ the changes in heartbeat were extracted from recordings of a 44-channel fNIRS monitor of infants (3 to 12 months of age) with a high ($n=40$) versus low risk ($n=48$) of having autism spectrum disorder (ASD). The results showed group differences between baseline heart rates. More importantly, the heart rate response to stimulation (three-syllabic pseudowords) also differed between the two groups. This finding may be valuable for research on the physiological differences between high- versus low-ASD-risk children. However, it also highlights that systemic hemodynamic parameters can be stimulation-correlated and need to be corrected when functional cerebral activation is targeted.

3.3.

What are the Potential Perspectives?

In terms of developmental disorders that affect language and communication, fNIRS may have the potential to improve prognostic evaluation. A “diagnostic” function, in the sense of providing a measurement to decide whether a disorder is present, appears to be currently unrealistic. It should be noted that this pertains to all functional imaging tools, as behavioral measurements in the form of standardized and psychometrically validated tests must be considered as the gold standard. (In other words: a child would not be considered dyslexic if his/her brain shows an abnormal activation pattern but rather if a test for reading/writing shows results outside the normative range). For the development of language and communication, even behavioral screening may be underused,⁶⁸ especially for adolescents, making the more demanding assessment by functional imaging an unrealistic goal in a broader population. Therefore, a realistic target for future research is to use fNIRS in concert with other assessment tools to eventually sharpen the diagnostic criteria and prognosis for “at-risk” populations. For example, such predictive validity of ERPs has been shown for dyslexia.^69,70 The diagram in Fig. 2 illustrates potentially fruitful designs that combine standardized assessment at an early age and correlating the results with later development.

Fig. 2

Potential design for supporting the prediction of the occurrence (e.g., dyslexia) or persistence (e.g., stuttering) of a neurodevelopmental disability, clinical diagnosis (e.g., ADHD), or a response to a pharmaceutical treatment. fNIRS measurements could be a component of a multimodal assessment strategy associated with (later) clinical-behavioral testing to diagnose the condition. Longitudinal assessment is mandatory for such an approach.

In summary, the establishment of new clinically reliable tools for individual diagnosis/ prognosis is very rare. However, fNIRS may have the potential to contribute to group-wise predictions concerning the development or persistence of developmental deficits affecting language and communication. One great advantage of this methodology, especially in developmental language research, is the ease with which it can be combined with EEG and the fact that “natural” environments can be assessed. The option of simultaneously scanning multiple participants (hyper-scanning)⁷¹ may also be of interest to capture abnormalities in communication and social interaction. Aside from the conditions discussed in this section, this also applies to the following section on ADHD. This option has not yet been used in a clinical setting. Fascinating as it is, the challenge lies in the potential increase in variability between participants. To explore the clinical perspectives, however, large-scale studies ideally using similar paradigms for different languages and a correlation with later outcome appear to be the most promising avenues. Recent studies have documented the first steps in this direction for those who use CIs and the question of whether developmental stuttering will persist or resolve in later life.

4.1.

What has been Studied?

The scoping review search yielded 151 entries. The search terms used in this section were: [(attention deficit disorder) OR (attention deficit) OR (ADD) OR (ADHD)]. All titles and abstracts were scanned and false and or redundant hits were removed. Fifty-two original articles remained and were reviewed in detail. For a summary of the main information related to these articles, please refer to Table S3 in the Supplementary Material. Among the 52 fNIRS papers on ADHD and children/adolescents, almost half (24/52) used fNIRS to better understand the pathophysiology of the disorder, only seven (7/52) used fNIRS for the screening or diagnosis of ADHD, and 21 used fNIRS to assess the impact or efficacy of an intervention, mostly pharmaceutical treatments. The main findings and global conclusions revealed by this body of literature are reported below.

4.2.

What are the Challenges?

There are several challenges and limitations associated with fNIRS research in ADHD. First, there is large heterogeneity between studies in terms of diagnostic criteria and the inclusion or exclusion of comorbidities. The diagnostic criteria for ADHD and its subtypes (i.e., inattentive, hyperactive/impulsive, combined) were most often not mentioned, and some studies included children with comorbidities (e.g., ASD), whereas others excluded them. This makes it difficult to compare the results between studies. In most publications, the sample sizes were relatively small, probably due to recruitment issues and the difficulty of acquiring good quality data from children with ADHD. These limitations preclude researchers from controlling for important sociodemographic and clinical variables and from drawing strong and generalizable conclusions. Although many studies made the effort to match groups for sex and age, most studies with nonmatching groups did not control for it in the analyses, leading to bias and even invalid interpretations in some cases. Furthermore, there have been no fNIRS studies on the effect of sex, age, or gender in children or adolescents with ADHD, which would be of interest in clinical care.

An important methodological limitation in most ADHD publications (but not limited to this clinical entity) is the highly limited scalp coverage. Most studies used probes placed only over the bilateral frontal areas, although some also included additional coverage of the temporal and/or parietal regions. Although the established impact of ADHD on executive functioning may justify focusing on the frontal/prefrontal areas, such narrow coverage precludes researchers from taking a large-scale network approach to study the spatial specificity underlying the pathophysiology of this disorder. This may be of particular relevance when targeting the effect of medication on brain hemodynamics for drugs that affect larger brain networks rather than only frontal brain regions, which was not brought to light in these articles. The underlying brain network of ADHD also probably involves deep cortical and subcortical structures that cannot be detected using fNIRS due to the shallow penetration of light, highlighting a general limitation of the technique in the investigation of pathological mechanisms, especially when they involve large-scale networks that support cognitive functioning.

4.3.

What are Potential Perspectives?

Despite the above-mentioned limitations, we believe that the current fNIRS literature on ADHD offers several promising avenues for the development of approaches with a potential clinical impact. Using machine learning algorithms, fNIRS data combined with behavioral, clinical, and socio-demographic data offers relatively good diagnostic accuracy for ADHD (over 80%). fNIRS will surely not replace clinicians and behavioral testing for diagnosing ADHD, but it could eventually contribute by providing a computer-aided diagnostic tool to clinicians. This will be especially relevant in cases in which the clinical presentation is not typical or clear or in contributing to better defining the type of ADHD. Similar to its application in developmental disorders that affecting language and communication (see Sec. 3), the prediction of later development and the prognosis may be a prime target for fNIRS, ideally in concert with other neurophysiological and behavioral measurements (see Fig. 1). In terms of limitations, this literature is still young and the small cohort sizes included in most of the studies clearly call for an fNIRS database that includes typically and atypically developing children to build normative developmental fNIRS data. Another facet of the literature suggests that fNIRS may provide predictive markers for patients who will have a good response to medications, leading to a personalized medicine approach that could eventually be implemented in clinical practice. Not least, fNIRS-neurofeedback has been suggested as a promising nonpharmaceutical tool for the treatment of ADHD. However, only a nonrandomized parallel-group study and a pilot study have been published thus far. More evidence is still needed before considering offering fNIRS-neurofeedback in the clinical setting.