(i) Studies applying fNIRS;

(ii) Studies including a clinical population with a confirmed diagnosis (according to DSM-5) or a well-established condition of risk for NDDs;

(iii) Studies including preschool children (sample mean age < 7 years).

(i) Articles published in languages other than English;

(ii) Meta-analyses or literature reviews;

(iii) Other reasons referring to report characteristics, as not published clinical studies (e.g., conference paper, thesis, book chapter) or studies that do not involve human participants (e.g., focused on methodological issues).

One hundred and sixty-three potentially relevant studies were identified from PubMed, 136 from Scopus, 127 from Web of Science; in addition, four articles that met the selection criteria were retrieved from screened review articles. After omitting duplicates, 224 articles were examined. On the basis of the title, a total of 52 studies that did not meet the inclusion criteria were disregarded. Finally, within the172 articles screened for abstract or full-text evaluation, a total of 13 studies were selected for this review. An in-depth detail of paper selection is shown in the PRISMA diagram (Figure 1).

Crucial information was extrapolated from included papers. Tables 1, 2 summarize the mean age of participants, the sample size and gender ratio, the experimental procedures (technical aspects of recording sessions and experimental design) used during fNIRS measurement.

Although the initial aim was to include different type of NDDs (see keywords used for the search), only studies focused on children with idiopathic ASD or high-risk (HR) infants (infants at high familial risk of autism) satisfied the inclusion and exclusion criteria.

Regarding other NDDs, studies on ADHD included children older than 7 years, due to the pharmacological aim of their experimental design: at the time of our last search, indeed, 55 studies focused on fNIRS and ADHD, but none of them was performed on the preschool population. Moreover, within the growing bulk of literature that uses fNIRS for evaluating the emerging language ability in infants, only two studies (Chang, 2014; Hosseini et al., 2018) focused on children with Language Disorders; however, both of them, did not meet the inclusion criteria (mean age of participants older than 7 years) and were therefore excluded from this review. fNIRS was rarely used to explore neuropsychological features in learning disorders such as Dyslexia (Zhu et al., 2012; Pecyna and Pokorski, 2013; Song et al., 2013; Sela et al., 2014; Cutini et al., 2016), but this field obviously includes only school-aged children. Application of fNIRS to Intellectual Disability (ID) is only anecdotal yet (only one scientific report found in literature database consulted) (Bembich et al., 2021).

As mentioned above, all studies included in this review applied fNIRS in preschool children with ASD, a clinical condition characterized by socio-communicative impairment associated with restricted interests and repetitive behaviors, or HR infants (younger siblings of children with ASD). Only four studies included more than 40 participants (Keehn et al., 2013; Jia et al., 2018; Li and Yu, 2018; Li et al., 2018); in most cases subjects were equally distributed between the clinical population and the control group. Interestingly, the cohorts included both males and females in all studies, but the two populations were not homogeneously distributed in terms of sample size, except for two investigations (Lloyd-Fox et al., 2018; Pecukonis et al., 2021).

Five studies included in this review (Kikuchi et al., 2013; Li and Yu, 2016, 2018; Jia et al., 2018; Li et al., 2018) comprised ASD children with a mean age was between 3 and 7 years and compared the ASD group with age- and gender-matched typically developing (TD) peers in resting state condition (see Table 1). It is to note that all selected studies included children with an average IQ score and no comorbidity with ADHD.

The remaining eight studies (Fox et al., 2013; Keehn et al., 2013; Lloyd-Fox et al., 2013, 2018; Edwards et al., 2017; Braukmann et al., 2018; Bhat et al., 2019; Pecukonis et al., 2021) focused on infants/toddlers (children younger than 3 years old), and compared fNIRS data between HR group and in Low-Risk groups (LR; condition defined as subjects with no family history of ASD) in task-related study designs (see Table 2). Of these studies, only two (Lloyd-Fox et al., 2018; Pecukonis et al., 2021) analyzed the correlation between the fNIRS brain responses of infants during the first year of life and their diagnostic outcome at 24 or 36 months.

A total of four different fNIRS recording systems were employed in the reviewed studies (Foire 3000, LABNIRS, Hitachi ETG-4000 system, UCL-NIRS topography system). The number of channels used ranged from a minimum of 2 (Kikuchi et al., 2013) to a maximum of 44 (Li and Yu, 2016, 2018; Jia et al., 2018; Li et al., 2018). The remaining investigations used an intermediate number of channels (24–26 pairs of optodes), but the density of montages may be considered comparable among almost all studies, due to the younger age of the cohort investigated and the relatively smaller dimension of the probe. Despite the variability of probe geometry, in all cases optodes placement followed the international 10–20 EEG system.

Within the 13 articles included, eight studies explored the brain activation evoked by specific tasks (task-related paradigm or T-R; an overview of experimental procedures is reported in Table 2) and the remaining five investigated functional brain organization in resting-state (R-S) condition (an overview of experimental procedures is reported in Table 1). Considering the significant difficulties to remain still under stimulation-free conditions for young children, all resting-state data were recorded while participants watched a cartoon or, only in one case, a picture-card show (Kikuchi et al., 2013) in order to optimize subject compliance. Papers exploring functional cortical activation in task-related condition can be sub-grouped into two categories that reflect the nature of the task proposed: (i) social stimuli, including social perception (Lloyd-Fox et al., 2013, 2018; Braukmann et al., 2018), face processing (Fox et al., 2013) and naturalistic social interactions (Bhat et al., 2019); (ii) language-related stimuli (Keehn et al., 2013; Edwards et al., 2017; Pecukonis et al., 2021).

Resting-state fNIRS measurements have been proposed as auxiliary indexes for the objective assessment of ASD in both young (Xu et al., 2021) and adult population (Yanagisawa et al., 2016).

Regarding preschoolers, Kikuchi et al. (2013) first demonstrated an aberrant functional connectivity between the right and the left anterior prefrontal cortex (aPFC) in young children with ASD under resting state conditions. Notably, they found a significantly higher inter-hemispheric connectivity in the ASD compared to the TD group, reinforcing previous data about the aberrant cortical organization at structural level in ASD children (McCaffery and Deutsch, 2005; Courchesne et al., 2011). The higher inter-hemispheric connectivity in ASD reported by Kikuchi et al. (2013) was positively correlated with the severity of social deficit, as scored with the Autism Diagnostic Observation Schedule (Lord et al., 2012). It is to note that the anterior prefrontal cortex has already been implicated in ASD pathophysiology in previous structural and functional studies, as being involved in social perception and processing abilities that are altered in the ASD condition (Uddin et al., 2013; Pagnozzi et al., 2018).

In contrast, studies from Li and Yu (2016, 2018) and Li et al. (2018) showed weaker functional network efficiency in young children with ASD compared to TD peers, mainly in short- and long-range connectivity of the right prefrontal cortex (Li and Yu, 2016). Moreover, a subsequent study of the same research group (Li et al., 2018) showed an inverse association between network efficiency, age (especially evident in the Deoxy- and TotHb-based networks) and severity of autistic behaviors (especially evident in the OxyHb-based-network) in high-functioning ASD subjects. Statistically significant findings were also obtained when comparing ASD and TD groups through spatial complexity of functional connectivity (SCFC) analysis (Li et al., 2018).

Other approaches to analyze resting state recordings also confirmed that fNIRS might be a candidate tool to investigate the pathophysiological mechanism of ASD, looking at the temporal dynamics of neuronal oscillations (Jia et al., 2018). Specifically, the long-range temporal correlations (LRTCs) of hemoglobin concentration signals were studied by quantifying a detrended fluctuation analysis (DFA) exponent. Comparing data between ASD and TD group, significant differences were found over the left temporal region for OxyHb signal, and over bilateral temporo-occipital regions for DeoxyHb signals, suggesting a shift-to-randomness of brain oscillations in children with ASD. Moreover, the relationship between age and DFA exponents revealed that this association could be modulated by autism. Furthermore, the DFA exponents of OxyHb in the left temporal region were negatively correlated with autistic symptom severity. Between-group differences remained significant also looking at correlation coefficients between age and DFA.

Altogether, these results confirm that fNIRS can detect a different pattern of functional connectivity in the brain of ASD children. Accordingly, the considerable bulk of prospective studies focused on neurostructural and neurofunctional measures of brain development validated the concept of ASD as characterized by early altered connectivity patterns (Bosl et al., 2011; Wolff et al., 2012; Solso et al., 2016). In particular, early social brain network alterations were previously reported in structural (Shen et al., 2018) and functional MRI (Emerson et al., 2017) as well in observational studies (Elsabbagh and Johnson, 2016). However, no resting-state fNIRS data are yet available in cohorts younger than 3 years, thus restricting the predictive value of resting state analyses so far. Moreover, the mathematical algorithms used for the interpretation of resting-state data are very complex to be applied in the clinical setting and the limited coherence among models still prevents from drawing conclusive interpretations.

Younger siblings of children with ASD are considered as HR infants because around 20% of them receive a diagnosis of ASD at the age of 3 years (Ozonoff et al., 2011) and a further 20–30% exhibit other neurodevelopmental problems (Messinger et al., 2013). In addition, a high rate of behavioral traits overlapping with those observed in children with an ASD diagnosis (referred as the Broader Autism Phenotype) has been documented in the sibling population (Charman et al., 2017). Thus, the prospective study of the HR population allows detecting behavioral risk signs or biomarkers of NDDs at a very early age, prior to the full clinical manifestation (Zwaigenbaum et al., 2015).

Studies on younger siblings of children with ASD compared HR infants with infants without familial history of ASD or NDDs (LR infants), regardless their developmental outcome or differentiating HR infants with or without a later diagnosis of ASD. In this review, we included both types of studies as we are interested in understanding how fNIRS signal can be used as a potential biomarker of atypical development, and not only as an ASD related outcome.

In the past decade, the number of studies exploring the applicability of fNIRS technique in HR infants has rapidly grown, focusing especially on core ASD neurocognitive domains. Accordingly, our study selection includes five papers testing social perception and three others assessing possible alterations of language development in the HR group.

With regard to social processing, starting from previous data in typically developing infants (Lloyd-Fox et al., 2012), Lloyd-Fox et al. (2013) examined whether the temporal lobe specialization for processing visual and auditory social stimuli during the first months of life differed between HR and LR infants. They found significantly diminished evoked hemodynamic responses in HR compared to LR infants in the superior temporal sulcus (STS), both in visual and auditory trials, suggesting a lack of cortical specialization to social stimuli, already within the first 6 months of life. Notably, a more recent prospective study in the same cohort of HR infants (increased by only 2 more subjects) demonstrated that the reduced activation to visual social and vocal stimuli across cortical regions of interest was specific for infants with a later diagnosis of ASD (Lloyd-Fox et al., 2018).

Moreover, Braukmann et al. (2018) provided further evidence for a social processing difference in infants at risk of autism, highlighting a reduced hemodynamic response evoked by social visual stimuli in the right posterior temporal cortex. Similarly, a reduced bilateral cortical activation in HR children compared to LR infants was reported in the social perception within the infant-mother dyad (Bhat et al., 2019). Of note, this is the only study using a fNIRS experimental protocol with live stimuli (face-to-face naturalistic social interaction between infant and mother).

Social behavior was also explored by Fox et al. (2013), recording fNIRS hemodynamic responses to a face perception task (facial identity and emotion perception on video clips performed by infant’s mothers) during the first year of life. This study suggested differences in patterns of functional connectivity in the frontal lobe (short- and long-range connections) between the HR and LR groups. HR infants showed significantly greater DeoxyHb responses in frontal channels and lower OxyHb responses in right temporal regions during face processing task; a significant difference between groups was detectable in response to mother’s face as to a stranger face (greater DeoxyHb response within the HR group across both frontal and lateral regions). For this cohort, no longitudinal data were available for discriminating infants with or without later ASD diagnosis and establishing the predictive value of early fNIRS recordings.

To summarize, these results suggest that socially evoked fNIRS signals might be a suitable and predictive biomarker in early assessment of HR infants; however, further studies are needed to confirm these promising findings.

Growing evidence on the validity of fNIRS for studying early aberrant functional brain networks in HR infants emerged from the language domain as well.

Keehn et al. (2013) explored functional connectivity during the first year of life in HR infants using a task-related experimental paradigm (language processing task). They found that from 3 to 12 months HR children exhibited a pattern of decreasing connectivity with reduced intra-hemispheric connectivity for the left hemisphere compared to LR infants at 12 months.

Moreover, an alteration of language processing can be detected very precociously with fNIRS: Edwards et al. (2017) analyzed precursors of language development in HR and LR 3-month-old infants: they found that HR females exhibited significantly higher OxyHb responses to auditory stimuli containing syllable repetitions than HR males, LR females or LR males in both left and right anterior brain regions. It is worth noting that HR females showed different temporal response patterns with respect to LR females across different type of auditory pattern (repeating vs. non-repeating syllabic sequences) presented randomly (speech-like stimuli), indicating a potential gender endophenotype in ASD. Using the same experimental protocol, Pecukonis et al. (2021) performed a longitudinal study on functional specialization of language-related brain regions during the first year of life in HR and LR group, suggesting a possible role of language-evoked fNIRS measurements to predict the diagnostic outcome of infants at 24 months.

Altogether, these data demonstrate that fNIRS studies on siblings can provide a unique window into the earliest neurobiological atypical trajectories and have the potential to shed light on possible ASD-related endophenotypes and biomarkers of brain function. In particular, studies correlating fNIRS brain responses in the first year of life to the diagnostic outcome at 24–36 months (Lloyd-Fox et al., 2018; Pecukonis et al., 2021) highlight that atypical brain responses are detectable in ASD infants as early as 6 months, thus indicating that fNIRS might be the an useful tool to early predict ASD.

Theoretical speculations about possible left-hemisphere dysfunction (Fein et al., 1984) or predominant right-hemisphere impairment (Happé, 1999) have historical roots in autism research. Many reports of atypical hemispheric asymmetries in ASD came from anatomical (Herbert et al., 2005) and functional imaging studies (Kleinhans et al., 2008), and typically focused on the abnormal lateralization of language domain (Just et al., 2004), even at very early stages of infants’ development (Eyler et al., 2012). Nevertheless, it is only in the last decade that attention on functional asymmetries related to non-verbal processing and the possibility to consider atypical brain asymmetry as a candidate for clinically meaningful stratification in ASD has grown (Floris et al., 2021). Indeed, recent evidence from resting-state fMRI studies supported the hypothesis that atypical rightward asymmetry shift may be a pervasive feature of functional brain organization in ASD, affecting not only the language network, but also the sensorimotor function and higher cognitive domains (Cardinale et al., 2013). Task-related – fNIRS studies focusing on hemispheric asymmetry in ASD individuals are still very limited and evidence reported are quite inconsistent (Doi and Shinohara, 2017). No systematic findings on hemispheric asymmetry in ASD population during resting-state fNIRS recording are currently available. Even in the studies included in this review, hemispheric asymmetry was not highlighted as a primary research purpose, allowing us to prompt only a few speculations on this topic.

Functional near-infrared spectroscopy studies on language perception and processing seem to confirm previous literature about ASD atypical lateralization for the language domain, with a reduced functional connectivity in the left hemisphere of HR infants compared to the LR peers, both at 6- (Pecukonis et al., 2021) and 12-months of age (Keehn et al., 2013). In contrast, Edwards and colleagues (Edwards et al., 2017) found left- or even hyper-lateralization of neural activity in 3-month-old HR females in response to a language auditory task, speculating about a gender-specific ASD endophenotype related to language processing. However, the limited spatial resolution of fNIRS prevent these results from being conclusive.

Similarly, task-related fNIRS studies of social processing were not consistent in defining hemispheric asymmetry. Indeed, Lloyd-Fox et al. (2013, 2018) found cortical activation patterns in response to social vs. non-social stimuli showing maximal differences between HR and LR groups in the left STS region. Moreover, Braukmann et al. (2018) detected different evoked responses in the right temporal region between groups. In contrast other studies, despite reporting cortical hypoactivation in HR both in naturalistic social interactions (Bhat et al., 2019) and face processing (Fox et al., 2013), did not highlight a clear hemispheric asymmetry of responses.

Finally, resting-state fNIRS studies found a hypoactivation with rightward asymmetry shift in the prefrontal cortical area of ASD preschool children (Li and Yu, 2016; Li et al., 2018). Only two studies reported ASD-related alterations bilaterally (Kikuchi et al., 2013; Jia et al., 2018). However, these results could be influenced by technical aspects, such as the different number of channels used.

Even if no systematic assessment of hemispheric asymmetry has been performed with fNIRS in preschoolers, these findings and evidence from other neuroimaging technique such as structural (Conti et al., 2016) and functional MRI (Müller et al., 2011) would encourage studying more in depth atypical lateralization in ASD population, also at early stage.

Over the last years of research, the biological dimension of ASD has been extensively investigated both in the preclinical (Ebert and Greenberg, 2013; Del Pino et al., 2018) and the clinical settings (Billeci et al., 2016; Bralten et al., 2018). However, most studies only focused on neuronal mechanisms underlying ASD.

Preclinical studies on animal models can offer important insights about specific mechanisms underlying the pathophysiology of ASD (Bartz et al., 2008). Recent studies suggested the presence of structural and functional neurovascular abnormalities in ASD. Indeed, an early dysfunction of endothelial cells, altered angiogenesis and impaired vasodilation reactivity have been shown in a mouse model of 16p11.2 deletion ASD syndrome (Ouellette et al., 2020). Similarly, cerebrovascular deficiencies have been recognized in a mouse model of Creatine Transporter Deficiency, an inherited metabolic condition characterized by intellectual disability and autistic-like features (Mazziotti et al., 2020). The vascular hypothesis is also supported by clinical studies. Indeed, postmortem analysis of ASD brains suggested an impairment of cerebral angiogenesis (Azmitia et al., 2016) and resting-state imaging highlighted the alteration of cerebral blood flow in distinct brain regions (Jann et al., 2015; Bjørklund et al., 2018). Interestingly, it has been speculated that an exacerbated inflammatory response might be one of the possible causes of the deteriorated integrity of the blood-brain barrier (Fiorentino et al., 2016; Kumar and Sharma, 2016) and of the defective blood flow documented in the ASD brain (Morgan et al., 2010; Bjørklund et al., 2018). Moreover, there is evidence of glial dysfunction in ASD (Petrelli et al., 2016), with the alteration of astrocyte population (Fatemi et al., 2008; Edmonson et al., 2014; Wang et al., 2021) potentially contributing to the deregulation of neurovascular homeostasis.

We believe that the potential vascular contribution in the etiology of ASD should be explored more in-depth, since the interplay with the vascular system is crucial for the proper maturation and function of neuronal networks (Andreone et al., 2015; Segarra et al., 2018). In this framework, the combined analysis of hemodynamic responses measured with fNIRS and bioelectrical signals obtained through EEG recordings might allow dissecting vascular vs. neuronal aspects of ASD.