IEEE Computational Intelligence Magazine - February 2022 - 17

in brain) analysis paradigms is of significance
as it allows investigations into the interactions
between different brain regions. However,
despite the potential of multivariate analysis to
shed light on the interactions between developing
brain regions, artificial intelligence (AI)
techniques applied render the analysis nonexplainable.
The purpose of this paper is to
understand the extent to which current stateof-the-art
AI techniques can inform functional
brain development. In addition, a review of
which AI techniques are more likely to explain their learning
based on the processes of brain development as defined by
developmental cognitive neuroscience (DCN) frameworks is
also undertaken. This work also proposes that eXplainable AI
(XAI) may provide viable methods to investigate functional
brain development as hypothesized by DCN frameworks.
I. Introduction
H
uman brain development is a complex and dynamic
process that begins prenatally and extends through
to late adolescence [1]. The human brain has an
estimated 100 billion neurons at birth [2] whose
interconnections form neural networks, which become specialized
over time and mediate the functional capabilities of
the human brain [3]. This specialization results not only from
the structural development of the brain but also as a consequence
of optimization of inter-regional interactions in the
developing brain [3]. Over the past 50 years, the field of
developmental cognitive neuroscience (DCN) has examined
the relations between the structural and functional development
of the human brain [4], elucidating the developmental
mechanisms underlying cognitive processes such as perception,
attention, memory, and language.
DCN research can inform us about the influence of genetic
variations and environmental factors in the specialization of
neural networks [3]. In addition, DCN studies can extend
insights into how these specialized networks mediate newly
acquired social and cognitive functions, shedding light on typical
and atypical trajectories of human brain development [5]. A
greater understanding of brain development trajectories can
have profound implications for early detection and the subsequent
intervention of developmental disorders [4]. Furthermore,
a better understanding of the interplay between
structural and functional brain development can be leveraged
to inform clinical, educational and social policies [6].
In order to examine the neural underpinnings of cognitive
processes and their changes across development, functional NearInfrared
Spectroscopy (fNIRS) [7], [8], and Electroencephalogram
(EEG) [9] have been widely used in DCN studies with
infants and children. These neuroimaging modalities are both
non-invasive, portable, wearable, and relatively inexpensive-
compared to functional magnetic resonance imaging (fMRI),
which has instead proved pivotal in adult brain neuroimaging. In
particular, fNIRS and EEG allow for the young participants to
Developmental cognitive neuroscience is an
interdisciplinary field that aims to understand how
functional and structural changes in a developing
brain lead to the formation of highly optimized cortical
networks in developed (adult) brains. Indeed, a typical
young child's brain develops incredibly fast, with 80%
of adult size reached by age 3 and 90% by age 5.
stay engaged in tasks whilst recording their brain activity in more
naturalistic postures (e.g., sitting upright vs laying down) and
ecologically valid settings such as their homes [8]. Nevertheless,
fMRI has been successfully used in developmental studies with
asleep infants [10], [11], and more recently with awake infants
[12], [13]. As fNIRS and EEG are considered the most commonly
used and 'infant-friendly' modalities to investigate neural
substrates in DCN studies, the present review paper will focus on
these two modalities and their respective data analysis paradigms.
fNIRS is an optical neuroimaging modality that uses NearInfrared
(NIR) light on the scalp to record changes in blood
hemoglobin that occur as a result of cerebral activity. More
specifically, fNIRS measures the relative changes in hemoglobin
(Hb) concentration in the blood, based on NIR light
absorption by the Hb molecules, which is inferred as a measure
of the cortical brain activity [7]. The fNIRS cap, comprising
of pairs of sources and detectors, can be flexibly adapted
based on the brain areas of interest (see for an example Fig. 1a).
The strength of fNIRS lies in its good spatial localization
(within 2 cm) that allows for conclusions to be drawn about
the localized cortical activity from different anatomical locations
of the cortical structures, as recorded by the fNIRS
channels located on the participant's head (Fig. 1a). An illustration
of the fNIRS principle (Fig. 1b) along with a representative
signal (oxy-Hb in red, and deoxy-Hb in blue is
shown in Fig. 1c) is shown in Fig. 1.
While fNIRS relies on changes in blood oxygenation to
measure brain activity, EEG measures electrophysiological brain
activation. More specifically, EEG records electrical changes on
the scalp, allowing the measurement of rapid cognitive processes
[19] with high temporal accuracy in the order of milliseconds
[20]. The EEG principle is represented in Fig. 1e. The
EEG net has electrodes fitted in it using a standardized 10/20
electrode placement system that covers the whole head (see
Fig. 1d). Since EEG can record activity on the time scale of
underlying neuronal activity, EEG signals (representative EEG
signal shown in Fig. 1f) are best suited for connectivity analysis.
However, EEG is also more sensitive to motion artifacts [21]
and due to its limited spatial resolution [18], EEG makes it difficult
to map brain electric activity read by the electrodes to
their corresponding anatomical regions in the brain.
One of the most recent advancements in neuroscience
research is the combined use of neuroimaging techniques (e.g.,
[22]). In particular, in DCN, multimodal imaging can provide a
FEBRUARY 2022 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 17
IEEE Computational Intelligence Magazine - February 2022

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - February 2022
