Attention deficit hyperactivity disorder (ADHD) is one of the most common disorders amongst children and has attracted substantial attention from the scientific community over the last decades. This is not surprising given that 3-8% of school-aged children have a clinical diagnosis of ADHD (Rubia et al., 2014). It has been estimated that 65% of ADHD cases will persist into adulthood (Polanczyk et al., 2015; Rubia et al., 2014). In a national survey conducted in the United States in 2016, Danielson et al. estimated approximately 6.1 million ever-diagnosed ADHD children in the age range between 2-17 years old (Table 1) (Danielson et al., 2018a). The high prevalence of ADHD in children during the last decades has generated questions regarding the cause and neurophysiology of the disorder.

Amongst these questions, many highlighted the apparent rising prevalence of the diagnosis in the population. Along with this rising prevalence, economic, societal and personal costs started to increase (Dijk et al., 2021). It has been estimated that the cost per child/adolescent with ADHD in the United States was between $621 and $2720 (Doshi et al., 2012). When including patients of all ages, the economic burden of ADHD ranges between $143 and $266 billion (Doshi et al., 2012). Given the economic burden coupled with the rise in prevalence, it is not surprising that many questioned the quality and consistency of the diagnostic process across time, and between clinicians. The foundations of the diagnosis are based on subjective evaluations from either family members and/or teachers as well as observations by clinicians (Rubia et al., 2014) as described in the DSM-V manual (Diagnostic and statistical manual of mental disorders: DSM-5TM, 5th ed., 2013). Meanwhile, the underlying mechanism or cause has been completely excluded from this diagnostic approach (Posner et al., 2020). As a result of this “observational” diagnostic method, there is variability between clinicians, countries or even cultures (Rubia et al., 2014).

As proposed in the Canadian ADHD Resource Alliance (CADDRA) guidelines, there are two main categories of treatment: pharmacological treatment and psychosocial interventions (CADDRA, 2020). The pharmacological approach consists of prescribing psychostimulants such as methylphenidate and atomoxetine. The outcome of this treatment is assessed by trial and error based on symptom response. This type of pharmacological treatment has proven to be safe and efficient in the short term (Tourjman et al., 2022), but the long-term benefits have yet to be established (Posner et al., 2020). Although psychostimulants have proven to improve school performance and quality of life in children and adolescents as well as reduce adverse events (Jangmo et al., 2019; Tourjman et al., 2022; Tsujii et al., 2020), many prefer psychosocial intervention as opposed to pharmacological treatment (CADDRA, 2020). As described in the CADDRA guidelines, psychosocial interventions can consist of behavioural therapy, mind-body training (meditation, yoga, etc.), parent training in behaviour management, or behaviour management in a school setting (Shrestha et al., 2020).

These interventions can be combined with the use of psychostimulants. In the same US national survey by Danielson et al., amongst children with ADHD, 62% received medication, 47% received behavioural treatment and 32% received both treatments (Table 1) (Danielson et al., 2018b). These treatments provide beneficial results by reducing adverse events and improving school performance and quality of life (Jangmo et al., 2019; Tourjman et al., 2022; Tsujii et al., 2020), but they do not address the core pathophysiology of ADHD. On the other hand, psychostimulants are known to cause secondary effects such as appetite loss, growth delay, headaches, and an increase in anxiety. To emphasize the non-specificity of psychostimulants, in our society focused on results and performance, the use of psychostimulants such as methylphenidate is seen in healthy adolescents or adults to improve their performance in attention-related tasks in school or a professional setting (Urban and Gao, 2017). It is estimated that 1 in 10 college students use or have used psychostimulants to improve their cognitive performance in a school setting (Maher, 2008; Sharif et al., 2021; Steward and Pickersgill, 2019).

In addition, many other disorders are associated with ADHD. In the United States in 2016, 6 out of 10 ADHD children had at least one or more emotional or behavioural disorders (Danielson et al., 2018a). As a result, the ADHD phenotype is heterogenous and can be seen as concealing multiple endophenotypes presenting different cognitive deficits (e.g., working memory, inhibition, etc.)(Posner et al., 2020). In another study, researchers estimated that 10-90% of ADHD children present comorbidities, which are not always reported in studies (Gnanavel et al., 2019). The absence of a homogeneous ADHD population (without comorbidities) represents an obstacle to properly assessing cognitive constructs in children with ADHD. With a homogeneous population, we could potentially identify specific neural correlates that would otherwise not be detectable in a more heterogeneous population. Such a homogeneous ADHD population will reduce the external validity of the findings. These various sub-phenotypes can be driving the ADHD diagnosis in some children by influencing attentional capabilities and therefore obscuring the core pathophysiological cause away. To address these core mechanisms underlying the different deficits encompassed within the ADHD diagnosis, many research teams turned their focus to neuroimaging methods.

MRI studies have taken an important place in clinical neuroscience as researchers have tried to elucidate anatomical, functional or structural phenotypes that could explain disorders such as ADHD. The main research goal was to, ultimately, develop objective criteria to help reduce the variability in the diagnostic process. Meta-analyses have reported the findings in functional MRI (fMRI) studies (Hart et al., 2013; McKenna et al., 2017; Saad et al., 2020; Zhang et al., 2017) and therefore fMRI studies will not be discussed in this review article. Technological advances in computer science during the last decades have led to an emerging technique to reconstruct and visualize white matter connectivity and microstructure. Diffusion MRI (dMRI) has grown in popularity in recent years by allowing the evaluation of specific white matter bundles relevant to cognitive functions. Since the development of diffusion imaging, many acquisition schemes have been used to improve acquisition quality or decrease acquisition time. Simple schemes such as diffusion tensor imaging (Basser et al., 1994) have evolved into more complex structures of q-space sampling like diffusion spectrum imaging (Callaghan et al., 1988), single-shell and multiple-shell high angular resolution diffusion imaging (HARDI) (Tuch et al., 2002) allowing for more robust reconstructions of crossing fibers and edges. Diffusion tensor imaging (DTI) is probably the technique most often used in studies due to its reasonable acquisition time compared to the quality of reconstruction. As a result, the DTI measures such as fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD) and axial diffusivity (AD) are the main dMRI measures used to assess WM microstructure. These measures have been, to some extent, validated in post-mortem histological samples with some limitations and challenges (Seehaus et al., 2015). All these measures quantify water displacement in a voxel via the diffusion tensor model (Figure 1.). FA represents a measure of isotropy versus anisotropy (weighted from 0 to 1, respectively), the MD value is the mean water displacement inside the voxel in mm2/s, radial diffusivity is the water displacement rate perpendicular to the main vector and axial diffusivity is the water displacement rate parallel to the main vector, both measured in mm2/s (Figure 1.) (Alexander et al., 2007; Descoteaux, 2015; Le Bihan et al., 2001).

With the development of new imaging techniques, clinical neuroscience has bloomed in the last two decades and the number of studies examining WM microstructure in various psychiatric disorders has increased exponentially. Some of these studies have tried to identify core neural mechanisms of ADHD through these imaging techniques, revealing findings from structural to functional experiments. Amongst these findings, ADHD has been associated with dysfunction and/or abnormal morphology in multiple functional networks comprising the frontoparietal, the frontostriatal and the fronto-cerebellar networks that are responsible for the executive functions system (Rubia et al., 2014). Evidence regarding structural, functional and connectivity deficits in the default mode network (DMN) comprising the medial prefrontal cortex, posterior cingulate cortex, precuneus, lateral parietal and medial temporal cortices have been associated with ADHD (Raichle et al., 2001; Raichle and Snyder, 2007; Rubia et al., 2014). This connectivity deficit is thought to be related to a poor inactivation of the DMN during task performance causing an intrusion upon the task-positive cognitive system (Raichle et al., 2001; Raichle and Snyder, 2007; Rubia et al., 2014). More areas such as the prefrontal cortex, the caudate nucleus, the putamen and the cerebellum have also been linked to ADHD for their regulation of attention, motivation, emotions, behaviour and actions in various studies (Arnsten and Pliszka, 2011; Kasparek et al., 2015; Rubia et al., 2014; Samea et al., 2019; Sharma and Couture, 2014; Shaw et al., 2007; Tripp and Wickens, 2009). On a cellular level, researchers have found that ADHD is related to a deficit in dopamine and norepinephrine receptors which are congruent with the regions identified above (Faraone and Biederman, 1998). Regions like the caudate nucleus and putamen are rich in dopaminergic synapses as well as the frontal cortex, which is the main target of methylphenidate (Biederman and Faraone, 2005; Sharma and Couture, 2014). In a review article, authors listed various white matter connections implicated in the pathophysiology of ADHD in children. The reported tracts were connecting the striatum, cerebellar, corpus callosum, fornix, parietal lobe, occipital lobe, frontal lobe, corona radiata, and internal capsule (Figure 2.). The corticospinal tracts as well as the superior longitudinal fasciculus were also listed for their implication in the pathophysiology of ADHD (Figure 2.) (Cortese, 2012). Studies also reported late development of cortical thickness and surface area of the frontal cortex when comparing ADHD children and neurotypical children which suggests a developmental aspect in the pathophysiology of ADHD (de Zeeuw et al., 2012; McAlonan et al., 2009; Shaw et al., 2012, 2007).

Although the findings from these case-control studies are a great addition to our understanding of the neurophysiology of ADHD, most of them rely on clinical diagnosis as the group classifier for their analysis. It is often seen that studies would use subtypes described in the DSM-V criteria to classify ADHD children either in the inattentive, hyperactive/impulsive or combine subgroups rather than using a specific symptom score (Diagnostic and statistical manual of mental disorders: DSM-5TM, 5th ed., 2013). Whilst these studies have identified many regions that could potentially provide insights to better our understanding of the disorder, the literature is still heterogeneous. We suggest that this heterogenicity stems from the group classification which is inherent to the design of case-control studies and the imprecise phenotyping of ADHD children. These phenotypes represent a wide range of specific cognitive deficits (or “endophenotypes”) and degrees of severity expressed as ADHD behaviour. Our premise is based on the rationale that if there is a specific cognitive construct (ex: working memory), which is primarily affected in a child with ADHD, we should be able to visualize a WM “signature” on a dMRI scan. The grouping of this child with another one who has a different cognitive deficit (ex: inhibition control) within the same study group will impact the regression to the mean during statistical analysis (e.g., comparison of these cases as a group with controls). The wrongful grouping will hide the specific WM correlates resulting in heterogeneous statistical findings. As stated in the Research Domain Criteria Initiative (RDoC) of the National Institute of Mental Health (NIMH), it is imperative to consider the interaction between every human cognitive domain outside and across clinical diagnoses.

Many studies have used diffusion MRI techniques to examine the underlying mechanisms by identifying tracts that are associated with inattention and hyperactivity/impulsivity symptom severity in ADHD. Many meta-analyses and reviews have covered this subject (Aoki et al., 2018; Connaughton et al., 2022; van Ewijk et al., 2012; Zhao et al., 2022). Therefore, this review will briefly report the main findings concerning WM bundles from dMRI studies while suggesting the next steps to better characterize and understand the pathophysiology of ADHD (Table 2). Ultimately, our goal is to propose a model to better understand the key component underlying the development of ADHD. This discussion could very well take place with any mental health disorder, as they all share the same diagnostic approach, but, due to the rise in ADHD rates and variability of endophenotypes, this review will focus on ADHD.