Individualised Structural and Functional Frontostriatal Connectivity Mapping in Adult ADHD

2.1. Participants

Forty individuals with ADHD and 36 NT controls took part in this study. Adults with ADHD were recruited by psychiatrists at clinics in the Helsinki metropolitan area (the Neuropsychiatry outpatient clinic of the Helsinki University Hospital; Diacor Healthcare Services, Helsinki, Finland; and ProNeuron, Espoo, Finland) and were pre‐screened in the clinic. NT adults were recruited via e‐mail lists of universities, adult high schools, polytechnics, and vocational schools, and via the personal contacts of the authors. To be accepted to the study, the participants had to be native Finnish speakers, have normal hearing, normal or corrected‐to‐normal vision, cognitive abilities at or above average level, and no contraindications to MRI. Participants were excluded if they had moderate or major psychiatric or neurological conditions (apart from ADHD in the ADHD group), such as learning disorders, depression or anxiety disorders, tic disorder, or past head trauma that had demanded treatment. The study was conducted according to the Declaration of Helsinki. All participants gave written informed consent to participate in the study. The study was reviewed and approved by the Ethics Committee for Gynaecology and Obstetrics, Pediatrics and Psychiatry of the Helsinki University Hospital. The participants were compensated by 60 € if they participated only to the parts described in the present study, and by 240 € if they continued to a working memory training trial described elsewhere (Salmi, Soveri, et al. 2020).

ADHD diagnoses were made by experienced psychiatrists based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM‐IV; American Psychiatric Association 1994). All participants with ADHD met criteria for both inattention and hyperactivity/impulsivity or only for inattention, according to the Conners' Adult ADHD Diagnostic Interview for DSM‐IV (Epstein et al. 2001). Two of the participants had hypothyroidism, four had migraines, and two had experienced mild symptoms of epilepsy in childhood requiring no later treatment. The participants with hypothyroidism and migraine had prescription medication, and one participant had a selective serotonin re‐uptake inhibitor prescription for mild depression. In addition, 33 participants with ADHD were using stimulants but were instructed to have a 24‐h wash‐out period before the MRI session.

2.2. Neuropsychological Assessment and Self‐Ratings

General cognitive abilities were assessed with two tasks (Matrix Reasoning and Vocabulary) from the Wechsler's Adult Intelligence Scale III (WAIS‐III; Wechsler 2005). Working memory was assessed with the Digit Span test from WAIS‐III (Wechsler 2005) and with spatial and verbal running memory tasks (Pollack et al. 1959; see Salmi, Soveri, et al. 2020 for full details). For self‐assessment of ADHD symptoms, we used the Adult ADHD Self‐Report Scale (ASRS; Kessler et al. 2005). Depression symptoms were rated with the Depression Scale (DEPS; Salokangas et al. 1995), and alcohol use with the Alcohol Use Disorders Identification Test—Consumption (AUDIT‐C; Bush et al. 1998).

2.3. MRI Acquisition and Preprocessing

2.4. fMRI Conditions

2.5. Defining Regions of Interest

The striatum was divided into regions of interest (ROIs) based on a parcellation by Janssen et al. (2015) (Figure 1). The caudate nucleus, putamen, and ventral striatum were separated to six regions on both hemispheres: dorsal caudate nucleus, ventral caudate nucleus, rostral caudate nucleus and nucleus accumbens, rostral putamen, dorsal putamen, and caudal putamen. Similar clustering of the striatum based on its functional connectivity has also been reported in other studies (e.g., Jung et al. 2014; Liu et al. 2021). Striatal ROIs were transformed to the participants' DW‐MRI space for tractography and DW‐MRI analyses, or to T1‐weighted space for fMRI analyses with ANTs' symmetric image normalisation (SyN; Avants et al. 2008; RRID:SCR_004757).

A cerebro‐cortical ROI pair for each striatal region was then defined by selecting areas that were most densely connected to it based on the number of white matter streamlines. Specifically, first a seed‐based, anatomically‐constrained tractography using constrained spherical deconvolution (CSD) was conducted from each striatal ROI using MRtrix3 (Smith et al. 2012; Tournier et al. 2007, 2019; RRID:SCR_024123). One million streamlines were created from each striatal seed ROI. Tractography was constrained to the frontal lobe unilaterally with individual participant‐specific inclusion and exclusion masks. The inclusion masks included all areas of the frontal cortex from the Destrieux atlas (Destrieux et al. 2010). The exclusion masks covered the rest of the brain (temporal, parietal, occipital, and insular cerebral cortices and the cerebellum), including subcortical structures (the thalamus, pallidum, hippocampus, and amygdala), derived with FSL's FIRST algorithm (Patenaude et al. 2011; Smith et al. 2004; RRID:SCR_002823), and the mid‐sagittal plane. After tractography, a streamline endpoint map was created for each seed‐point with MRtrix3's tckmap‐ends_only, and cerebro‐cortical voxels that included less than 100 streamline endpoints were excluded. Then, neighbouring voxels, either by the voxel face, edge, or vertex, were defined as clusters with MRtrix3's maskfilter‐connectivity, and only clusters including at least 10,000 streamline endpoints in total were selected. The selected clusters formed cerebro‐cortical ROIs and were used as is for DW‐MRI analyses and expanded to cover the whole cortical ribbon for the fMRI analyses with FreeSurfer's surf2vol. See Figure 1 for a schematic overview of the cerebro‐cortical ROI formation.

These voxel‐ and cluster‐wise thresholds were chosen by visually inspecting the cerebro‐cortical endpoints and by selecting thresholds that yielded ROIs which fitted best to known frontostriatal models. Visual inspection was done with 18 participants from the NT group and validated with the rest of the NT group. However, as the threshold selection was based on subjective decisions, the analyses were repeated with several thresholds to ensure the robustness of the findings. Complementary analyses were conducted with voxelwise thresholds of 75 and 125, and a clusterwise threshold of 7500.

To further evaluate the robustness of individual tract reconstructions, we compared them to tracts generated with pre‐defined cerebro‐cortical regions. Connection density (defined as the number of streamlines divided by the mean number of voxels in the striatal and cortical ROIs; see Section 2.7) was used as an index of how accurately each striatal ROI was linked to its cerebro‐cortical counterpart. The pre‐defined cerebro‐cortical ROIs were selected from the Schaefer 200 parcellation (Schaefer et al. 2018) based on prior literature (Alexander et al. 1986; Gordon et al. 2021; Haber 2016; Jung et al. 2014) and used as cortical inclusion masks. The cerebro‐cortical ROIs were clipped with a whole‐brain endpoint mask to get a corresponding voxel count compared with the individually defined ROIs. The chosen regions are illustrated in Figure S1.

2.6. Excluded Participants

Some participants were excluded from further analyses during the data processing steps. From resting‐state analyses, two participants from the ADHD group were removed due to inaccurate image field placement during fMRI. Furthermore, because the resting‐state scan was shorter than the other fMRI paradigms, we applied a conservative criterion for the proportion of motion outliers to ensure sufficient data for reliable functional connectivity estimation (Van Dijk et al. 2010). Participants with more than 30% of motion outlier volumes during the resting‐state acquisition were excluded. This resulted in the removal of three additional participants from the ADHD group and two from the NT group. In the longer n‐back and naturalistic attention paradigms, no participants were removed since the number of volumes remained sufficient after motion outlier removal. The proportion of motion outlier volumes was also within reasonable limits and did not exceed 40% in any participant in these tasks. The final sample size was 40 ADHD and 36 NT participants for the n‐back tasks and for the naturalistic attention condition, and 35 ADHD and 34 NT participants for the resting state condition. Furthermore, all cerebro‐cortical ROIs were not reconstructed (i.e., no clusters containing enough streamline endpoints emerged) in one participant from the NT group. The participant is included in analyses with successful ROI formation (bilateral dorsal caudate circuits and left‐sided caudal, dorsal and rostral putamen circuits). Additional analyses where the participant was removed were conducted as a control. In these analyses, the results remained unchanged.

2.7. Functional and Structural Connectivity Measures

To estimate structural connectivity, whole‐brain CSD‐based anatomically‐constrained tractography was first performed in MRtrix3 with 10 million streamlines seeded from the grey matter—white matter interface (Smith et al. 2012; Tournier et al. 2007). Streamlines were further weighted to correct their density distributions (Smith et al. 2013, 2015). The streamlines connecting each cortical‐striatal ROI pair were extracted from the weighted whole‐brain tractograms by using the ROI inclusion masks (see Section 2.5). The number of extracted streamlines was used as a measure of structural connectivity. Furthermore, because the size of the cerebro‐cortical ROIs could potentially vary substantially between participants due to the definition method, we also measured the streamline density by dividing the number of streamlines by the mean number of voxels in the cortical and striatal ROI.

For functional connectivity, the Nilearn toolbox (Abraham et al. 2014 RRID:SCR_001362) was used to measure Pearson's correlation between the time‐series of each cortico‐striatal ROI pair. Correlation values were Fisher transformed before any further analyses.

2.8. Microstructural Analyses

White matter microstructure was analysed using three conventional measures: FA, AD, and RD. Diffusion tensors were first estimated from the preprocessed DW‐MRI data using weighted least‐squares with MRtrix3's dwi2tensor (Basser et al. 1994; Veraart et al. 2013). Mean FA, AD, and RD were then computed from the tensors along the streamlines extracted between each cortico‐striatal ROI pair (see Section 2.7).

2.9. Spatial Overlap of Structural and Functional Connectivity

We quantified spatial overlap between structural and functional connectivity using the Dice similarity coefficient. For each striatal seed, we first generated binary top 25% masks within a frontal cortical mask, retaining voxels that exceeded the 75th percentile of the connectivity strengths. Dice coefficients were then computed between the structural and functional top 25% masks for each task × seed combination as Dice = 2|A ∩ B|/(|A| + |B|).

2.10. Medication Effects on Structural and Functional Connectivity

We evaluated medication effects on functional and structural connectivity using ordinary least squares regression. Analyses were conducted within the ADHD group, because medication was only present in the participants with ADHD. For overall stimulant use, we fit y ~ medication use (stimulant users n = 33, non‐users n = 7) separately for each connectivity outcome. To examine specific medication types, we modelled y ~ medication group (methylphenidate users n = 29, (lis)dexamfetamine users n = 4, non‐users n = 7) with reference “non‐user”, yielding the contrasts methylphenidate versus non‐user and (lis)dexamfetamine versus non‐user. We used heteroskedasticity‐robust standard errors (HC3) to account for unequal group sizes.

2.11. Statistical Analyses

Normality of the variables was assessed through visual inspection of Q–Q plots and the D'Agostino–Pearson test. For variables in streamline connectivity, streamline density, and microstructure, clear outliers were identified and data points exceeding ±3 standard deviations from the group mean were removed. Between‐group differences in functional connectivity, number of streamlines, and streamline density were tested with independent‐samples t‐tests. Analyses with pre‐set hypotheses were conducted with one‐tailed tests and experimental analyses with two‐tailed tests. Overall microstructural properties were examined with one‐way multivariate analysis of variance (MANOVA), examining the impact of group on the multivariate pattern of FA, AD, and RD. For MANOVA results showing significant group effects, post hoc independent‐samples t‐tests were performed on each of the three microstructural measures. The relationship between post‐quiz performance and functional connectivity during the naturalistic attention paradigm was examined with Pearson's r. Correction for multiple comparisons was performed using the false discovery rate (FDR), applied across the 12 frontostriatal ROI pairs or across the three microstructural measures in the MANOVA post hoc analyses. Because some variables deviated from normality and failed the D'Agostino–Pearson test, all t‐tests were re‐run using the nonparametric Mann–Whitney U‐test. The pattern of results remained unchanged.

3.1. Background Variables

The ADHD and NT groups did not statistically differ in age, gender, handedness, education level, general cognitive abilities, alcohol consumption, mood, or in quiz performance after the naturalistic attention task. The ADHD group had more inattention and hyperactivity symptoms than the NT control group and performed worse in a spatial running memory working memory task. During the n‐back tasks, the participants made more errors when the working memory load increased, but there were no statistically significant group differences in hit rates at any load level. The ADHD participants had faster reaction times in all n‐back levels during the visuospatial task, and in the 2‐back level during the digit task. See Table S1 for statistics related to background variables in this sample (also Tolonen et al. 2023).

3.2. Reconstructed Frontostriatal Circuits

Cerebro‐cortical ROIs defined with tractography from the six striatal seed points in each hemisphere followed well the patterns described in previous literature (Figure 2). More specifically, in the putamen, the caudal part was mainly connected to primary motor areas, dorsal area to premotor areas and rostral part to OFC. In the caudate nucleus, the dorsal section was primarily connected to the lateral PFC, mostly its dorsal and ventral areas and extending to the frontal pole. The cerebro‐cortical region also covered caudal and dorsal medial areas. The cerebro‐cortical counterpart of the ventral caudate nucleus overlapped largely with the one of the dorsal caudate nucleus but was centred slightly more ventral and rostral, and included more ventromedial PFC areas. The rostral caudate nucleus and nucleus accumbens were predominantly connected to the ACC and orbitofrontal areas, overlapping with the rostral putamen connectivity pattern. There were clear individual differences in the specific cortical endpoint regions, but overall, they followed a similar pattern across participants and across both groups, although the ADHD group showed slightly more inter‐individual variability (Figure 2). Tracts constructed individually for each participant included more streamlines relative to ROI size than the tracts formed by tracking to pre‐selected cortical areas (all p < 0.001, see Figure S2). Cortical ROI maps for each seed point including all participants and transformed to MNI‐space are available as Supporting Information of this article.

3.3. White Matter Group Differences

3.4. Functional Connectivity

3.5. Spatial Overlap of Structural and Functional Connectivity

Mean dice coefficients between structural and functional connectivity varied between 0.16 and 0.24, depending on striatal seed point and task. Largest dice coefficients were observed for the dorsal and ventral caudate seeds, and the smallest for rostral caudate and putamen seeds. See Table S3 for further statistics.

3.6. Complementary Analyses

4.1. Individual Reconstruction of the Frontostriatal Circuits

Individualised definition of brain areas is crucial in neuroscience, as brain structure and functional patterns vary substantially across individuals (Gordon et al. 2017; Kong et al. 2021; Mueller et al. 2013). This is particularly relevant in disorders such as ADHD, which are underpinned by heterogenous variations in brain structure, function, and connectivity (Lei et al. 2014; Pan et al. 2025; Qian et al. 2019; Saad et al. 2017; Shang et al. 2018; Wolfers et al. 2020), driven by developmental changes shaped by genetic variability and environmental factors (Faraone et al. 2024). The inter‐individual variability of both symptoms and their biological correlates is a core challenge in ADHD research (Bu et al. 2025; Faraone et al. 2024; Firouzabadi et al. 2022). By selecting individually defined white matter tracts and the associated cerebro‐cortical ROIs, we aimed to mitigate the loss of true participant‐specific variability that could occur when using atlas‐based ROI selection.

While functional co‐fluctuations between brain regions depend on both direct and indirect structural connections, the relationship between structural and functional connectivity is complex (Fotiadis et al. 2024). Structural connectivity reflects the brain's long‐term architectural framework and tends to remain relatively stable over time. In contrast, functional connectivity is dynamic, context‐sensitive, and influenced by cognitive states and environmental demands. This divergence suggests a gradient in structure–function coupling across the brain: strong coupling is typically found in primary sensory and motor areas, whereas it becomes weaker in transmodal regions responsible for higher‐order processes such as attention, emotional regulation, and working memory (Fotiadis et al. 2024). The discrepancy between structural and functional connectivity was also apparent in the present study, showing only modest spatial overlap across frontostriatal circuits and task types. Similar modest structure–function coupling in frontostriatal pathways has been reported in macaques (Tang et al. 2025). Together, these findings underscore the need for further work on frontostriatal structure–function relationships across task types.

The structural connectivity patterns were topologically largely reproducible across participants, but the specific spatial location and size of each cerebro‐cortical endpoint region showed some individual variability which was slightly more pronounced in the ADHD group. Similar inter‐individual variability has been observed previously regarding both frontostriatal connectivity (Gordon et al. 2021; Tziortzi et al. 2014) and global functional organisation of the brain more generally (Gordon et al. 2017; Kong et al. 2021; Mueller et al. 2013). Comparing individually parcelled tracts with tracts conducted using pre‐selected cerebro‐cortical areas revealed that the individual tracts had denser connections (i.e., included more streamlines per voxel) to the frontal areas. This indicates that our individual tract formation approach was successful in pinpointing the most relevant cerebro‐cortical counterparts for different striatal regions. Individual connectivity disparities emphasise the need for more participant‐specific ways of describing brain function and structure rather than relying on atlas‐based brain parcellations. It has been proposed that mapping of individual connectomes is an important step forward in analysing the brain both in health and disease (Gordon et al. 2017). Furthermore, individualised, tractography‐based mapping could help refine targeting of relevant cerebro‐cortical areas for neuromodulation therapies used in psychiatric and neurological disorders (Lioumis et al. 2025; Ukharova et al. 2025). This approach may also facilitate improved detection, prognosis, and personalised treatment in these conditions, analogous to possibilities enabled by precision functional mapping (Demeter and Greene 2024). Our research suggests that DW‐MRI tractography can be utilised to define individual connectivity patterns for structural and functional analyses, at least in frontostriatal regions.

4.2. ADHD‐Related Differences in Frontostriatal White Matter

Our findings add to previous literature highlighting the centrality of OFC in ADHD symptomology (Norman et al. 2016; Yu et al. 2023), as we observed structural alterations in frontostriatal tracts originating from these regions. The present results align with previous reports of white matter abnormalities in striato‐OFC pathways, identified in both volumetric (Cupertino et al. 2020) and microstructural studies (Konrad et al. 2010; Schweren et al. 2016). The OFC plays a central role in regulating emotional and motivational behaviour (Rolls 2023), and aberrant OFC‐connectivity may underlie the atypical reward‐related decision making and emotional difficulties frequently reported in ADHD (Shaw et al. 2014). Supporting this view, individuals with ADHD have been shown to exhibit disrupted connectivity (Ceceli et al. 2020; Ma et al. 2016; Park et al. 2016) and altered activation patterns (see review by Rubia 2018) in the OFC and striatum during tasks involving motivational and emotional processing. Although the differences were observed only in the left hemisphere, there were no statistically significant group × hemisphere interactions. Overall, the structural aberrancies were susceptible to choices made in ROI formation, making them less robust, and must be interpreted accordingly.

Previous investigations of frontostriatal structural connectivity in adult ADHD are scarce. While previous evidence points to altered RD, AD, and FA in frontostriatal tracts in ADHD (Chaim et al. 2014; Chiang et al. 2022; Gehricke et al. 2017; Konrad et al. 2010; Schweren et al. 2016), no microstructural changes were observed in the present study in the main analyses. However, using more stringent thresholds in ROI selection we observed increased RD and AD in the ADHD group in tracts originating from orbitofrontal regions, but due to the complementary nature of these analyses the results are not discussed further. Regarding connectivity strength, some studies have reported both increased and decreased structural connectivity of striatal and frontal regions, but their network‐based analytical approaches do not allow us to examine frontostriatal connections specifically (Sidlauskaite et al. 2015; Tolonen et al. 2023; Wang et al. 2021). However, despite methodological differences, our findings are broadly consistent with those of the mega‐analysis by Cupertino et al. (2020), which reported reduced white matter volume in an OFC‐striatal component: Both results converge on the interpretation of diminished connectivity between orbitofrontal regions and the striatum in adult ADHD.

4.3. Aberrant Functional Connectivity During Working Memory Tasks

The present study sheds light on the neural basis of working memory deficits in ADHD (Alderson et al. 2013; Willcutt et al. 2005). While general frontostriatal circuit involvement was previously highlighted in neurocognitive models (Castellanos and Tannock 2002; Sonuga‐Barke 2002), our findings pinpoint precise pathways impacted. Both verbal and spatial working memory hypoactivation in ADHD appeared in somewhat similar frontostriatal circuits, including dlPFC, premotor, and motor regions. While the dlPFC's role in working memory deficits in ADHD is well established (Bédard et al. 2014; Cortese et al. 2012; Massat et al. 2012; McCarthy et al. 2014; Tolonen et al. 2024), our results indicate that altered connectivity between the dlPFC and striatum is a key mechanism behind these aberrations. The left dlPFC was affected during the verbal working memory task, along with the left vlPFC and OFC circuits, whereas the right dlPFC showed effects during the spatial task. This pattern accords with canonical hemispheric lateralisation of working memory domains (Nagel et al. 2013; Walter et al. 2003). However, no significant group × task × hemisphere interactions were detected in the present sample. Aberrant connectivity from the ventral caudate nucleus during the verbal task may relate to overlapping cerebro‐cortical representations with the dorsal caudate nucleus or the vlPFC's role in verbal working memory and related deficits observed in ADHD (Cortese et al. 2012; Rottschy et al. 2012).

Divergent connectivity in motor and premotor circuits, while perhaps unexpected, aligns with their supporting role in working memory processes (Marvel et al. 2019; Rottschy et al. 2012; Wang et al. 2019; Yaple et al. 2019; Yaple and Arsalidou 2018). Moreover, aberrations within the motor network have repeatedly been proposed as a key feature of ADHD (Cortese et al. 2012). Similarly, aberrant OFC connectivity might reflect its well‐documented involvement in ADHD (Norman et al. 2016; Yu et al. 2023), or its role in networks suppressed during tasks (Girn et al. 2024). Evidence also suggests that the OFC supports complex functions like maintaining the relative recency of presented stimuli (Barbey et al. 2011; Johnson et al. 2022). Collectively, these findings highlight that ADHD‐related working memory deficits involve not only canonical frontostriatal pathways but also altered contributions from motor and orbitofrontal circuits.

Despite connectivity findings, no differences in in‐scanner n‐back task accuracy were noted between the groups. However, the ADHD participants showed faster reaction times in many task levels, which might reflect their impulsive behaviour tendencies. Nevertheless, discrepancy between brain activation and behavioural findings in ADHD studies is not unique to our study (see e.g., Massat et al. 2012; Valera et al. 2005). The MRI environment might influence task engagement and performance (e.g., van Maanen et al. 2016), and potentially mask group differences that might appear elsewhere. Indeed, the adults with ADHD performed worse than the NT adults on the out‐of‐scanner spatial running memory task. However, we found no associations between running memory performance and functional connectivity during the spatial task, limiting interpretation of the group differences in functional connectivity. This dissociation can also result from the two task types likely engaging partially distinct cognitive processes (Bunting et al. 2006; Huang et al. 2025). Differences in functional connectivity may arise from factors not captured by hit rates or reaction times, such as engagement of additional brain systems reflecting alternative task strategies (Chauvin et al. 2021; Fassbender and Schweitzer 2006).

Contrary to a mega‐analysis on children with ADHD (Norman et al. 2024), we found no group differences in frontostriatal connectivity during resting state in adults. However, developmental changes in symptomatology and brain organisation (Franke et al. 2018; Grayson and Fair 2017; Lebel et al. 2008; Mills et al. 2016) complicate direct child‐adult comparisons. Though adult studies have noted altered striatal and frontal connectivity during rest (Hoekzema et al. 2013; Lin and Gau 2015; McCarthy et al. 2013; Sörös et al. 2019; Zhao et al. 2017), reports of disrupted connectivity between these regions specifically are scarce. The present results also align with our previous findings, showing differences in working memory‐related but not in resting‐state whole‐brain connectivity (Tolonen et al. 2024). Despite the growing popularity of resting‐state paradigms and their important discoveries of connectivity alterations in clinical populations, there are major difficulties in meaningfully interpreting their findings (Finn 2021). A focus on more relevant task designs could better reveal brain correlates related to symptom manifestation in everyday life.

However, differences between task and resting conditions do not explain the lack of group differences in the naturalistic attention task, which involves watching a film with distractors to mimic daily life attentional lapses (Salmi, Metwaly, et al. 2020). Our earlier work showed ADHD‐related desynchronisation in attention‐related frontal areas during this task (Salmi, Metwaly, et al. 2020), and other studies have similarly reported reduced frontal activation and aberrant connectivity during attentional processing in ADHD (Hart et al. 2013; Kowalczyk et al. 2022). Furthermore, in our prior work, recall accuracy after movie viewing correlated with coherence of brain activation in the inferior frontal gyrus among participants with ADHD and in the frontal eye fields among NT controls (Salmi, Metwaly, et al. 2020). However, no association between frontostriatal functional connectivity and recall accuracy was found in the present study. The role of frontostriatal circuits in this paradigm and other naturalistic tasks remains unclear due to limited research. Given that naturalistic tasks evoke strong neural responses alongside diverse emotional and cognitive processes (Sonkusare et al. 2019), the absence of detectable group differences here was unexpected. Nevertheless, the current findings underscore that the detectability of neural alterations in ADHD may be highly context dependent.

Group differences emerged during working memory tasks but not during the other functional conditions, suggesting that frontostriatal abnormalities may be most detectable when circuits are engaged by core operations such as working memory updating. Group × task interactions were observed in motor circuits implicated in motor planning and execution, suggesting potential differences in psychomotor processing between the adults with ADHD and NT controls, particularly given that these were the only tasks requiring overt motor responses. Notably, however, motor circuits also have a strong supporting role in working memory processes (Marvel et al. 2019; Rottschy et al. 2012; Wang et al. 2019; Yaple et al. 2019; Yaple and Arsalidou 2018). Taken together, our findings likely reflect task‐evoked, domain‐specific dysfunction related to working memory deficits (or psychomotor operations) rather than stable, trait‐like alterations that generalise across conditions.

4.4. Limitations

Several limitations of the present study should be acknowledged. The relatively modest sample size limited the statistical power of our analyses, particularly given the need to correct for multiple comparisons. Some effects may not have survived FDR correction with this sample size. It also precluded more detailed subgroup investigations, such as examining potential gender differences. In addition, because of the limited a priori information about the effect sizes of the studied phenomena, we did not conduct formal power calculations, which represents an additional limitation. Furthermore, in order to reduce the impact other clinical conditions might have on the results, individuals with comorbid psychiatric or neurological conditions were excluded from the present study. While this approach improved the specificity of the results to ADHD, it also reduced the generalisability of our findings, as the sample represents a more restricted subset of the broader ADHD population. A substantial proportion of the participants in the ADHD group were also receiving stimulant treatment, and consequently, the differences between the ADHD and NT participants cannot fully be separated from effects resulting from stimulant exposure. This also prevented adjusting for medication in group comparisons. In the within‐group analysis, stimulant use was associated with small effects that varied in direction, which makes the interpretation more complex. For functional connectivity, the association with stimulant use was negative and for structural connectivity it was positive. However, these findings should be viewed as provisional given the small (lis)dexamfetamine (n = 4) and non‐stimulant (n = 7) subgroups. In terms of imaging parameters, the DW‐MRI data were acquired using a traditional single‐shell, that is, single diffusion‐weighting, sequence in contrast to a more advanced multi‐shell acquisition, limiting the range of possible analysis methods that could be implemented, especially regarding microstructural features. The use of single‐shell, single‐tissue CSD may also have affected tractography accuracy because it can overestimate the proportion of white matter in voxels that also contain grey matter or CSF, potentially introducing spurious FOD peaks in those regions (Jeurissen et al. 2014). Furthermore, our complementary analyses indicated that the structural group differences were dependent on the chosen voxel‐wise threshold, suggesting limited robustness and sensitivity to subjective threshold selection. Lastly, our connectivity analyses focused exclusively on direct connections between striatal and cerebro‐cortical regions, which may have overlooked more complex topological alterations in connectivity patterns that could be revealed through more comprehensive analytical frameworks.