White matter alterations in Attention-Deficit/Hyperactivity Disorder (ADHD): a systematic review of 129 diffusion imaging studies with meta-analysis

Aberrant anatomical brain connections in attention-deficit/hyperactivity disorder (ADHD) are reported inconsistently across diffusion weighted imaging (DWI) studies. Based on a pre-registered protocol (Prospero: CRD42021259192), we searched PubMed, Ovid, and Web of Knowledge until 26/03/2022 to conduct a systematic review of DWI studies. We performed a quality assessment based on imaging acquisition, preprocessing, and analysis. Using signed differential mapping, we meta-analyzed a subset of the retrieved studies amenable to quantitative evidence synthesis, i.e., tract-based spatial statistics (TBSS) studies, in individuals of any age and, separately, in children, adults, and high-quality datasets. Finally, we conducted meta-regressions to test the effect of age, sex, and medication-naïvety. We included 129 studies (6739 ADHD participants and 6476 controls), of which 25 TBSS studies provided peak coordinates for case-control differences in fractional anisotropy (FA)(32 datasets) and 18 in mean diffusivity (MD)(23 datasets). The systematic review highlighted white matter alterations (especially reduced FA) in projection, commissural and association pathways of individuals with ADHD, which were associated with symptom severity and cognitive deficits. The meta-analysis showed a consistent reduced FA in the splenium and body of the corpus callosum, extending to the cingulum. Lower FA was related to older age, and case-control differences did not survive in the pediatric meta-analysis. About 68% of studies were of low quality, mainly due to acquisitions with non-isotropic voxels or lack of motion correction; and the sensitivity analysis in high-quality datasets yielded no significant results. Findings suggest prominent alterations in posterior interhemispheric connections subserving cognitive and motor functions affected in ADHD, although these might be influenced by non-optimal acquisition parameters/preprocessing. Absence of findings in children may be related to the late development of callosal fibers, which may enhance case-control differences in adulthood. Clinicodemographic and methodological differences were major barriers to consistency and comparability among studies, and should be addressed in future investigations.


INTRODUCTION
Attention-deficit/hyperactivity disorder (ADHD) is a neurodevelopmental condition characterized by age-inappropriate inattentive and/or hyperactive-impulsive symptoms [1].Cognitively, individuals with ADHD may present with deficits in executive functions, such as motor inhibition, attention, and/or working memory [2].It is commonly diagnosed in childhood, with community prevalence between 2-7% [3], but its impairing symptoms persists in adulthood in up to 65% of cases, and are associated with poor social and occupational outcomes [4].Co-occurrent disorders, from autism spectrum disorder (ASD) to affective and substance abuse disorders, are often observed [5,6].Multiple genetic and environmental factors contribute to ADHD, but it is unclear how they interplay with brain development to produce symptoms and cognitive deficits [2].Understanding the underlying neuropathophysiology is crucial to develop and tailor behavioral, pharmacological or brain-based treatments.
Meta-analyses of structural and functional neuroimaging studies have identified several case-control differences, but mainly focused on regional alterations [2].However, brain regions operate as neural networks, and there is increasing evidence that anatomical brain connections are also affected in ADHD [7].Diffusion-weighted imaging (DWI) is the only non-invasive imaging method that allows us to study the anatomy of brain connections in the living human.It measures the diffusion of water molecules, which in the brain is restricted by structures such as myelin and axons, providing information on the microstructural organization of white matter tracts [8,9] (Box 1).
The first systematic review of diffusion imaging studies in ADHD was published in 2012 and included 15 studies, mostly in pediatric samples [10].The meta-analysis of nine of the included studies revealed diffuse alterations mainly affecting fronto-striatocerebellar connections [10].The two following meta-analyses, respectively published in 2016 and 2018, included VBA and/or TBSS studies and mainly identified regions of reduced FA in posterior commissural fibers [11,12].The separate analysis of TBSS and VBA studies also allowed the identification of regions of increased FA in the corpus callosum (CC) and cingulum [11].These evidence syntheses were important to elucidate the most consistent findings in ADHD.Still, they did not consider studies using techniques not amenable to meta-analysis, and included a very limited number of studies in adults.Since then, DWI has benefitted from considerable technological advances, and the quality and number of published studies in the field of ADHD has progressively increased.A more comprehensive review with metaanalysis is therefore timely to provide a broader view of the findings, identify the most robust evidence, and highlight methodological considerations.Therefore, we conducted a comprehensive systematic review of DWI studies in ADHD, including a quality assessment of imaging data acquisition, preprocessing and analysis.We then performed, wherever possible, meta-analyses including individuals with ADHD of any age and, separately, children and adults.Finally, we conducted meta-regressions to test the effect of age, sex, and medicationnaïve status, followed by a sensitivity analysis including only highquality datasets.

METHODS
This study followed a preregistered protocol (PROSPERO 2021 CRD42021259192) and is reported in line with the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement [13].

Data sources
We searched the following electronic databases: PubMed (Medline), Ovid databases (Ovid MEDLINE ® , EMBASE Classic+EMBASE, PsycINFO), and Web of Knowledge (including Web of Science, Biological Abstracts, BIOSIS, Food science and technology abstracts), without language and date restrictions.The search was first conducted on the 22 nd June 2021 and updated on the 26 th March 2022.Search terms and syntax for each electronic database are reported in the Supplementary material.The reference lists of previous reviews were hand-searched for any additional eligible studies that could have been missed in the electronic searches.

Identification and selection of studies
First, two authors (VP and TI) independently screened titles and abstracts of all nonduplicated papers and agreed on a final list of studies that proceeded to full-text screening.Then, these two authors independently assessed the eligibility of these studies for the systematic review and meta-analysis.Any discrepancy between the two authors was resolved by a third senior author (SC).

Study selection
Studies were included in the systematic review if they: 1. were peer-reviewed, indicating methodological adequacy, in line with recent meta-analyses [11,12]; 2. recruited individuals diagnosed with ADHD based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders (DSM-III or following editions) or International Classification of Diseases (ICD-9 or 10) and typically developing (TD) controls; 3. collected diffusion imaging data from both ADHD participants and TD controls.
According to our pre-registered protocol, we assessed the feasibility of conducting a meta-analysis of eligible whole-brain studies (i.e., TBSS and VBA studies), provided their number was increased sufficiently (i.e., by 50%), as compared to the last published meta-analysis before the start of this study [11], to justify a new meta-analysis.Additionally, studies were eligible for the meta-analysis only if they compared whole-brain diffusion imaging data (any metric) between individuals with ADHD and TD controls.

Data extraction
Two authors (TI and YL) independently extracted information from the studies selected for the systematic review.Any discrepancy

BOX 1
Traditional (tensor based) DWI studies in ADHD have mainly analyzed case-control differences using four different approaches: voxel-based analysis (VBA), region of interest (ROI) analysis, tract-based spatial statistics (TBSS), and tractography.The first diffusion imaging study in ADHD [19] used a VBA approach, which offers the advantage of identifying whole-brain white matter alterations without a priori anatomical hypotheses.Nevertheless, this approach is affected by the choice of normalization and interpolation techniques, and the need for multiple comparisons correction may reduce the power to detect significant group differences [146][147][148][149]. Conversely, ROI approaches extract diffusion measurements from circumscribed brain areas selected based on a priori hypotheses.Thus, they are sensitive to potential differences in the selected regions but may be limited by the manual selection of ROIs, especially in early studies.Further, they are not amenable to be included in whole-brain meta-analytic syntheses, as they could inflate group differences in the selected regions [150,151].TBSS has been developed to partly overcome the limitations of both VBA and ROI methods [152].It uses a whole-brain white matter 'skeleton' mask to restrict the analyses to the central part of the major tracts.This minimizes potential misalignment artefacts while increasing the statistical power to detect group differences.However, TBSS may not capture potential differences in the periphery of a tract and is still based on comparisons at the voxel level.Finally, tractography allows the analysis of group differences along specific white matter tracts [149].Its main advantage is the possibility of extracting diffusion measurements from the entire course of individual tracts, although it may be less accurate in regions with complex fiber organization (e.g., crossing fibers) [153].Further, the high variability among tractography algorithms and the placement of the ROIs used to identify the tracts may limit comparisons among studies.DWI allows the extraction of quantitative indexes of white matter microstructural organization, such as fractional anisotropy (FA), mean diffusivity (MD), and axial and radial diffusivity (AD and RD, respectively) [154,155].FA ranges from 0 to 1, according to the degree of tissue anisotropy, and is the most used DWI metric.MD represents the average diffusion, whilst AD and RD respectively reflect diffusion along the main axis and perpendicular to it [156].Although there is not a precise correspondence between these measures and the underlying biological attributes, multiple aspects of tissue microstructure, including the number and size of axons, myelination, and membrane permeability may contribute to FA.Similarly, axon size and density may affect AD, whilst myelination may contribute to RD [127,156].As these indices vary during brain maturation and in pathological conditions [157,158], they are useful for case-control comparisons or to study age-related changes.between the two authors was resolved by a third author (VP).Data extracted from all studies in the systemic review included: sample size, demographic and clinical characteristics (i.e., age, sex, total intelligent quotient (IQ), ADHD presentation, comorbidities, and medication-naïve status); analytic method; significant case-control comparisons and associations between diffusion metrics and symptom severity or cognitive performance.Further, data on imaging data acquisition, preprocessing and analysis were extracted for the quality assessment (see below).Authors were contacted for missing data on imaging parameters included in the quality assessment.
For the meta-analyses, we extracted peak coordinates and their effect sizes for FA, MD, AD and RD contrasts.In studies not providing exact effect sizes for peak coordinates, the study threshold for significance was interpreted as the effect size, as in previous reports [11].In cases where peak coordinates and/or effect sizes were not provided, we contacted the corresponding authors to obtain the missing data.P-values of peak coordinates were converted to t-values using the anisotropic effect size signed differential mapping (AES-SDM) utility (http://www.sdmproject.com/utilities/?show=Statistics)[14].Peak coordinates are available as Supplementary material.

Quality assessment
In the absence of an established tool to rate the quality of DWI studies, criteria were identified based on published recommendations (listed in Supplementary Table S3).A traffic light system was then implemented, and studies were assigned a low/medium/ high-quality rating separately for imaging data acquisition, preprocessing, and analysis, as well as an overall rating based on the worse single rating.Two authors (SL and TTN) independently completed the quality assessment and discrepancies were resolved by a third author (JHZ).

Meta-analysis and meta-regressions
We ran meta-analyses for diffusion metrics that had peak coordinates available from at least five suitable studies.As in previous meta-analyses [11,12], we used the Signed Differential Mapping (SDM) software, version 6.21 [15] (https:// www.sdmproject.com/),to analyze regional differences in tract metrics between ADHD and TD control groups; and used the TBSS template for TBSS studies.Also consistently with prior studies [11,12], we used a random effects model, and the same statistical threshold that was previously applied (p < 0.005 at the voxel level with an extent threshold of 10 voxels).Meta-analyses were repeated in the pediatric sample (i.e., children/adolescents <18 years) and adults separately; and then in children (<12 years) and adolescents separately (Supplementary Table S2).We also conducted a post-hoc sensitivity analysis only including datasets judged of high quality.Finally, we ran meta-regression analyses to test the linear effects of age, percentage of males and medication-naïve subjects.We chose these variables because they have been associated with variation in white matter characteristics [16,17].Further, given that a previous report suggested that stimulant treatment could affect FA measures in children, but not adults [16], we tested the potential confounding effect of treatment exposure in the metaregression of age.

Systematic review
As shown in the PRISMA flow diagram (Fig. 1), from a pool of 956 possibly relevant references, we included 129 studies (96 in children, 25 in adults and 8 including both age groups), for a total of 6739 ADHD participants and 6476 controls.As many studies in pediatric samples included both children and adolescents, in the  narrative synthesis we aggregated them under 'children' (i.e.<18 years).However, in the meta-analysis, we also considered them separately (Supplementary Table S2).With regards to the imaging data analytic approach, the retained studies used one or a combination of the following: TBSS (43 studies), tractography (38 studies), ROI (22 studies), VBA (16 studies), network/graph analysis (17 studies) or other techniques (e.g., fixel-based analysis) (3 studies).Here, we first summarize the results of these studies as a narrative review, according to the topographical organization of the main brain connections [18].These can be grouped in projection (cortico-subcortical) pathways, which relay sensorymotor information; association (intra-hemispheric) pathways, which integrate functions of brain regions within the same hemisphere; and commissural (inter-hemispheric) pathways, which support information transfer between the two hemispheres [18].Associations between diffusion metrics and symptoms/ cognitive deficits are summarized in Figs. 2 and 3 and detailed in Supplementary material (page 5).Detailed characteristics of included studies are reported in Tables 1 and 2. Excluded studies are reported, with reasons, in the Supplementary material (page 3 and Supplementary Table S1).We then discuss the quality assessment and present the results of the meta-analyses and meta-regressions.
Cerebellar pathways: Eight studies reported alterations in the cerebellum or the middle cerebellar peduncle (MCP) in children with ADHD [19,20,25,32,37,49,64,83].Among these, reduced FA was identified by three VBA [19,32,49], two TBSS [25,37] and one ROI study [83].No study reported increased FA.Only two studies identified alterations in the cerebellum/MCP of adults with ADHD [56,74].In sum, reduced FA was the most consistently reported alterations within the cerebellum/MCP.Only one study reported reduced FA in the inferior cerebellar peduncle [37].Tract  c).Significant associations between tract metrics and cognitive deficits or symptoms are reported.Additional abbreviations: ATR anterior thalamic radiation, FS frontostriatal tract.
metrics were significantly associated with symptoms and cognitive performance (Fig. 2).
Finally, a tractography study in a mixed pediatric-adult sample reported reduced FA in the splenium of the CC [57], and another reduced FA in callosal fibers [60].Overall, reduced FA was the most consistently reported alteration, especially in the splenium of the CC.Tract metrics were significantly correlated with both clinical and neuropsychological characteristics (Fig. 3).
Conversely, one VBA study reported increased FA [53] and one ROI study both increased and reduced FA [74].Overall, both increased and reduced FA in the cingulum have been reported, bilaterally or in either hemisphere with equal frequency.Diffusion characteristics have been associated with variation in both clinical and cognitive profiles (Fig. 3).[37,38,43,53,71,74,82,87,94,97].Of these, TBSS studies either reported increased [43,94] or decreased FA [37,82] in ADHD children.In adults, two VBA study reported increased FA [53,71], whilst a tract-based analysis observed reduced FA [97].Overall, main differences were observed either bilaterally or in the left uncinate, and were associated with inattentive-emotional symptoms, and cognitive deficits (Fig. 3).

Quality assessment
Among the included studies, 68.7% were judged to be of low overall quality, mainly due to factors related to pre-processing (rated low-quality in 54% of studies) and/or acquisition parameters (rated low-quality in 46% of reports).Low-quality ratings for preprocessing were mainly due to lack of motion correction/ quality checks; those for acquisition to the use of non-isotropic voxels or lack of information.The quality assessment of individual studies is reported in Supplementary Tables S4 and 5.
Meta-analyses and meta-regressions Meta-analysis of TBSS studies comparing ADHD versus controls.Of the 43 TBSS studies included in the systematic review, 25 had available peak coordinates for case-control differences in FA and 18 in MD from non-duplicated datasets (see Supplementary Table S2).Therefore, these studies were included in two separate metaanalyses.The first meta-analysis included 25 TBSS studies, for a total of 32 datasets (26 in children and 6 in adults), comparing FA between 1348 individuals with ADHD and 1354 TD controls.As shown in Fig. 4 and Supplementary Table S6, the SDM analysis identified two clusters showing reduced FA values in the ADHD group compared with the TD control group.The right splenium of the CC, extending to the posterior cingulum, showed the most significant effect size and cluster extent (Fig. 4).The right body of CC exhibited the second-largest effect size and cluster extent.No clusters exhibited increased FA values in the ADHD group as compared with TD controls.The second meta-analysis included 18 studies, for a total of 23 datasets (19 in children and 4 in adults), comparing MD between 1051 participants with ADHD and 1101 controls, and did not identify any significant clusters.Due to the limited number of studies reporting peak coordinates for AD and RD (three and four, respectively), it was not possible to carry out metaanalyses for these metrics.Finally, the systematic review identified 16 studies that used a VBA approach.However, among these, only one study with available peak coordinates was not included in the previous review by Aoki et al., 2018 [11].Therefore, we did not rerun the meta-analysis based on our pre-published protocol.
Meta-analyses of TBSS studies in children and adult samples.The meta-analyses of TBSS studies investigating case-control differences in FA in the pediatric sample (<18 years) showed no significant clusters.Similarly, we did not observe significant clusters when we separated children (<12 years) and adolescents.By contrast, the meta-analysis including only adult studies identified five clusters showing reduced FA in individuals with ADHD as compared to controls (Supplementary Table S7).The right splenium of the CC exhibited the largest effect size and cluster extent, and the right ATR showed the second-largest effect size and cluster extent.No clusters exhibited increased FA values in adults with ADHD as compared to controls.No significant cluster was identified when repeating the meta-analysis on casecontrol comparisons in MD separately in children and adults.
Meta-analysis of high-quality studies.In the sensitivity analysis considering only the 6 TBSS datasets rated of high quality (Supplementary Table S2), we did not observe significant differences in FA between ADHD and TD controls.
Meta-regression analyses.We ran three meta-regressions including the 25 TBSS studies (32 datasets) comparing FA between ADHD and Fig. 4 Results of the meta-analysis of TBSS studies.The meta-analysis of TBSS studies comparing fractional anisotropy (FA) between individuals with ADHD (any age) and typically developing (TD) controls showed that individuals with ADHD had lower FA in the splenium of the corpus callosum (CC), extending to the posterior cingulum, and in the body of the CC.Peak coordinates are reported in Supplementary Table S6.
TD controls.The first meta-regression tested the linear influence of age and identified two clusters with significant negative associations.In the right splenium and body of the CC, the differences between groups in reduced FA amplified with advancing age (Supplementary Table S8).No clusters displayed significant positive associations with age.These findings held after including treatment exposure as additional regressor (Supplementary Table S9).Metaregressions accounting for the percentage of medication-naïve participants or the percentage of males included 23 studies (30 datasets) and 24 studies (31 datasets), respectively.We observed that ratios of medication-naïve/medicated individuals and males/ females had no significant influence on FA.

DISCUSSION
We conducted the most comprehensive systematic review of DWI studies in the field of ADHD, meta-analyzing studies amenable to quantitative synthesis.The systematic review identified widespread alterations (mainly reduced FA) in individuals with ADHD as compared to controls, mostly in the fronto-striatal pathways, cingulum, and CC.The meta-analysis of TBBS studies included 32 datasets (of which six in adults) and identified the most consistent FA reduction in the right splenium (extending to the posterior cingulum), followed by the body, of the CC.Meta-regressions showed that these effects were not affected by sex or exposure to ADHD medication.However, lower FA was related to older age, and case-control differences did not survive in the pediatric metaanalysis.Conversely, the meta-analysis in adults mainly identified reduced FA in the right splenium of the CC and ATR.
Alterations in the splenium and the body of the CC are in line with their roles in supporting cognitive and motor functions affected in ADHD.The splenium of the CC connects the occipital, temporal and posterior parietal lobes of the two hemispheres [105] and has been associated with visuospatial information transfer, processing speed, IQ, and behavior [106].Posterior cortical regions underpin attention and fronto-parietal cognitive control networks [107,108], and thus contribute to cognitive functions, such as attention, working memory and executive control, which are commonly impaired in ADHD [109,110].Further, the body of the CC primarily connects premotor, supplementary and primary motor cortices between the two hemispheres, and contributes to the modulation of motor activity [105].Our findings are in line with previous meta-analyses of structural magnetic resonance imaging (MRI) data reporting consistent volumetric reductions in the splenium of the CC in individuals with ADHD as compared to controls [111,112]; although more recent meta-and mega-analyses observed significant case-control differences only in children [113,114].Further, functional MRI meta-analyses have identified reduced parietal and temporal activations during cognitive control, attention and timing tasks [115][116][117][118].Of note, the cluster located in the splenium of the CC extended to the cingulum, which connects regions subserving the default mode network (DMN) [119].There is fMRI evidence for a poor anticorrelation between the DMN and task-positive networks, such as the fronto-parietal control and ventral attentive networks [120,121].This, according to the default mode interference theory of ADHD [122], may cause lapses of attention during cognitively demanding tasks [123,124].Notably, a prior study reported that white matter disruption in the splenium and body of the CC was associated with decreased resting-state functional connectivity in the DMN posterior cingulate cortex [125].Taken together, our findings and prior studies suggest that altered anatomical connectivity within the splenium and body of the CC may disrupt the function of brain networks supporting cognitive and motor functions affected in ADHD or their interaction with the DMN.Importantly, our findings not only confirm but also extend those of previous meta-analyses of DWI studies.Although the first meta-analysis combining nine VBA and TBSS studies reported more consistent alterations within fronto-striato-cerebellar connections, subsequent meta-analyses mainly identified altered FA within commissural fibers [11,12,126].The first included ten TBSS studies and reported reduced FA in the splenium of the CC, right sagittal stratum and left tapetum, extending to the cingulum, ILF and IFOF [12].The second performed two separate meta-analyses, including 12 TBSS and 13 VBA studies respectively.It confirmed a consistent reduced FA in the CC, in addition to the ILF, IFOF and SLF.However, the meta-analysis of VBA studies also identified regions of increased FA, e.g., in the midcingulate and anterior CC [11].Finally, a more recent meta-analysis combining 24 TBSS and VBA studies confirmed a consistent FA reduction in the splenium of the CC, extending to the body and right posterior corona radiata [126].The high consistency among the more recent metaanalyses may be due to their largely overlapping pool of included studies (e.g., almost 90% of articles included in Zhao et al., 2022 [126] were included in Aoki et al., 2018 [11]).Our findings from 32 TBSS datasets further support the consistently reported callosal alterations in ADHD extending to the cingulum.Different mechanisms may lead to reduced FA, such as altered myelination, axonal density/diameter or fiber crossing [127,128].As FA is a composite measure, we encourage future studies to consistently report additional metrics, e.g.AD and RD, as this may help understand whether white matter alterations are primary (e.g.related to myelination) or secondary to those in the gray matter from which the tract origins from (e.g.number of neurons).Both these mechanisms have been suggested as plausible in relation to ADHD pathophysiology [113,129].Further, both genetic and environmental factors may potentially contribute to the observed alterations.For instance, a recent genome-wide association (GWA) meta-analysis in ADHD identified, among others, genes related to myelination [130].However, their patter of methylation, which may reflect environmental adversity, affect ADHD symptoms trajectories [131].Thus, genetic and environmental factors may interplay to cause white matter abnormalities, which in turn are associated with ADHD symptoms.Overall, there are likely multiple alternative pathophysiologic pathways underpinning ADHD and brain alterations are not necessarily causal to symptoms, but might be co-occurrent manifestations or consequence of ADHD behaviors.
As reported by Chen et al. [12], we also observed a negative association between FA and age, with lower FA values in older individuals.However, in this prior meta-analysis, findings in the splenium of the CC survived in the pediatric meta-analysis, in contrast to our study.The absence of significant findings in our meta-analysis restricted to children is somehow unexpected given that prior structural MRI meta-and mega-analyses reported significant case-control volumetric and morphometric differences in children but not in adults with ADHD [113,114].For instance, reduced surface area and cortical thickness were identified in fronto-cingulate-temporo-occipital regions of ADHD children, but not in adults, compared to controls [114].However, we believe that our findings could provide new important insight into the pathophysiology of ADHD across the lifespan.A possible explanation of differences between prior and our results might be related to the distinct developmental trajectories of the gray and white matter.While gray matter structural measures increase and reach their peak in childhood (~age 2 for cortical thickness; ~age 6 for gray matter volume; and ~11-12 years for surface area/ cerebral volume) and then decrease in a curvilinear fashion [132]; FA in the CC increases in a curvilinear fashion with age, peaking between 21 and 29 years [133,134].Hence a delayed, less steep increase in FA with age in ADHD compared to healthy subjects would be reflected in larger case-control differences in older subjects; whilst a less steep decrease in gray matter measures would be more pronounced in childhood.The reduction in FA we observed in the splenium and body of the CC of individuals with ADHD could hence reflect a delay in white matter maturation, parallel to that observed in posterior cortices [135], which are interhemispherically connected by the posterior CC.However, longitudinal studies investigating developmental trajectories are needed to clarify the course of brain structure and connectivity alterations in ADHD.
We also observed a lack of spatial convergence in the sensitivity analysis only including high-quality studies.These findings need to be interpreted with caution because, although the quality assessment was based on published recommendations, there is not an available, established gold standard quality rating tool for DWI studies.Further, this meta-analysis only included six studies.Nevertheless, these findings suggest that the case-control differences identified in DWI meta-analyses might be influenced by non-optimal acquisition parameters or preprocessing.Future studies should be encouraged to follow the existing recommendations on imaging data acquisition, preprocessing, and analysis (referenced in Supplementary Table S3).For example, studies should (1) opt for isotropic voxels to avoid bias; (2) include sufficient gradient directions to ensure rotation invariance and improve precision; (3) spread multiple gradients unweighted among weighted volumes in the scan for similar noise profiles; (4) report sufficient acquisition and preprocessing information; (5) include additional scans for distortion correction; and ( 6) perform data quality control on the motion and report motion thresholds.Further, research in ADHD may benefit from methodological advances in the field.For instance, higher encoding resolution may improve sensitivity to signal and biophysical properties [136]; acceleration techniques may improve applications to younger populations [137]; and newer preprocessing methods may handle more types of artefacts [138].Overall, methodological developments may improve the study quality and reliability of findings, thus fostering our understanding of brain microstructure and connectivity in ADHD.
Prior meta-analyses of ADHD studies in other imaging modalities also reported no significant spatial convergence [139,140], and related this to methodological differences among studies and the high clinical and neurobiological heterogeneity of ADHD.In support to the latter observation, our systematic review indicates that differences may exist between ADHD presentations, females and males, treated versus untreated individuals, and participants with/without comorbidities.Further, a recent metaanalysis has shown that FA reduction in the splenium of the CC is a common feature of both ADHD and ASD, although the latter is characterized by additional white matter alterations in frontostriatal pathways [126].These findings, together with the limited converging results in the whole sample meta-analysis and the absence of significant results in the pediatric meta-analysis, should encourage future studies to extend their investigations beyond case-control differences and determine whether subgroups of individuals with ADHD could be discerned based on white matter characteristics.The need to parse neuroanatomic heterogeneity has also been raised by studies in other MRI modalities [141], as this may improve our understanding of ADHD pathophysiology.Further, the stratification of the heterogenous ADHD population based on differences in the underlying neuroanatomy may pave the way to the development of new more targeted treatments (Parlatini et al., under review).

Limitations
Some limitations of this work should be considered, mostly related to limitations of the included studies.For instance, most studies included in the narrative review used traditional tensor-based methods, and only a minority included other imaging (e.g., q-ball) and analysis (e.g., graph theory) techniques.We encourage the use of state-of-the-art methods as this can also advance our understanding of ADHD.As in previous meta-analyses, we used a coordinate-based approach rather than the original t-statistic maps, as these are not publicly available, but this may limit the accuracy of the results [14].We conducted meta-analyses of TBSS studies but did not replicate the one of VBA studies, according to our pre-published protocol, as we could only have included one additional study compared to Aoki et al. [11].This reflects the general tendency in the current literature towards the use of TBSS as compared to VBA approaches.As in this previous work [11], we preferred not to combine TBSS and VBA approaches in a single meta-analysis because this would violate the assumption under the null hypothesis that the expected FA differences are equal at every voxel, and because there is evidence that the two approaches may produce non-converging findings [11].We conducted separate meta-analyses of FA and MD, but could not investigate other diffusion metrics (e.g., RD and AD) due to the limited number of studies reporting them.We contacted authors for missing data, but peak coordinates were not available for eight studies.Our meta-analysis included a much greater number of studies than prior syntheses, which enhances robustness of findings; however, we encourage future investigations to provide full details of their results to limit potential reporting bias.
The studies included in the systematic review/meta-analyses mostly recruited small samples and were heterogeneous in terms of clinicodemographic characteristics.They included varying proportions of males and females and different ADHD presentations.Most studies included subjects previously exposed to medication, which can represent a potential confounder in connectivity measurements [16].Our metaregression analysis did not show any significant effect of previous exposure to stimulants; however, longitudinal studies in large samples are needed to disentangle the effects of development and treatment.Further, only few studies restricted the recruitment to comorbidity-free individuals.Most included different disorders or did not provide information, thus we were unable to control for comorbidities.This heterogeneity may have potentially influenced consistency among studies and our findings.For instance, about one fifth of individuals with ADHD also have ASD [142].However, comorbidity was not officially allowed until the DSM-5 was published (2013) [1]; therefore, the most recent studies and those in pediatric samples may be enriched with comorbid cases.This may also have potentially contributed to the lack of convergent findings in our pediatric submeta-analysis.For instance, recent meta-analyses highlighted both shared and specific alterations in individuals with ADHD and ASD [126,143].One of these identified reduced FA in the CC of both ADHD and ASD individuals, but also clusters of increased FA in those with ASD [126].This may have confounded and could be responsible for the negative findings in children with ADHD.Finally, comorbidities, such as affective and substance use disorders, are more common in adults with ADHD [5] and could also have confounded the results.Restricting recruitment to comorbidity-free individuals does not reflect daily clinical practice and limits our understanding of the biological basis of brain disorders.An alternative for future studies is that of taking a dimensional approach, in line with the National Institute of Mental Health Research Domain Criteria (RDoC) framework [5,144].In our meta-analysis, we considered the potential effect of sex and treatment exposure on the findings.However, the number of studies comparing ADHD presentations was limited, thus we could not meta-analyze them, and this should be subject of future investigations.Finally, we separately analyzed studies in children and adults, which allowed us to identify adult-specific alterations; however, the number of studies in the adult population was limited, thus findings should be interpreted with caution.Nevertheless, the association we observed between callosal FA and age, as well as findings from previous structural MRI studies, should encourage the longitudinal investigation of differences in the developmental trajectories of the white and gray matter in ADHD.

CONCLUSIONS
Clinicodemographic and methodological differences among studies are major barriers to our understanding of the neurobiology of ADHD.Future studies are needed to disentangle the potential biological differences related to sex, age, presentations, and comorbidities.They may also investigate tracts that have so far received less attention, such as the cerebellar peduncles, especially in adults.Finally, methodological improvements are recommended, especially optimizing imaging parameters, controlling and reporting for head motion, as well as efforts to enhance the comparability among studies, e.g., by using anatomical atlases to identify ROIs and tracts.While currently DWI studies do not have direct clinical implications in the field of ADHD [145], our findings may help future studies to stratify individuals with ADHD according to the underlying pathophysiology, which may guide the development of more tailored treatments.To this purpose, future study may also benefit from the combination of multiple imaging approaches, e.g., fMRI, functional Near Infrared Spectroscopy (fNIRS), and positron emission tomography (PET), as these can provide complementary information on functional and metabolic changes associated with ADHD.

Fig. 2
Fig. 2 Projection pathways and associated impairment.This figure displays the distinct components of the frontostriatal and thalamofrontal connections (panel a); cerebellar pathways (panel b); and the corticospinal tract (panel c).Significant associations between tract metrics and cognitive deficits or symptoms are reported.Additional abbreviations: ATR anterior thalamic radiation, FS frontostriatal tract.

Fig. 3
Fig. 3 Association and commissural pathways and associated impairment.This figure shows the three branches of the superior longitudinal fasciculus (SLF) (Panel a); the cingulum (CING), uncinate (UNC), inferior longitudinal fasciculus (ILF) and inferior fronto-occipital fasciculus (IFOF) (Panel b); and the main subdivisions of the corpus callosum (CC) (Panel c).Significant associations between tract metrics and cognitive deficits or symptoms are reported.

PRISMA 2020 flow diagram for new systemaƟc reviews which included searches of databases, registers and other sources
Results for each database are reported in Supplementary material, page 2 b) References of excluded studies are reported, with reasons, in Supplementary material, page 3 c) References of included studies are reported in Supplementary material, TableS2 and page 20

Table 1 .
Diffusion imaging studies in children with ADHD.

Table 2 .
Diffusion imaging studies in adults or mixed pediatric/adult samples with ADHD.