Functional Connectivity and Complexity Analyses of Resting-State fMRI in Pre-Adolescents Demonstrating the Behavioral Symptoms of ADHD

Abstract

Attention deficit hyperactivity disorder (ADHD) has been characterized by impairments among distributed functional brain networks, e.g., the frontoparietal network (FPN), default mode network (DMN), reward and motivation-related circuits (RMN), and salience network (SAL). In the current study, we evaluated the complexity and functional connectivity (FC) of resting state fMRI (rsfMRI) in pre-adolescents with the behavioral symptoms of ADHD, for pathology-relevant networks.

We leveraged data from the Adolescent Brain and Cognitive Development (ABCD) Study. The final study sample included 63 children demonstrating the behavioral features of ADHD and 92 healthy control children matched on age, sex, and pubertal development status. For selected regions in the relevant networks, ANCOVA compared multiscale entropy (MSE) and FC between the groups. Finally, differences in the association between MSE and FC were evaluated.

We found significantly reduced MSE along with increased FC within the FPN of pre-adolescents demonstrating the behavior symptoms of ADHD compared to matched healthy controls. Significant partial correlations between MSE and FC emerged in the FPN and RMN in the healthy controls however the association was absent in the participants demonstrating the behavior symptoms of ADHD.

The current findings of complexity and FC in ADHD pathology support hypotheses of altered function of inhibitory control networks in ADHD.

1. Introduction

Attention deficit hyperactivity disorder (ADHD) is characterized by a persistent pattern of inattention and/or hyperactivity and impulsivity that causes impairment in functioning and development. ADHD occurs worldwide and is estimated to affect 5% of children and approximately 2.5% of adults (Polanczyk et al., 2007). Although numerous studies have been conducted on ADHD, the psychopathology of the disorder is still not fully understood. In recent years, ADHD has been increasingly viewed as impairments among distributed functional networks or circuits (Polanczyk et al., 2007). Neuroimaging has been used to identify networks, which consist of brain regions that show synchronized, correlated fMRI signal dynamics (Biswal et al., 1995; Fox et al., 2009). These functionally connected networks (FCNs) are believed to reflect distinct mental states or processes (Friedman & Rapoport, 2015; Paloyelis et al., 2007). Aberrant communication between or within FCNs may reflect deficits in cognitive and affective functioning (Kessler et al., 2014; Zang et al., 2007).

In ADHD, altered functional connectivity has been reported in the frontoparietal network (FPN), default mode network (DMN), reward and motivation-related circuits (RMN), and salience network (SAL; Cai et al., 2018; Castellanos & Aoki, 2016; Posner et al., 2014; Sripada et al., 2014). Importantly, FCN findings in ADHD are inconsistent across different studies and age groups. Three recent meta-analyses summarized the rsfMRI functional connectivity findings of ADHD vs. healthy controls based on seed-based analyses (Cortese et al., 2021; Gao et al., 2019; Sutcubasi et al., 2020). With respect to the seeds in FPN, Gao et al. (2019) reported hyperconnectivity to the left orbital frontal cortex, caudate, putamen, and insula and hypoconnectivity to the precentral gyri among those with ADHD. Gao et al. (2019) found ADHD was associated with hyperconnectivity between the seeds in DMN and the left superior temporal gyrus and right supramarginal gyrus extending to the right angular gyrus. ADHD also showed hypoconnectivity between the seeds in DMN to the left middle frontal gyrus and subcallosal cingulate cortex. For the seeds in RMN, hyperconnectivity to the dorsolateral prefrontal cortex was observed in ADHD. Sutcubasi et al. (2020) reported somewhat similar results in the FPN and DMN but no agreement for RMN. In terms of the seeds in the FPN, ADHD showed elevated connectivity to the anterior prefrontal cortex, while for the seeds in the DMN, they found reduced connectivity to the dorsal posterior cingulate cortex but elevated connectivity to the dorsomedial prefrontal cortex in ADHD. Cortese et al. (2021) did find no significant spatial convergence of ADHD-related hyperconnectivity or hypoconnectivity across studies. However, these inconclusive patterns of hyper and hypoconnectivity in ADHD also persist when focusing on children and adolescents only (Gao et al., 2019; Sutcubasi et al., 2020) but suggesting a more widespread connectivity alteration as compared to adult ADHD groups. Additionally, they both reported altered functional connectivity between seeds in the SAL and DMN in children and adolescents. The findings in this subgroup could potentially be attributed to neurodevelopmental changes or delays in functional connectivity from childhood to adulthood (Fair et al., 2009).

Alongside functional connectivity analyses that assess the signal coherence between nodes, nonlinear analyses of neural signals from fMRI that characterize the signal complexity within a node have been proposed as measures for information processing capacity of brain areas and networks (McDonough & Nashiro, 2014; Smith et al., 2014; Wang et al., 2018; Wang et al., 2014), or indices of pathological brain function (Fernández et al., 2013; Sun et al., 2020; Takahashi, 2013). Sample entropy (SampEn; Richman & Moorman, 2000) has attracted considerable attention in complexity analysis because of its simplicity and reduced dependency on time series length compared to other forms of entropy such as approximation entropy (Pincus, 1991). SampEn measures the randomness and predictability of a stochastic process and generally increases with greater complexity. Multiscale entropy (MSE) was subsequently introduced to differentiate complex processes from random fluctuations more accurately, by calculating SampEn of a signal at multiple coarse-grained time scales (Costa et al., 2002). Because MSE is evaluated across different time scales, it is also capable to identify frequency-dependent neuropathophysiological processes in different brain regions. Two studies investigated the complexity of fMRI data in adults with ADHD (Guan et al., 2023; Sokunbi et al., 2013) and both reported reduced complexity in the resting state in the frontal cortical areas, but no study as of yet has investigated fMRI complexity differences in children or pre-adolescents when ADHD is most likely to be initially diagnosed. Additionally, no study has assessed the relation between functional connectivity and MSE (McDonough & Nashiro, 2014; Wang et al., 2018) in pre-adolescents with ADHD.

In the current study, we aimed to address this critical gap in the extant literature and evaluated the functional connectivity and complexity of rsfMRI in pre-adolescents who demonstrated behavioral features of ADHD, within the four main networks commonly associated with ADHD: FPN, DMN, RMN, and SAL. We combined MSE complexity analysis with functional connectivity analysis to elucidate the functional alterations present in those with behavioral features of ADHD on network nodes, at node-to-node edges, and on nodal complexity and node-to-node connectivity interaction levels. According to the literature, we hypothesized that compared to the healthy controls, pre-adolescents with ADHD symptoms would demonstrate lower MSE in the FPN and altered functional connectivity in the FPN, DMN, RMN, and SAL. Due to the lack of previous studies, we did not postulate a hypothesis in terms of the difference between children demonstrating behavioral features of ADHD and the control group on the association between MSE and functional connectivity.

2. Methods

2.1. Participants

We leveraged the baseline demographic, clinical, T1 structural, and rsfMRI data from the Adolescent Brain and Cognitive Development (ABCD) Study (Casey et al., 2018). The ABCD study is the largest pediatric brain imaging study in the United States, involving 21 research sites across the country. The ABCD baseline included 11878 participants in total, and each was evaluated based on 144 psychiatric diagnoses. In the first full data release, ADHD diagnoses were made according to DSM-5 criteria (American Psychiatric Association. & American Psychiatric Association. DSM-5 Task Force., 2013), according to parent reports by the Kiddie-Schedule for Affective Disorders and Schizophrenia (KSADS)-parent report. 1087 baseline participants were diagnosed with ADHD (i.e., Attention-Deficit/Hyperactivity Disorder Present). We excluded the individuals from the current study if they were diagnosed with any other psychiatric comorbidity. The final sample was reduced to 63 subjects (age = 9.91 ± 0.640 months; pubertal development status [PDS] = 1.665 ± 0.542; 25 female/38 male), each with the T1 structural data and at least two complete rsfMRI scans. However, due to an anomaly within the ABCD Study infrastructure, what was originally confirmed to represent ADHD diagnoses was subsequently found to be invalid, on the grounds that parents were asked whether “symptoms interfere with social, academic or occupational functioning”. Full DSM-5 criteria mandate that the impairment caused by symptoms extends to at least two domains. As such, full threshold DSM-5 diagnoses could not be confirmed, and our sample here comprised subjects who met all the behavioral criteria for ADHD diagnoses, without knowing whether the impairment extended to multiple domains. Therefore, our sample related to children with all behavioral features of ADHD.

We extracted data for 92 healthy controls (age = 10.009 ± 0.651 months; PDS = 1.687 ± 0.522; 29 female/63 male) to match the ADHD sample in terms of age (t = 0.859, p = 0.392), sex (χ² = 0.767, p = 0.381), and PDS (t = 0.255 p = 0.799; see Table 1).

Table 1.

An overview of sample characteristics delineated by group.

	ADHD group (N = 63)		Control group (N = 92)		Statistic	p value
Sex (f/m)	25/38		29/63		χ2 = 0.767	0.381
	Mean	Standard deviation	Mean	Standard deviation
Age (years)	9.910	0.640	10.009	0.651	t = 0.859	0.392
Pubertal development status	1.665	0.542	1.687	0.522	t = 0.255	0.799
ADHD score	5.412	2.152	1.228	1.570	t = −13.995	< 0.001

Pubertal development status was measured by the self-reported pubertal development scale (Petersen et al., 1988). The ADHD score was measured by the parent-reported ADHD CBCL-DSM5 scale (Achenback & Rescorla, 2001). ADHD group refers to children demonstrating behavioral symptoms of ADHD.

2.2. Imaging Acquisition, Preprocessing, and Denoising

All T1-weighted scans were acquired with voxel resolution = 1mm³, 256 × 256 matrix, flip angle = 8°, and 2x parallel imaging. Other scan parameters slightly varied by scanner platform, i.e., Siemens Prisma, Philips, or GE 3T scanner¹⁶. Each participant had 3–4 eyes-open (passive crosshair viewing) rsfMRI scans, each of which was approximately 5-minutes in duration. All rsfMRI scans were collected using a gradient-echo EPI sequence of 383 volumes in total (voxel resolution = 2.4 × 2.4 × 2.4 mm³, 60 slices, 90 × 90 matrix, FOV = 216 × 216, TR = 800 ms, TE = 30 ms, flip angle = 52°, 6-factor multiband acceleration). Head motion was monitored during scan acquisition using real-time procedures to adjust scanning procedures as necessary.

For the current study, we used rsfMRI run 1 and run 2. Spatial and temporal image preprocessing was performed in the CONN toolbox (Conn: fMRI functional connectivity toolbox): Default parameter setup was used including motion-realignment, coregistration and normalization to MNI152 template space and smoothing with a 6mm FWHM Gaussian kernel.

2.3. Data Analysis

2.3.1. Complexity Analysis

Voxel-wise multiscale entropy (MSE) was computed using the in-house developed LOFT Complexity Toolbox (github.com/kayjann/complexity) for each preprocessed rsfMRI sequence for each participant. SampEn is defined as the natural logarithm of the conditional probability that a pattern length of m points will repeat itself, excluding self-matches, for m + 1 points within a tolerance of r in a time series of length N (Richman & Moorman, 2000). Multiple “coarse-grained” time series of lengths N/1, N/2, …, and N/a were formed by averaging consecutive data points of increasing length. SampEn for each coarse-grained time series is calculated and MSE is the average of SampEn over all the temporal scales.

In the current MSE calculation, we set the pattern length m = 2, the sensitivity threshold r = 0.3, and the number of temporal scales a = 15. MSE was computed for each voxel for each rsfMRI time series for each participant. The MSE map was then obtained by averaging MSE over the two rsfMRI runs for each voxel for each participant. Finally, mean MSE within each region-of-interest (ROI) of the four networks was calculated for each participant: ROIs based on the Harvard-Oxford atlas (Desikan et al., 2006) were selected in the middle frontal gyrus and superior frontal gyrus for the FPN, supramarginal gyrus, medial frontal cortex, and posterior cingulate gyrus for the DMN, frontal orbital cortex, caudate, putamen, and nucleus accumbens for the RMN, and insula the for SAL. The anterior cingulate gyrus sometimes is divided into subareas that have been associated with different functional networks. However, in the atlas used for this study, the entire anterior cingulate was used and thus we considered association with three networks: FPN, DMN, and SAL.

A factorial 2 (Group: Participants demonstrating behavioral features of ADHD, Control) by 2 (Sex: Male, Female) analysis of covariance (ANCOVA) on the mean MSE in each ROI was conducted with PDS (self-reported) and scan-site as covariates. We were only interested in the main effects of Group here and in this study’s other analyses. To reduce skewness and kurtosis (Bliss, Greenwood, & White, 1956), PDS scores were Rankit-transformed and then z-scored. These values were used as the continuous covariates in all the analyses in the current study. A Benjamini-Hochberg correction (Benjamini & Hochberg, 1995) was performed for each network individually. The same ANCOVA was repeated for SampEn at each temporal scale.

In order to evaluate the association between complexity and ADHD behavior symptoms, the partial correlation between MSE and CBCL DSM5 ADHD scores controlling for sex, PDS z-score, and scan-site was computed for the whole sample. We also computed the partial correlation between SampEn at each temporal scale and the CBCL DSM5 ADHD scores controlling for the same covariates for the whole sample. Note that the CBCL DSM5 ADHD scores were Rankit-transformed and then z-scored in this partial correlation analysis.

2.3.2. Functional Connectivity Analysis

A denoising process was implemented on the preprocessed rsfMRI data. This process included white matter, CSF, motion scrubbing, and motion realignment parameters. After denoising, the first-level analyses on functional connectivity were conducted for each participant across both runs using the CONN toolbox default setup. Time series were averaged over voxels in each ROI, correlated between each pair of ROIs using Pearson’s correlation analysis, and then Fisher’s Z-transformed. The same ANCOVA including PDS z-score and scan-site as covariates was then performed for functional connectivity (Fisher’s Z-transformed) using a seed-to-seed analysis. A Benjamini-Hochberg correction (Benjamini & Hochberg, 1995) was performed on the resultant main effects of Group for each network.

In order to evaluate the association between functional connectivity and ADHD behavior symptoms, the partial correlation between functional connectivity and CBCL DSM5 ADHD scores controlling for sex, PDS z-score, and scan-site was computed for the whole sample. Note again that the CBCL DSM5 ADHD scores were Rankit-transformed and then z-scored in this partial correlation analysis.

2.3.3. Analysis of the Association between Complexity and Functional Connectivity

To investigate the association between complexity and functional connectivity, we computed the partial correlation (i.e., Fisher’s z score) between functional connectivity (between each seed ROI and each other ROI) and MSE in each seed ROI for each group by controlling for sex, PDS z-score, and scan-site. A group comparison (Participants demonstrating behavioral features of ADHD vs. controls) was conducted on the partial correlations for each (seed-MSE, seed-to-seed connectivity) linkage. The same analysis was repeated for the partial correlation between SampEn at each temporal scale and functional connectivity.

2.3.4. Follow-Up Analyses

Medication effects.

In order to investigate the potential medication effects, a factorial 2 (Medication: medicated participants demonstrating behavioral features of ADHD, non-medicated participants demonstrating behavioral features of ADHD) by 2 (Sex: Male, Female) ANCOVA on the mean MSE in each ROI was conducted with PDS z-score and scan-site as covariates. The same ANCOVA was repeated for SampEn at each temporal scale and for functional connectivity at each seed-to-seed edge.

Test-retest reliability.

The intraclass correlation coefficient (ICC) is commonly used to measure test-retest reliability (Spitzer et al., 1992). In the current study, ICC of mean MSE in the whole brain as well as in each ROI across run 1 and run 2 was computed for each group. ICC is expressed as (BMS-EMS)/BMS+(s-1)EMS, modeled by two-way ANOVA with random subject effects and fixed session effects (Shrout & Fleiss, 1979). The BMS is the between-subject mean squared error, the EMS is the within-subject mean squared error, and s is the number of repeated sessions. For the current study, s = 2.

Additionally, the factorial 2 (Group: Participants demonstrating behavioral features of ADHD, Control) by 2 (Sex: Male, Female) analysis of covariance (ANCOVA) on the mean MSE in each ROI with PDS z-score and scan-site as covariates was performed separately for each run. Thereafter, to test the repeatability of these results, ICC of the F values across run 1 and run 2 was computed.

3. Results

3.1. Clinical Data

There were no group differences in age (t = −0.859, p = 0.392), sex (χ² = 0.767, p = 0.381), or PDS (t = −0.255, p = 0.799; see Table 1). As expected, children demonstrating behavioral features of ADHD scored significantly higher than the controls on the CBCL DSM5 ADHD score (t = 13.995, p < 0.001; see Table 1). There was no significant group difference in terms of the framewise displacement (FD; p > 0.05 for both runs; Power et al., 2014).

3.2. Complexity Analysis

The average MSE maps over the two rsfMRI runs for the children demonstrating behavioral features of ADHD and the control group are displayed in Figures 1(a) and (b). Qualitatively, it can be seen that complexity in the grey matter was higher than in the white matter. Moreover, children demonstrating behavioral features of ADHD appeared to have lower MSE than the controls, specifically in the frontal cortex. The ANCOVA revealed significant main effects of Group in multiple ROIs, i.e., the anterior cingulate gyrus (FPN, RMN, & SAL; F = 4.417, p < 0.05, ${\eta }_p^2=0.033$) bilateral middle frontal gyrus (FPN; F = 5.491 & 5.557, p < 0.05, ${\eta }_p^2=0.040\&0.041$, respectively), bilateral superior frontal gyrus (FPN; F = 4.423 & 4.831, p < 0.05, ${\eta }_p^2=0.033\&0.036$, respectively), and frontal medial cortex (RMN; F = 4.626, p < 0.05, ${\eta }_p^2=0.034$). The Benjamini-Hochberg correction was performed for each network separately (false discovery rate = 0.05) and all the ROIs in FPN survived the correction. Post-hoc t-tests indicated that the children who demonstrated behavioral features of ADHD had lower MSE values than the controls in all the ROIs in the FPN (MSE of children demonstrating behavioral features of ADHD vs. control = −0.078 to −0.042; t = −1.768 to −0.715; see Figure 1(c) and Table 2).

Figure 1.

(a) MSE map for the participants demonstrating behavioral features of ADHD. (b) MSE map for the control group. (c) t value map of MSE for the participants demonstrating behavioral features of ADHD vs. control. Only significant main effects of Group from ANCOVA were displayed, Benjamini-Hochberg corrected for each network (p < 0.05, false discovery rate = 0.05).

Table 2.

Brain regions displaying significant main effects of Group, obtained from the ANCOVAs for MSE and functional connectivity.

Regiona	F-value	lvalue	${\eta }_p^2$
For MSE
Anterior cingulate gyrus	4.417	−0.895	0.033
Left middle frontal gyrus	5.491	−1.396	0.041
Right middle frontal gyrus	5.557	−1.523	0.041
Left superior frontal gyrus	4.423	−1.768	0.033
Right superior frontal gyrus	4.831	−1.754	0.036
For functional connectivity
Anterior cingulate gyrus-to-right superior frontal gyrus	6.852	1.801	0.052
Anterior cingulate gyrus-to-left middle frontal gyrus	7.052	3.007	0.053

^aAccording to the Harvard-Oxford Atlas (Desikan et al., 2006). All results presented at p < 0.05, Benjamini-Hochberg corrected (false discovery rate = 0.05). ${\eta }_p^2=\text{partial}\text{eta}\text{squared}$.

Figure S1 in the supplementary material displays the t value of children demonstrating behavioral features of ADHD vs. control for mean MSE and SampEn at each temporal scale in each ROI which showed a significant main effect of Group (p < 0.05, uncorrected). It reveals that at scale 1 (frequency = 1.25Hz), every ROI showed a significant main effect of Group except the left anterior supramarginal gyrus. As the scale number increased, some significant main effects of Group disappeared although all the ROIs in the FPN remained till scale = 9 (frequency = 0.111Hz). At scale = 15 (frequency = 0.083Hz), there was no significant difference between the children demonstrating behavioral features of ADHD and the control group in terms of SampEn (p > 0.05).

We further tested whether there was an association between MSE and the CBCL DSM5 ADHD score but the partial correlation analysis taking into account sex, PDS z-score, and scan-site for the whole sample did not reach statistical significance in any ROI (p > 0.05). We also computed the same partial correlation between SampEn and the CBCL DSM5 ADHD score at each temporal scale and none was significant (p > 0.05).

3.3. Functional Connectivity Analysis

Figures 2(a) and (b) display the significant functional connectivity over the two rsfMRI runs for the children demonstrating behavioral features of ADHD and the control group, respectively, Benjamini-Hochberg corrected for each network (p < 0.05, false discovery rate = 0.05). It can be seen that most ROIs were strongly connected within each network. The ANCOVA revealed significant main effects of Group at two edges in the FPN after Benjamini-Hochberg correction (false discovery rate = 0.05), i.e., anterior cingulate gyrus-to-right superior frontal gyrus (FPN; F = 6.852, p < 0.05, ${\eta }_p^2=0.052$) and anterior cingulate gyrus-to-left middle frontal gyrus (FPN; F = 7.052, p < 0.05, ${\eta }_p^2=0.053$). Post-hoc t-tests indicated that the children demonstrating behavioral features of ADHD had stronger functional connectivity than the controls at these edges (functional connectivity of ADHD vs. control = 0.097 & 0.094; t = 1.801 & 3.007, respectively; see Figure 2 (c) and Table 2). For the uncorrected results, interested readers shall refer to Figure S2 in the supplementary material for the functional connectivity map for each group and the t-value map for children demonstrating behavioral features of ADHD vs. control and Table S1 in the supplementary material for the statistical summary of the main effects of Group (p < 0.05, uncorrected).

Figure 2.

Functional connectivity maps (Fisher’s z) for (a) the participants demonstrating behavioral features of ADHD and (b) the control group, and (c) the t-value map for the participants demonstrating behavioral features of ADHD vs. the control group. Only significant edges are displayed in blue or red colors, Benjamini-Hochberg corrected for each network (p < 0.05, false discovery rate = 0.05). For (a) and (b), blue/red indicates p < 0.05 for the t-test against 0. For (c), blue/red indicates p < 0.05 for the main effect of Group. The black indicates data was not available. SFG: superior frontal gyrus; MidFG: middle frontal gyrus; AC: anterior cingulate gyrus; aSMG: anterior supramarginal gyrus; pSMG: posterior supramarginal gyrus; MedFC: frontal medial cortex; PC: posterior cingulate gyrus; FOrb: frontal orbital cortex; Acc: Accumbens; r: right; l: left.

The partial correlations between functional connectivity and the CBCL DSM5 ADHD score controlling for sex, PDS z-score, and scan-site were significant at the edges of the left posterior supramarginal gyrus-to-the right caudate (r = 0.161, p = 0.045), the left posterior supramarginal-to-the left caudate (r = 0.160, p = 0.046), the frontal medial cortex-to-the right nucleus accumbens (r = −0.176, p = 0.028), the right frontal orbital cortex-to-the left caudate (r = 0.217, p = 0.007), and the right nucleus accumbens-to-right insula (r = 0.202, p = 0.012). However, none of them survived from Benjamini-Hochberg correction.

3.4. Analysis of the Association between Complexity and Functional Connectivity

Figures 3(a) and (b) display the significant partial correlation (Fisher’s z) between MSE (in a specific seed) and functional connectivity (at the edge of this seed to another ROI) for each group after controlling sex, PDS z-score, and scan-site, Benjamini-Hochberg corrected for each network (p < 0.05, false discovery rate = 0.05). Children demonstrating behavioral features of ADHD did not have (seed-MSE, seed-to-seed connectivity) linkages that were significantly different from zero. By contrast, the healthy controls had significant and positive partial correlation (Fisher’s z) between MSE and functional connectivity in a few linkages within the FPN while the associations were significant and negative in many linkages within the RMN. However, Benjamini-Hochberg corrected results (corrected for each network; false discovery rate = 0.05) indicated no significant between-group difference for any linkage (see Figure 3(c)). Please refer to Table S2 in the supplementary material for the summary of the uncorrected results.

Figure 3.

Maps of partial correlations (Fisher’s z) between MSE and functional connectivity for (a) the participants demonstrating behavioral features of ADHD, (b) the control group, and (c) the z value map of partial correlations for the participants demonstrating behavioral features of ADHD vs. control. Only significant (seed-MSE, seed-to-seed connectivity) linkages are displayed in blue or red colors, Benjamini-Hochberg corrected for each network (p < 0.05, false discovery rate = 0.05). For (a) and (b), blue/red indicates p < 0.05 for the t-test against 0. For (c), blue/red indicates p < 0.05 for the statistical comparison of partial correlations in the participants demonstrating behavioral features of ADHD against the control group. The black indicates data was not available. SFG: superior frontal gyrus; MidFG: middle frontal gyrus; AC: anterior cingulate gyrus; aSMG: anterior supramarginal gyrus; pSMG: posterior supramarginal gyrus; MedFC: frontal medial cortex; PC: posterior cingulate gyrus; FOrb: frontal orbital cortex; Acc: accumbens; r: right; l: left.

Figure S3 in the supplementary material displays the significant partial correlations (Fisher’s z) between MSE or SampEn at each temporal scale and functional connectivity for each group (p < 0.05, uncorrected; Figures S3(a) and (b)) and the z values of partial correlations for children demonstrating behavioral features of ADHD vs. control (only significant linkages are shown, p < 0.05, uncorrected; Figure S3(c)). Across all the temporal scales, children who demonstrated behavioral features of ADHD appeared to have fewer (seed-MSE, seed-to-seed connectivity) linkages within networks and between networks, significantly different from zero than the controls. Children demonstrating behavioral features of ADHD showed significantly lower partial correlations between SampEn and functional connectivity relative to the controls within the DMN at each temporal scale except scale = 1 (p < 0.05, uncorrected), specifically in linkages of (frontal medial cortex, frontal medial cortex-to-right anterior supramarginal gyrus), (frontal medial cortex, frontal medial cortex-to-left anterior supramarginal gyrus), (right anterior supramarginal gyrus, right anterior supramarginal gyrus-to-medial frontal cortex), and (left anterior supramarginal gyrus, left anterior supramarginal gyrus-to-medial frontal cortex). Children demonstrating behavioral features of ADHD showed significantly lower partial correlations relative to the controls in the linkage of (left putamen, left putamen-to-left nucleus accumbens) within the RMN up to scale = 6 (p < 0.05, uncorrected). Children demonstrating behavioral features of ADHD also showed significantly higher partial correlations relative to the controls within the RMN (p < 0.05, uncorrected): The linkage of (right putamen, right putamen-to-right caudate) was significant at each temporal scale; Some significant within-RMN linkages emerged at the higher scale number (scale > 10), i.e., (left putamen, left putamen-to-right caudate), (right caudate, right caudate-to-right putamen), and (right caudate, right caudate-to-left putamen). Children demonstrating behavioral features of ADHD showed a significantly higher partial correlation relative to the controls within the FPN only at scale = 1 in the linkage of (right superior frontal gyrus, right superior frontal gyrus-to-anterior cingulate gyrus) (p < 0.05, uncorrected). In terms of cross-network linkages, we observed that children demonstrating behavioral features of ADHD had a lower partial correlation between SampEn and functional connectivity than the controls in the linkages of (frontal medial cortex, frontal medial cortex-to-right putamen) and (frontal medial cortex, frontal medial cortex-to-left putamen) at almost every temporal scale up to scale = 13 (p < 0.05, uncorrected). Children demonstrating behavioral features of ADHD had a higher partial correlation than the controls in the linkages of (right insula, right insula-to-right anterior supramarginal gyrus), (left insula, left insula-to-right anterior supramarginal gyrus), and (right anterior supramarginal gyrus, right anterior supramarginal gyrus-to-right insula) at scale = 1, and in the linkage of (right putamen, right putamen-to-right middle frontal gyrus) at scale = 11 and 12 (p < 0.05, uncorrected).

3.5. Follow-Up Analyses

3.5.1. Medication Effects

In order to investigate potential medication-related effects, a factorial 2 (Medication: medicated participants demonstrating behavioral features of ADHD, non-medicated participants demonstrating behavioral features of ADHD) by 2 (Sex: Male, Female) ANCOVA on the mean MSE in each ROI was conducted with PDS z-score and scan-site as covariates. Please refer to Table S3 in the supplementary material for the list of prescriptions. Children demonstrating behavioral features of ADHD who consumed at least one prescription were categorized as medicated. The result indicated a main effect of Medication in the left anterior supramarginal gyrus (DMN; F = 4.563, p < 0.05, ${\eta }_p^2=0.186$). However, it did not survive the networkwise Benjamini-Hochberg correction (false discovery rate = 0.05). Therefore, we concluded there was no significant difference in MSE between the medicated children demonstrating behavioral features of ADHD and the non-medicated children demonstrating behavioral features of ADHD in any ROI. The same ANCOVA was replicated for SampEn at each temporal scale. Main effects were found in multiple regions at multiple temporal scales (p < 0.05, uncorrected). More details can be accessed on Page 1 in the supplementary material. However, none of them survived the Benjamini-Hochberg correction (false discovery rate = 0.05).

The same ANCOVA was replicated for functional connectivity at each seed-to-seed edge. It revealed significant main effects of Medication at three edges, i.e., left middle frontal gyrus-to-right frontal orbital cortex (FPN-to-RMN; F = 4.509, p < 0.05, ${\eta }_p^2=0.184$), medial frontal gyrus-to-posterior cingulate gyrus (DMN-to-DMN; F = 4.574, p < 0.05, ${\eta }_p^2=0.186$), and right anterior supramarginal gyrus-to-right insula (DMN-to-SAL; F = 6.604, p < 0.05, ${\eta }_p^2=0.248$). However, none of them survived the Benjamini-Hochberg correction (false discovery rate = 0.05).

3.5.2. Test-Rest Reliability

In practice, ICC values are usually divided into five common intervals: 0 < ICC ≤ 0.25 indicates poor reliability; 0.25 < ICC ≤ 0.4 indicates low reliability; 0.4 < ICC ≤ 0.6 indicates fair reliability; 0.6 < ICC ≤ 0.75 indicates good reliability, and ICC > 0.75 indicates very good reliability. ICC of mean MSE of the entire brain across run 1 and run 2 for the current whole sample revealed ICC = 0.860, F(154, 146) = 13.5, p = 0.000. For ROI-wise test-retest reliability in each group, we achieved ICC = 0.749 – 0.893, F(62, 63) = 7.132 −18.951, p = 0.000 for children demonstrating behavioral features of ADHD, and ICC = 0.792 – 0.930, F(91, 91.7) = 8.608 – 27.588, p = 0.000 for the controls (Figure 4). The results indicated very good test-retest reliability. For more details, please refer to the text on Page 1–2 in the supplementary material.

Figure 4.

ICC maps of MSE for (a) the participants demonstrating behavioral features of ADHD and (b) the controls. p < 0.001.

In order to further investigate test-retest reliability of complexity analysis, the factorial 2 (Group: Participants demonstrating behavioral features of ADHD, Control) by 2 (Sex: Male, Female) analysis of covariance (ANCOVA) on the mean MSE in each ROI with PDS z-score and scan-site as covariates was performed for each run separately. The results are displayed in Table S4 and Figure S4 in the supplementary material. Generally, children demonstrating behavioral features of ADHD showed reduced complexity in both runs, thus confirming the main results when averaged across runs. However, it can be observed that the group differences of mean MSE in the ROIs were stronger in run 1 than in run 2, indicated by greater F values and smaller p values in run 1. ICC of the F values across run 1 and run 2 was computed and the results revealed ICC = 0.302, F(20, 9.85) = 2.52, p = 0.069. ADHD’s greater increased restlessness in the later run (see Page 1–2 in the supplementary material) could potentially diminish the group difference in mean MSE in the ROIs and result in low test-retest reliability in the outcome of ANCOVA on mean MSE.

4. Discussion

In this study, we report two main observations: (1) reduced rsfMRI complexity in areas of the FPN in pre-adolescents who demonstrated behavioral features of ADHD as compared to the matched controls, and (2) increased functional connectivity between anterior portions of the FPN in pre-adolescents who demonstrated behavioral features of ADHD, relative to the control group. We did not observe a significant group difference in terms of the association between complexity and functional connectivity in any interested linkage.

The FPN has been hypothesized to play a role in instantiating and flexibly modulating cognitive control (Marek & Dosenbach, 2018). Cognitive control is commonly compromised in many forms of psychopathology, many of which emerge during childhood and adolescence, while the control function is refined throughout development. Emerging literature has suggested an important role of the FPN in the pathophysiology of ADHD. Previous structural MRI studies consistently reported that ADHD was associated with abnormal morphometry (Albajara Sáenz et al., 2019; Frodl & Skokauskas, 2012) and developmental trajectories (Shaw et al., 2012; Shaw et al., 2007) in the prefrontal and cingulate structures. Both qualitative review (Rubia, 2011) and meta-analyses (Cortese et al., 2012; Hart et al., 2013) on task-fMRI reported hypoactivation in the regions of FPN in ADHD across different tasks. rsfMRI meta-analyses also suggested functional alternation of FPN in ADHD (Gao et al., 2019; Sutcubasi et al., 2020). Specifically, Gao et al. (2019) reported children/adolescents with ADHD demonstrate hyperconnectivity between the seeds in the FPN and the dorsal anterior cingulate cortex, a finding that our results corroborated. In addition to altered functional connectivity in the FPN, the current study revealed significantly different complexity in pre-adolescents who demonstrated behavioral features of ADHD relative to the controls, thus extending existing literature and providing additional evidence for the hypothesis of altered cognitive control in ADHD.

The observation of significant reductions in MSE in rsfMRI time series in pre-adolescents who demonstrated behavioral features of ADHD was consistent with the findings in adults with ADHD diagnosis using the same modality (Guan et al., 2023; Sokunbi et al., 2013). Guan et al. (2023) observed significant decreases of SampEn in multiple ROIs in every investigated brain network under every atlas in subjects with ADHD compared to controls. They also observed reduced SampEn of the gray matter at every temporal scale under every atlas in ADHD. Sokunbi et al. (2013) reported significant reductions of SampEn in adults with ADHD relative to the controls in the superior frontal and anterior cingulate gyrus which was replicated for pre-adolescents who demonstrated behavioral features of ADHD in the current study (Figure 1(C)). Our results also aligned with existing literature in resting-state using other neuroimaging modalities such as EEG and MEG in which complexity was found lower in the participants with ADHD than in the healthy controls (Angulo-Ruiz et al., 2022; Fernández et al., 2009; Gu et al., 2022; Gómez et al., 2013; Hu et al., 2021; Rezaeezadeh et al., 2020).

While multiple previous studies support the conceptualization of ADHD as a dysconnectivity syndrome, findings relating to functional connectivity under rsfMRI have been highly heterogeneous in terms of the affected networks and network subcomponents, as well as the direction of effects. The current study observed enhanced connectivity within the FPN, i.e., the right superior frontal gyrus and left middle frontal gyrus using the anterior cingulate gyrus as the seed in pre-adolescents who demonstrated behavioral features of ADHD (Figure 2). This result was consistent with previous literature that reported enhanced connectivity between the anterior cingulate gyrus and middle frontal gyrus in children/adolescents with ADHD compared to healthy controls (Kumar et al., 2021; Lin et al., 2021; Liu et al., 2023) except for one study reporting decreased connectivity (Posner et al., 2013). Interestingly, this increased functional connectivity is predominantly between anterior nodes of the network and one might speculate that this represents a developmental stage of the network of increased short-range connectivity before long-range distributed network connections are established (Fair et al., 2009).

While an association between MSE and functional connectivity in healthy cohorts has been previously reported (McDonough & Nashiro, 2014; Wang et al., 2018), as far as we know, the current study is the first effort to conduct a comparison on the association between complexity and functional connectivity between healthy and diseased groups. While we found significant positive relationships between MSE and functional connectivity in the FPN and negative associations in the RMN in the healthy control group, such relationships were absent in the children who demonstrated behavioral features of ADHD (Figure 3). However, the direct statistical comparison did not reveal a significant group difference in the relation between MSE and functional connectivity between participants demonstrating behavioral features of ADHD and the controls. Future research in ADHD or other neuropsychiatric disorders following this approach might reveal disease and network-specific alterations in the association between nodal complexity and node-to-node connectivity, two complementary but interrelated network properties.

Several caveats should be considered with respect to the current results. First, an unforeseen anomaly in the parent ABCD dataset precluded us from being able to assess DSM-5 diagnoses of ADHD. While we originally conceived of the study as an assessment of DSM-5 ADHD, it recently emerged that parents participating in the parent ABCD study were not asked whether their child’s symptoms extended to multiple domains (i.e., social, academic, occupational functioning), which excluded the possibility of confirming DSM-5 diagnoses, which require impairment in at least two domains. As such, our sample was retrospectively reconceptualized to refer to children demonstrating all behavioral symptoms of ADHD, rather than DSM-5 diagnosed ADHD. Replication of these findings in a sample of children with full-threshold ADHD is required. Second, the anterior cingulate gyrus sometimes is divided into subareas that have been associated with different functional networks. For instance, the dorsal anterior cingulate gyrus is usually considered a part of SAL, and the rostral anterior cingulate gyrus is a part of DMN. However, in the atlas used for the current study, the entire anterior cingulate was used and thus we considered this region was associated with three networks (FPN, DMN, and SAL). As a result, the reported altered functional connectivity in children demonstrating all behavioral symptoms of ADHD within the FPN in the current study (i.e., anterior cingulate gyrus-to-right superior frontal gyrus and anterior cingulate gyrus-to-left middle frontal gyrus) indeed could also be considered as functional connectivity alteration between SAL and FPN or alteration between DMN and FPN. Third, in order to evaluate the effect of prescription, a factorial 2 (Medication: medicated participants demonstrating behavioral features of ADHD, non-medicated participants demonstrating behavioral features of ADHD) by 2 (Sex: Male, Female) ANCOVA on the mean MSE, SampEn at each temporal scale, and functional connectivity in each ROI was conducted with PDS z-score and scan-site as covariates. The result after correction indicated no significant main effects of Medication for MSE, SampEn at each temporal scale, and functional connectivity in any ROI. Fourth, in the case of children exhibiting behavioral features of ADHD, ICC ranged from 0.749 to 0.893 across run 1 and run 2, while for the control subjects, it ranged from 0.792 to 0.930. This suggested a high degree of reliability and reproducibility of MSE measurements taken on the same individuals during different rsfMRI scans. These findings indicated that the variations observed in MSE in the present study were due to the actual difference between subjects rather than random errors or measurement inconsistencies. However, the ICC for F values between two runs was 0.302, revealing low reliability for the main effects of Group derived from the ANCOVA on mean MSE. This inconsistency was suspected to be a consequence of greater increased restlessness among individuals with ADHD behavioral features during the later run compared with the controls, potentially diminishing group differences. Findings in previous test-retest reliability studies were diverse: Wang et al. (2014) investigated test-retest reliability on the whole brain mean SampEn and reported very good reliability, reflected by ICC = 0.967. Niu et al. (2020) computed ICC for region-based complexity measures using different resources of rsfMRI datasets and found ICC showed fair to good reliability for SampEn. Guan et al. (2023) computed ICC for various types of resting-state brain entropy across different numbers of rsfMRI scan repetitions under three different atlases. Their results revealed fair or worse than fair test-retest reliability in most cases. Further studies are needed to address the test-retest reliability issue on complexity. Fifth, it is worth mentioning that the patients included in the current work had no other previous or current psychiatric diagnoses. Many previous complexity studies in ADHD indeed examined ADHD with potential various comorbidities such as oppositional defiant disorder/conduct disorder that might confound the results. The current results removed this concern and further confirmed the finding of reduced complexity. Moreover, the lack of a unified understanding of functional connectivity in ADHD across literature might be due in part to different rates of comorbid conditions in those studies. The current finding of enhanced functional connectivity in children demonstrating behavioral features of ADHD relative to the controls in the FPN might potentially contribute to clarifying the mystery and provide a greater understanding of this field. Sixth, due to the limitation of networkwise Benjamini-Hochberg correction, the group differences in terms of cross-network functional connectivity and cross-network partial correlation between MSE and functional connectivity were absent. However, we showed significant cross-network group differences in Tables S1 and S2 and Figures S2 and S3 in the supplementary material. The reader should keep in mind that those effects were based on uncorrected statistics.

In conclusion, our analyses support the feasibility and utility of temporal signal entropy in investigating the changes in complexity of rsfMRI signals in pre-adolescents demonstrating behavioral features of ADHD when compared to healthy controls. We found reduced complexity in the anterior cingulate gyrus, bilateral middle frontal gyrus, and bilateral superior frontal gyrus in the FPN and greater functional connectivity in the right superior frontal gyrus and left middle frontal gyrus using the anterior cingulate gyrus as the seed in pre-adolescents who demonstrated behavioral features of ADHD. We suggest that combining analyses of complexity and functional connectivity is a useful and easily obtainable measure to reveal changes in ADHD brain dynamics.

Highlights.

• The current study evaluated the complexity and functional connectivity (FC) of resting-state fMRI in pre-adolescents demonstrating the behavior symptoms of ADHD for pathology-relevant networks using the Adolescent Brain and Cognitive Development (ABCD) Study data.

• We found significantly reduced multiscale entropy (MSE) and increased FC within the frontoparietal network for the pre-adolescents demonstrating the behavior symptoms of ADHD compared to matched healthy controls.

• Significant partial correlations between MSE and FC emerged in the FPN and RMN in the healthy controls however the association was absent in the participants demonstrating the behavior symptoms of ADHD.

• The current findings of complexity and FC in pre-adolescents demonstrating the behavior symptoms of ADHD support hypotheses of altered function of inhibitory control networks in ADHD.