Subjects

The experimental group consisted of 13 children with (C)APD (7 girls, 6 boys, aged: 7.3-16.0 years, mean = 12.2) and 15 control children (7 girls, 8 boys, aged: 7.0-16.0 years, mean = 11.7). There was a comparable number of girls and boys in each group and the groups were not significantly different with respect to chronological age (t = -0.5; p > 0.05).

Children with (C)APD were recruited from the patients of the Institute of Physiology and Pathology of Hearing, prior to their participation in a specifically-designed therapeutic programme serving to improve auditory processing and attention. The control group was recruited via an advertisement in local press. Control subjects obtained results within age-specific reference ranges in a (C)APD examination according to the Polish norms and did not have any history of neurodevelopmental disorders and/or learning problems.

(C)APD diagnosis was made by a multidisciplinary team comprising audiologists, speech-language therapists, psychologists, educators, and physicians. Children who had ADHD symptoms, language problems, psychiatric disorders and/or emotional disturbances were not included in the study.

All children participating in the study were right-handed, had normal hearing (as measured with tonal audiometry at the Institute of Physiology and Pathology of Hearing) and intellectual abilities within the normal range (Raven’s Progressive Matrices). They also had no history of neuropsychiatric disorders, head trauma or medication affecting the CNS. Subjects’ parents provided both oral and written informed consent in compliance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and guidelines of the Bioethics Committee of the Institute of Physiology and Pathology of Hearing. Participants were informed that they can withdraw from the study at any moment.

Imaging protocol

A 3 T Siemens Magnetom Trio Tim MR scanner equipped with a 12-channel matrix head coil and MR-compatible goggles (Nordic NeuroLab Visual system) were used to perform the imaging study. The scanning sessions included: 1) localization; 2) 3D T1-weighted image acquisition covering the entire brain (3D MP-RAGE sequence, TR = 1900 ms, TE = 2.26 ms, 0.9 × 0.9 × 0.9 mm voxels); 3) resting-state functional imaging (TR = 2000 ms, TE = 30 ms, flip angle = 90°, FOV 192 mm, matrix size 64 × 64, scanned volume-37 axial sequential slices of 3.5 mm thickness [no gap, 3 × 3 × 3.5 mm voxel]), AC-PC oriented, 150 volumes, iPAT = 2). The resting-state protocol was 5 minutes in duration, producing 150 brain volumes. During the study, the subjects were instructed to remain motionless, keep their eyes open (the subjects were presented with a black screen) and try to relax.

ReHo and ICA analysis

The first 10 volumes were discarded. The processing steps were the same for ReHo and ICA: 1) slice timing (correction to the first slice), 2) realignment (a detailed description is presented below in section Motion and data quality analysis), 3) normalization to the MNI template, 4) detrending, 5) for ReHo: temporal band–pass frequency filtering (0.01-0.08 Hz) and for ICA: spatial smoothing with a Gaussian kernel of 6 mm full width at half maximum (FWHM). The preprocessing was performed in DPARSF (http://rfmri.org/DPARSF).

Once pre-processing was completed, REST software (http://resting-fmri.sourceforge.net) was used to obtain ReHo maps (by calculating Kendall’s concordance coefficient value to measure the similarity of the ranked time series of a given voxel to its nearest 27 neighbor voxels in a voxel-wise way). After obtaining individual ReHo maps the data were smoothed with a Gaussian kernel (FWHM = 4 mm). First, one-sample t-tests were performed for the inter-subject analysis of each voxel of each image to identify voxels that had significantly higher ReHo. Then, a two-sample t-test was computed on the normalized individual ReHo maps to investigate ReHo differences between the two groups.

The Matlab toolbox, group ICA of fMRI Toolbox (GIFT; http://icatb.sourceforge.net), was used for ICA analysis. This included: principal component analysis (PCA) reduction, ICA separation and back-reconstruction to produce single-subject time courses and spatial maps from the raw data matrix [22].

The dataset was decomposed with different model orders: 29 (the optimal number of group’s components based on minimum description length criteria for source estimation); 20 (a default number of components in GIFT); 40 and 50 components. The rationale for using ICA with different numbers of independent components was that according to other studies, lower model orders provide a general insight into large-scale networks, whereas higher model orders can serve to detect more fine-grained networks corresponding to anatomical and functional segmentations [23, 24]. The data were separated by ICA using Infomax algorithm. For each model order, the ICA approach was repeated 30 times with ICASSO implemented in the GIFT toolbox.

Group ICA was run with all subjects included in one group, to ensure that the same components were identified in each subject. The default mode component was identified by spatially correlating all components with two default mode masks provided in GIFT: “DMN_ICA_REST.Nii” and “ref_default_mode.nii”. These masks contained regions that have been most commonly reported to comprise the DMN [25, 26]. We used different templates as they seem to be more weighted on different nodes of DMN: ref_default_mode.nii enables to sort the more posterior part of DMN, whereas DMN_ICA_REST sorts more anterior hubs. The component that most significantly (spatially) correlated with each template were selected as the default mode network. Then, the DMN components were transformed to z-scores using the GIFT software and exported to SPM8 (Wellcome Trust Centre for Neuroimaging, http://www.fil.ion.ucl.ac.uk/spm) where statistical analysis was performed. Statistical maps of the default mode network for patients and comparison subjects were created by entering the single-subject default mode component into a voxelwise one-sample t test and subsequently a two-sample t test, which was then thresholded at p < 0.05 with AlphaSim Correction. Regions were labelled using the Talairach Daemon Atlas [27].

Motion and data quality analysis

In order to eliminate the possible influence of motion on the data we followed the procedures described by van Dijk and collaborators [28]. First, separate matrices encompassing mean motion, maximum motion, and rotation of head motion were calculated based on translation and rotation parameters [28, 29]. Subjects whose mean motion parameters exceeded 0.5 mm were discarded from further analysis. Therefore, the final (C)APD study group consisted of 13 out of 23 recruited children with (C)APD and 15 out of 30 children were selected from among the examined healthy controls.

Secondly, the temporal signal-to-noise ratio (tSNR) parameter was estimated to evaluate the quality of the data and to eliminate any possible causes of data instability [28]. The mean tSNR for the (C)APD group was 289 (SD = 106), and the mean tSNR for healthy controls was 253.1 (SD = 113). A two-sample t-test showed no between-group differences [t (26) = 1.036, p = 0.310].

Clinical characteristics

The (C)APD group performed significantly poorer than controls in all tests that were applied (see: Table 1).

Resting-state brain activity patterns in (C)APD and in the control group

With the use of ReHo analysis in both groups we found a pattern of brain regions consistent with the DMN nodes provided in literature [9, 13, 26]. Nevertheless, two-sample t-test showed inter-group differences in regions encompassing: the precuneus (pCu)/posterior cingulate cortex (PCC) and the area of frontal pole corresponding to BA 10.

Table 2 provides a list of brain regions in DMN, along with coordinates of peak voxels and the number of activated voxels obtained in ReHo. Results obtained with the use of ReHo are presented in Figure 1. The final statistical map was set at a combined threshold of P < 0.005 and a minimum cluster size of 66 voxels, which resulted in a corrected threshold of P < 0.05 determined by AlphaSim.

Brain area	Cluster coordinates			Voxel peak T	Number of activated voxels
	x	y	z
pCu/PCC	2	-50	24	5.57	86
Superior Frontal Gyrus	-4	58	28	4.30	51

Abbreviations: pCu precuneus; PCC posterior cingulate cortex.

For ICA, the topographic distribution of DMN was similar in both groups and encompassed: middle frontal gyrus, posterior cingulate, precuneus, lateral parietal lobule, bilaterally. For the model with 20 components we found one component corresponding to DMN, regardless of the chosen template (correlation values for ref_default_mode.nii and DMN_ICA_REST. nii were 0.58 and 0.37, respectively). Nevertheless, we found that for model orders 29, 40, 50 the areas of significant co-activation start to decompose into the posterior and the anterior DMN. Interestingly, for the two DMN templates two different patterns of co-activation consistent with DMN (correlation values ranging from 0.43 to 0.65) were found. Consistently, depending on the model order and the components identified based on the correlation coefficient with the two templates inter-group results were inconsistent. Therefore, for the sake of clarity, we decided to present within-group results only (one-sample t-test). The final statistical map was set at the same threshold as in case of ReHo.Results obtained with the use of ICA are shown in Figure 2. In addition Figure 2 provides the correlation values for DMN identified based on the two described templates.

Rs-fMRI in (C)APD

In the present study we used a rs-fMRI paradigm to investigate brain activity in (C)APD (and not an active task). There were several reasons why we did so. First, previous fMRI studies in children with (C)APD showed that their concentration and tests performance decreases with increased duration of examination [5]. Then the possible group differences in brain activations may results from attentional lapses, only caused by the duration of the experiment rather than the modality-specific impairment. Resting state-fMRI, which is of very short duration, enables to overcome this limitation. Secondly, a paradigm that does not require children to cooperate actively, might diminish motion artifacts, resulting from reacting to stimulation.

Furthermore, a rs-fMRI paradigm allows us to examine a whole brain network rather than focal activations associated with a particular task. As (C)APD children in the present study showed deficits in all behavioral (C)APD screening tests, it would be advisable to apply as many as five different cognitive tasks in an MR scanner in each child. It would not be possible within a single fMRI session and it would be problematic to ask the subject to participate in an fMRI experiment several times in a short period of time.

Default mode network and attention

The ReHo results of the study are interesting, as the posterior cingulate cortex (PCC) and the precuneus not only form a central node in DMN in the human brain but they have also been suggested to play an important role in attention and goal-directed cognition [30, 31]. PCC shows high connectivity to the frontal network involved in cognitive control and attention demanding tasks [30]. Moreover, a specific pattern of PCC/pCu activity has been found in several rs-fMRI studies in ADHD [18].

Additionally, altered connectivity between the anterior (prefrontal cortex) and the posterior (posterior cingulate/precuneus) components of the DM network [18, 32] has been reported in subjects with ADHD. In our study we did not assess the functional connectivity directly, but we also found alternations in corresponding regions in ReHo analysis. As ReHo corresponds to connectivity at the local level, the revealed decreased ReHo values may reflect low regional metabolism [33] in PCC/pCu and the SFG (BA 10). This is in line with other studies linking attention disorders with disrupted consolidation of the DMN [18].

The atypical activity found in the region of PCC may suggest similar brain mechanisms underpinning (C)APD and ADHD. This can also indicate that rs-fMRI cannot differentiate between the modality-specific attention problems observed in (C)APD and general attention impairments.

A number of studies have suggested that dysregulation of DMN may be associated with lapses of attention and errors during cognitively engaging tasks [14]. According to default-mode interference hypothesis inefficacious transition from rest to task may account for impaired cognitive task performance [13]. It deserves further consideration, to investigate whether errors observed in (C)APD diagnostic batteries originate from general attention problems or a modality-specific perceptual dysfunction, as suggested by several authors [34]. Interpretative issues related to ICA with regard to the DMN analysis are discussed in the next section.

ICA or ReHo?

In the present study two different methods (ICA and ReHo) to analyze the rs-fMRI data were applied. Except for purely methodological studies, researchers usually choose only one approach to analyze the data. This may lead to some discrepancies between studies using different analysis techniques [35] or even only ICA but with different model settings [23].

One of the most popular techniques applied in resting-state fMRI studies has been ICA [36]. The main advantage of this method is lack of any initial assumptions about the spatial location of brain activations, which makes it suitable for exploratory fMRI analysis [36]. ICA is, however, not without its challenges, as it requires an arbitrary decision on the dimensionality reduction, as the number of components to estimate is not fixed and should be selected according to the data quality and the complexity of the results one wishes to obtain [e.g. resting state networks (RSN) may be separated into sub-networks reflecting their fine-grained nature] [36, 37]. Also, components representing networks of interest might be selected differently based on various templates which may lead to divergent results, as was presented in our study. Therefore, the reproducibility of the results obtained with ICA might be questioned, which has also been emphasized by other authors [23]. The issue of inconsistent ICA results has hardly been discussed and not resolved yet [23]. Thus, group ICA should be used with caution when drawing inferences about group differences.

Another approach widely used to the evaluation of rs-fMRI is regional homogeneity (ReHo), which investigates the temporal congruency of the regional BOLD signal in various brain regions [38, 39]. Both ReHo and ICA represent a connectivity analysis but they are performed with different mathematical approaches. While ReHo reveals correlations between slow fluctuations of the BOLD MR signal in various brain regions, ICA determines networks related to the decomposed temporal signal fluctuations. Importantly, ReHo is relatively insensitive to possible region-to region or trial-to-trial variability of the BOLD signal, which makes it complimentary to ICA [36]. The possible discrepancies in the results (which were demonstrated in our study) may be due to the fact that in ICA, the mean time course of a whole network is compared with the time course of individual voxels within that network [40], whereas ReHo seeks for correlations between the BOLD fluctuations of a small group of voxels with the adjacent ones. This might suggest that ReHo is more sensitive to local changes of functional connectivity and the ICA approach examines long-range connectivity [13].

Additionally, in ICA, due to differences in the model settings used in various studies results might be not reproducible. Also, possible inter-group differences might be concealed, as they may emerge in the components not included in further analysis.

Taking into account the aforementioned considerations, we recommend ReHo as the method providing more stable results in DMN network compared to the ICA.

The effect of motion

As a number of studies indicate that disruptions in BOLD signal resulting from head motion may create spurious patterns of correlations in rs-fMRI we made sure that in our study the two groups did not differ either in their motion parameters or in tSNR [29, 41]. Originally our study group was twice as big but after the motion parameter analysis we decided to eliminate participants whose motion might have confounded the results. We decided to exclude participants rather than apply “scrubbing” methods, as it was reported that this procedure does not always remove all motion-related signal and that these are most efficient when periods of motion are within a single TR [41]. In children, however, the periods of motion span across multiple TRs. Consequently, replacing time points in which movements occurred with adjacent ones may contribute to losing the signal of interest.

Limitations of the study

The presented findings are promising but several methodological limitations should be indicated (which are not confined to the present study). First, although there is a growing interest in rs-fMRI studies, interpretative issues remain to be resolved before the alternations in the intrinsic brain activity can serve as an indicator of cognitive dysfunction [36].

We suggest that group differences, showed in the ReHo analysis, reflect limited regional neuronal synchronization in (C)APD, which might mean that detected neurons do not behave in a coherent fashion during rest. However, considering the fact that physiology underlying spontaneous BOLD response might be different to that when performing a task, it is under debate whether alternations found in this study directly underlie behavioral problems manifested by individuals with (C)APD.

Finally, although we performed ICA with different model orders, we only selected components which represented DMN and compared that particular components with the results obtained with ReHo. The rationale was that from all the reported resting-state brain networks, only DMN is considered to be involved in the introspective modes of cognition, encompassing the shift between the internally and the externally directed attention [6]. Nevertheless, it is possible that selecting components reflecting different resting state networks would provide additional valuable results.

Since the ReHo approach revealed significant differences in DMN in children with (C)APD, relative to controls, we conclude that this matter deserves further consideration. Applying different neuropsychological and neuroimaging methods (including also task-based fMRI paradigms) might be helpful in differentiating developmental CAPD from other developmental cognitive disorders and further elucidate the neural basis of (C)APD.