Coherence is an important tool for the study of complex cortical network dynamics and temporal fluctuations in the coupling between neural signals. Previous studies have shown that measures of coherence reflect patterns of cortical connectivity in the brain and that decreased values of coherence are associated with reduced connectivity between distant neural networks [41, 57].
The results of the present study show a widespread and consistent reduction in interhemispheric coherence in the ASC group compared to the control group, during both visual tasks. These group differences are spread across the entire time-frequency spectrum, though they are more pronounced at frequencies lower than about 13 Hz and generally around 150 ms post-stimulus onset (Figures 1 and 2 – uncorrected for multiple comparisons). We hypothesize that these results are indicative of an overall impairment in functional interhemispheric connectivity during visual processing in people with ASC. This hypothesis is supported by previous reports of decreased structural and functional interhemispheric connectivity in ASC [14, 15, 22–24]. In addition, the tasks employed in this study involved object categorization, with participants needing to decide whether each presented image was of a chair or a human face. There is evidence that object categorization may be impaired in people with ASC [58–60]. It is also interesting to note that superordinate distinctions in object categorization can occur relatively soon after stimulus presentation. Van Rullen and Thorpe  reported electrophysiological differences associated with superordinate categorical differences (for example, animals vs. vehicles), peaking between 200 and 250 ms post-stimulus onset in typical controls. Similarly, Curran et al.  have reported ERP data indicating that feature analysis (supporting the process of finding similarities that link object exemplars into categories) precedes later processing stages associated with recognition of specific objects. Hence, it can be hypothesized that the relatively reduced coherence manifested in the current study by the participants with ASC from around 150 ms is related to atypical performance in categorization.
Consistent with this hypothesis, studies of the time-frequency responses of typically developing members of the general population to visual stimuli, including houses and faces, have shown that these responses could be explained by amplitude increases maximal in the 5 to 15 Hz frequency band, between 100 and 200 ms post-stimulus onset [63, 64]. These reported frequencies are similar to the ones at which the participants with ASC display decreased interhemispheric coherence in the current study. The paradigms employed by Rousselet et al.  and Tang et al.  differ significantly from the one used in the current study, and neither explored coherence of EEG activity between different electrode sites. Nevertheless, it is interesting to consider their results in light of the present study, where most group differences in interhemispheric coherence are found in a frequency band below around 13 Hz – the differences in coherence observed in the current study may relate to differences in brain activity associated with structural encoding of the observed images as part of their initial categorization as either faces or chairs. While no correlations were found between AQ or IQ and coherence for either the ASC or the control group, it is important to note that the power to detect a correlation is low given the small sample size in the current study. It is interesting to note, however, that in the ASC group the correlations between coherence and AQ scores were negative, for both tasks and all electrode pairs.
Observation of Figures 1 and 2 also seems to indicate that during the chair task, but not the face task, there is relatively decreased interhemispheric coherence in the ASC group earlier in the response window (less than 200 ms post-stimulus onset), at higher frequencies (>13 Hz), in more posterior regions of the cortex. While this observation is interesting, these group differences do not survive correction for multiple comparisons and without additional group-by-task interactions analyses further discussions on the interpretation of these findings would be speculative. Informed by the methods of previous WTC analysis of EEG data  and taking into account the small sample size of the current data set, in the current study it was decided to run the statistical analysis in a non-parametric context. In this case, group-by-task interaction analyses are not trivial to perform, and algorithms for non-parametric interaction analyses are still under development [65, 66].
In the current study, the lack of significant differences in EEG power spectra between groups or tasks also establishes a distinction between coherence measures and power spectrum analysis; changes in coherence values are not a reflection of changes in EEG power spectra in any frequency band. This is in accordance with some previous studies reporting the absence of abnormal patterns in EEG power spectra in individuals with ASC [67, 68].
While there are clear differences in interhemispheric coherence between the ASC and the control groups in this study, the small sample sizes limit the statistical power of the comparisons. Additionally, it is important to note that due to the size of the data matrices being analyzed in this study (36 frequency points by 401 time points giving a total of 14,436 data points) standard methods for correction for multiple comparisons, such as Bonferroni, were not suitable. However, despite the fact that previous studies using WTC for the analysis of EEG data do not correct for multiple comparisons , care must be taken when interpreting uncorrected statistical results. Although significant group differences are seen in well-defined time-frequency clusters, increasing the likelihood of these differences being meaningful , correction for multiple comparisons was still performed.
The multiple comparison problem for such a large data set (a large volume of data per participant, despite a low sample size) must be carefully considered . As mentioned above, conservative methods, such as Bonferroni correction, are less suitable as they lead to a high number of Type II (false-negative) errors, potentially losing true differences. However, the absence of any type of correction leads to the presence of Type I (false-positive) errors. Less conservative methods, such as False Discovery Rate correction (FDR; ), are commonly used in the statistical analysis of functional neuroimaging data, usually comprising hundreds of thousands of data points, and so FDR was considered suitable for use in the current study. However, it is important to note that as highlighted in a review on Type I and Type II error concerns in neuroimaging research , even FDR correction may be overly conservative when dealing with small effects. In a review by Lieberman and Cunningham , it is suggested that systematic meta-analyses should be used as an alternative approach in dealing with type I and type II errors, given that these random errors should not replicate across multiple studies, unlike true significant effects.
In the present study the only group difference that survives FDR correction is an area of decreased coherence for the ASC relative to the control group, for electrode pair T7-T8 in the faces task. This area is located at around 300 ms post-stimulus onset, on a frequency band between 7 and 10 Hz. It is interesting to note that temporal sites have previously been associated with visual processing of faces, albeit at earlier post-stimulus onset times [64, 72]. The measures used in these investigations differ from the one used in the current study: the former uses measures of localized brain activity indexed by ERP components, while the latter uses a measure of interhemispheric coherence. Nevertheless, these previous investigations provide evidence that temporal regions are functionally involved in facial processing and it can be hypothesized that the group differences identified in the present study reflect atypical face processing in people with ASC, indexed by decreased interhemispheric coherence between temporal sites in this group.
As pointed out by Srinivasan et al. , moderate to large EEG coherence can also arise from volume conduction effects. However, in the current study, the finding of specific time-frequency regions surviving FDR correction suggests that group differences in interhemispheric coherence are not simply the result of differences in magnitude of volume conduction between the two groups, but represent a difference in genuine source coherence. Similarly, previous studies have shown that reference electrodes may influence coherence calculations of EEG signals . Of particular interest to the current study are the findings of Essl et al. , showing that reference signals originating from a nose reference electrode may artificially increase coherence values. However, in the current study the reference electrode was the same for all individuals, and it is reasonable to assume that group and task differences in coherence would not be affected by the choice of reference electrode. It is also important to note that although group differences surviving FDR correction are quite limited, considering the small population size of the current study and taking into account the review by Lieberman and Cunningham  mentioned above, it is possible that in this case the FDR correction may have been overly conservative, and that other equally important, but small, effects are being missed.
Decreased interhemispheric coherence in ASC has been reported in previous studies [40, 41]. Isler and colleagues  found decreased interhemispheric synchrony in children with ASC, when compared to typically developing children, in occipital lobes, in and below the theta frequency band (<8 Hz), during a visual stimulation task. In an investigation of resting state EEG coherence in children with ASC, Coben et al.  found evidence of decreased interhemispheric delta (0 to 4 Hz) and theta (4 to 8 Hz) coherences in frontal regions, as well as decreased delta, theta and alpha (<13 Hz) interhemispheric coherences in temporal areas of the cortex. Coben et al.  also report a decrease in delta, theta and beta (<8 Hz and 13 to 30 Hz) interhemispheric coherences in parietal regions of the brain. Although the paradigms and population samples of the current study and those of Isler et al.  and Coben et al.  are not directly comparable (in that the current study was an investigation of task related coherence in adults and the others examined resting state and visual flash evoked coherence in children), all studies investigated a variety of brain regions and frequency bands, and the results of the current study can be considered as supported by and complementary to those of Isler et al.  and Coben et al. . Additionally, the investigation of functional brain coherence using other modalities confirms that decreased interhemispheric connectivity in people with ASC is a consistent finding [75, 76]. In a resting state MRI study that recruited individuals with and without ASC from late childhood to early adulthood, Anderson et al.  found evidence of impaired interhemispheric connectivity in ASC in sensorimotor cortex, anterior insula, fusiform gyrus, superior temporal gyrus and superior parietal lobule, while Dinstein et al.  investigated interhemispheric coherence in toddlers with ASC using MRI, and reported decreased interhemispheric connectivity in putative language areas, such as superior temporal gyrus.
The present study investigated task-related interhemispheric coherence during visual perception of chairs or faces in cortical regions, including frontal, temporal and parietal areas. These disparate regions are likely to have been involved in a variety of different components of the task, from visual processing to visual categorization learning [77–79]. The relatively extensive analysis of coherence performed in the current study, over a time-frequency range from 5 to 40 Hz and 1 to 400 ms post-stimulus onset, supports the conclusion that interhemispheric connectivity in ASC is impaired not only in posterior regions but also in frontal and temporal regions of the cortex (as reflected by group differences not corrected for multiple comparisons), in similarity to the results of the resting state studies of Coben et al. , Dinstein et al.  and Anderson et al. . In addition, the use of the WTC approach enabled evidence to be gathered suggesting that it was in lower frequency bands that group differences in EEG responses to the tasks were concentrated, as shown by the uncorrected group differences’ results. Previous studies have shown evidence relating low frequency theta and alpha synchronization with top-down working memory processes, subserving functional integration over multiple neural networks [80, 81]. The visual matching task included in the current study can be considered to involve working memory processes [81, 82], and we hypothesize that the decrease in low frequency coherence in the ASC group reflects atypical neural connectivity that results in an impairment of integration of information across neural networks. Additionally, previous studies have suggested the existence of a relation between the size and distance of a neural interaction and the frequency of the neural synchronization. In particular, it has been reported that lower frequency oscillations seem to be associated with larger neuronal assemblies and long range connectivity [80, 83–86]. The results of the present study are complementary to these reports, and show further evidence supporting theories of impaired long range connectivity in ASC [9, 10, 14, 15]. Our results suggest that interhemispheric connectivity in ASC is widely atypical, and it is hypothesized that this may have greater implications for tasks that require integration of information over neural networks spread across both cortical hemispheres.
As can be seen in Figures 4 and 5, some differences were found in within-group interhemispheric coherence between the chairs and the faces task, for both groups (uncorrected for multiple comparisons). Differences in coherence between tasks were not constrained to a particular region of the time-frequency map, occurring at both early (50 ms) and late (300 ms) post-stimulus onset times, from lower (7 Hz) to higher (23 Hz) frequencies. These differences were significant at a larger number of electrode pairs for the control group than for the ASC group, possibly reflecting an impairment in task differentiation in people with ASC relative to typically developing controls. This is consistent with previous investigations showing impairments in object categorization and face processing in people with ASC [43, 44, 59, 60]. It is also consistent with the results from an ERP investigation using the same paradigm as the current study . It is important to note that although within-group differences in coherence between tasks were found, these were not as significant as the group differences represented in Figures 1 and 2, and did not survive correction for multiple comparisons. This may be related to the behavioral results of this study, showing an absence of significant group differences in task performance in terms of speed and accuracy of image recognition. However, behavioral results also show that across both groups, the face task was performed a little less accurately than the chair task. This task effect in accuracy was driven by relatively lower accuracy for the ASC group in the face task, reflected in a group-by-task interaction that approached significance (F1, 28 = 3.661, P = 0.066). Despite this, both the ASC and the control groups performed the tasks with high degrees of accuracy and close to ceiling level (Table 2). The trend observed in the group-by-task interaction is probably the result of the majority of control subjects performing at ceiling level, and the participants in the ASC group making a larger, yet still small, number of mistakes. Clinically, these differences are not considered to be relevant, as the ASC group still presents accuracy scores of around 90% for the face task. The paradigm used in the current study may thus not have been sufficiently demanding to detect possible group differences in task performance or task differences in coherence. Further research is recommended to examine potential correlations between specific cognitive or behavioral functions and atypical patterns of interhemispheric coherence in people with ASC. Additionally, future investigations using the WTC algorithm should seek to improve statistical power of their analyses by using larger population sizes and correcting for multiple comparisons using FDR or similar method.