Introduction
Besides impairments in social interaction and communication as well as restricted and stereotyped interests and behaviours (APA,
2013), individuals with autism spectrum disorder (ASD) also show a range of unusual visual processes (e.g., Dakin & Frith,
2006). In particular, superior performance in some visuo-spatial tasks, such as Block Design and Figure-Disembedding (for a meta-analysis see Muth et al.,
2014) have led to the assumption of a more detail-focussed processing style (Fitch et al.,
2015; Frith & Happe,
2006), with reduced or spared global processing capacities depending on task requirements and instructions (Koldewyn et al.,
2013; Mottron et al.,
1999; O’Riordan & Plaisted,
2001; O’Riordan et al.,
2001; Plaisted et al.,
1999).
In general, visually presented information is organised according to classical Gestalt laws (e.g., Koffka,
1935), such as proximity (i.e. objects are grouped according to spatial closeness) or similarity (i.e. objects are grouped according to shared features). In an early classification of visual processing, a division into so-called type P and type N relationships was suggested (Pomerantz,
1983): While type P relationships are characterised by the placing of elements (e.g., closeness), type N relationships are characterised by the nature of visual elements (e.g., similar appearance). Individuals with ASD show particular weakness in the perception of type N relationships (e.g., Brosnan et al.,
2004; Falter et al.,
2010).
Nevertheless, whether visual grouping processes can be considered typical or atypical in ASD is still not settled. Studies employing tasks that explicitly measure grouping processes on the basis of participants’ introspection, showed impairments in perceptual grouping in ASD, particularly with respect to grouping by similarity (type N; e.g., Boelte et al.,
2007; Brosnan et al.,
2004). Similarly, a study measuring grouping by five different Gestalt principles showed weaker perception of grouping by similarity (type N) with respect to shape, orientation and luminance, and spared grouping by proximity (type P) and alignment (Farran & Brosnan,
2011).
The introduction of object-based attention tasks (Feldman,
2007) allowing the implicit measurement of how grouping principles influence perceptual judgments, showed a differentiated picture of impaired grouping by colour similarity (type N) and intact grouping by proximity (type P) in ASD (Falter et al.,
2010). Perceptual grouping in ASD also seems to depend strongly on general task requirements and instructions. For instance, symmetry perception in individuals with ASD has been found to be both superior (Perreault et al.,
2011) and inferior (Falter & Bailey,
2011), depending on task design and underlying constructs, with the authors of the two studies drawing contradicting conclusions (for a discussion see Falter,
2012). Thus, task design and instructions might play a role in shaping participants’ performance in grouping tasks. Similarly, sampling and inclusion criteria that vary between studies might likewise cause group differences to be found in some studies and not in others. Recent studies measuring grouping processes
implicitly in ASD showed typical influence of four different, both type P and N, grouping principles (similarity, proximity, closure, and good continuation) on stimulus distance estimations in ASD (Avraam et al.,
2019) and typical shape formation depending on grouping cues, such as closure, proximity and collinearity (Hadad et al.,
2019). Likewise, grouping of distractors was found to be typical in a visual search paradigm in individuals with ASD (Keehn & Joseph,
2016) and visual speed discrimination thresholds were equally strongly influenced by grouping of stimuli in children with and without ASD (Manning & Pellicano,
2015). These series of implicit measures therefore suggest typical grouping perception in ASD and a lack of impairments. Yet, such a conclusion could be premature: Even implicit measures of perceptual grouping processes show inconsistent results of reduced (Evers et al.,
2014) or typical grouping interference on multiple object tracking abilities in children with ASD (Van der Hallen et al.,
2018), again depending on exact task requirements and study design (e.g., ratio of grouped versus ungrouped stimuli; for a discussion see Van der Hallen et al.,
2018). Therefore, an answer to the question whether perceptual grouping is atypical or typical in ASD remains unsettled. Thus, in this study we turned our attention to the neural processing mechanisms of grouping principles in ASD. The rationale behind this approach is that, irrespective of typical or atypical performance levels in grouping tasks, we do not know whether the underlying neural mechanisms recruited by individuals with ASD resemble those used by typically-developing (TD) individuals.
Studies on neural processing of visual grouping in TD individuals found that grouping happens early in the visual processing stream and that different Gestalt laws appear to correspond to different neural mechanisms (Han et al.,
2002). In an event-related potential (ERP) study, grouping by the law of proximity (type P) has been shown to be associated with short-latency positivity over the medial occipital cortex followed by a right occipitoparietal negativity, whereas grouping by shape similarity (type N) elicited a long-latency occipitotemporal negativity (Han et al.,
2001). Similarly, grouping by colour similarity (type N) was associated with long-latency occipito-temporal modulations (Han et al.,
2002). These ERP-findings suggest that distinct neural substrates are associated with visual organisational processes based on different Gestalt laws in TD participants.
Although there are no neuroimaging studies investigating mechanisms of Gestalt grouping in ASD, there are suggestions of different processing strategies for Gestalt grouping. Farran and Brosnan investigated grouping by shape similarity (type N) with varying difficulty levels across stimulus displays. They found that when participants with ASD employed different processing strategies compared to TD controls, they yielded a similar accuracy. However, when they relied on the same strategies as TD controls, their performance was impaired (Farran & Brosnan,
2011).
The aim of the current study was to compare neural correlates of Gestalt grouping in individuals with and without ASD. Measuring neuromagnetic activity, participants were presented with two of the most commonly tested Gestalt grouping principles, proximity (type P) and similarity (type N), in the well-established paradigm used by Han and colleagues (Han et al.,
2001,
2002). Across groups, we expected grouping by proximity to be processed at an earlier and grouping by similarity at a later stage (Quinlan & Wilton,
1998). Furthermore, on the basis of previous behavioural results, we expected deviant patterns of processing Gestalt grouping principles in individuals with ASD compared to TD individuals.
Discussion
The aim of the current study was to explore the neural mechanisms underlying classic Gestalt grouping processing in visual displays in individuals with ASD using neuromagnetic activity. The main findings are summarised as follows: (i) stronger neuromagnetic activity during similarity grouping (SG) compared to proximity grouping (PG) in both participant groups, possibly suggesting lower processing demands during PG (see Fig.
2), which might be more readily discernible and differentiable compared to SG in both groups; (ii) significantly reduced and slower activity in the ASD group compared to the TD group within the first 200 ms after stimulus onset, irrespective of the grouping condition; (iii) a complex set of interactions between group and condition was found in different brain regions, with higher activities at later latencies (Fig.
5) in occipital and superior parietal areas in ASD and a more distributed activity in the right hemisphere in the TD group (see Fig.
3).
Against the backdrop of persistent inconsistencies in the findings of behavioural performance of individuals with ASD in Gestalt grouping tasks, as outlined in the introduction, the use of different neural strategies for Gestalt processing has been discussed (Farran & Brosnan,
2011). The rationale for using a direct measure in this MEG study was to explore the neural correlates of grouping in a categorical design. For this purpose, we chose a very simple Gestalt grouping design established before (Han et al.,
2001,
2002), which yielded in our study above 94% correct answers in both participant groups, allowing us to investigate any differences in neural mechanisms underlying Gestalt grouping processes during low task demands.
In both groups neuromagnetic activations during SG were stronger than during PG. Yet, despite this similarity between groups, we also found significantly higher and earlier amplitudes in the TD group during both types of grouping (SG and PG) as compared to the ASD group. Hence, despite the speculatively higher processing demand of SG over PG in both participant groups, the TD group seems to be better prepared for this task. The delayed processing of grouping information in ASD could be in accordance with previous reports of a sluggish cognitive tempo in ASD (e.g., Brewe et al.,
2020).
Comparing localisation of evoked activity, the contrast between SG and PG was found to be significantly different between groups in seven different ROIs. Two superior parietal areas (PreCC, SMG), three visual areas (FG, LG, PeriCC), one temporal area (STG), and one frontal area (PT) showed stronger activity at later latencies in ASD compared to TD (Fig.
5). This difference could be interpreted as high demand of SG as opposed to PG requiring increased neural activation in brain regions responsible for the processing of visuo-spatial material in persons with ASD. This specific information about grouping processing could explain why the performance in response to PG is faster and easier compared to SG (e.g., Falter et al.,
2010; Farran & Brosnan,
2011). Similarly, increased reliance on neural processing of visuo-spatial information in posterior regions in individuals with ASD has been reported before (e.g., Falter,
2012; Kumar,
2013).
A second neuroanatomical observation was the more distributed and stronger activation in the right hemisphere of TD persons, when compared to the ASD group. Thus, it seems that the TD group showed a more pronounced hemispheric specialisation (e.g., Allen,
1983), putatively enabling a more efficient resolution of perceptual demands. Our results corroborate several previous reports of reduced or atypical hemispheric asymmetry in ASD. For instance, significantly reduced white matter microstructure asymmetry has been found in ASD (Carper et al.,
2016) and atypical asymmetry patterns of activity were found in a previous MEG study on sentence reading (Ahtam et al.,
2020). Many previous findings relate atypical hemispheric asymmetries to language impairment in ASD (Lindell & Hudry,
2013) but several functional networks have recently been found to show atypical hemispheric asymmetry beyond language functionality (Cardinale et al.,
2013). Alternatively, less hemispheric asymmetry might be due to unrelated or noisy activity in one hemisphere in ASD.
Our interpretation of stronger specialisation in the allocation of neuronal processing in the TD group according to task requirements (Fig.
3), compared to a less specialised response in the ASD group, might be a general characteristic of the autistic neuro-cognitive profile in that individuals with ASD might adjust processing less specifically to conditions. Indeed, we previously also found less specific neural activity to task-specific requirements for auditory duration versus pitch perception in ASD compared to a matched TD group (Lambrechts et al.,
2018).
Particularly interesting is the significantly increased precuneus activity at 220–300 ms found for SG in contrast to PG in the ASD group only. Increased precuneus activation has previously been found in individuals with ASD during sustained attention: in the ASD group activation was found to progressively increase in the precuneus with increasing sustained attention load, in contrast to progressively decreasing activity found in the control group, which the authors interpreted as difficulty with default mode network suppression in ASD (Christakou et al.,
2013). Furthermore, increased precuneus activation was positively correlated with social symptom severity as measured with the ADOS (Christakou et al.,
2013). This finding again supports the proposal of a higher cognitive load during SG as opposed to PG that is specific to persons with ASD.
Although not directly comparable, our findings are generally speaking in line with studies of Gestalt grouping using fMRI. A recent fMRI study on spontaneous Gestalt processing showed particular activity in the superior parietal lobe and the anterior intraparietal sulcus associated with grouped illusory Gestalt perception (Zaretskaya et al., 2013). Likewise, the regions of interest found in the current study are in keeping with the report of intact versus disturbed global Gestalt perception in hierarchical stimuli (i.e. global shapes made of local elements) leading to activity in precuneus, temporo-parietal junction, and anterior cingulate cortex (Huberle & Karnath,
2012). In addition, Han et al. (
2005) showed proximity grouping associated with calcarine cortex, inferior parietal cortex (LH and RH) and right superior temporal cortex. Similarity grouping was associated with right middle occipital cortex, left middle temporal cortex, yet similarity was based on shape similarity, not colour similarity. Thus, an overlap of regions found responsible for Gestalt grouping in fMRI studies and the sources located in the current MEG study can be asserted. Employing MEG we could additionally carve out slower activity in the ASD group compared to TD controls.
Behaviourally, we observed a high accuracy with more than 94% correct answers, irrespective of diagnostic group or grouping process studied. Despite of the statistically significant difference between diagnostic groups, accuracy scores clearly showed a ceiling effect in an obviously low cost task; in addition, the percentage of incorrect trials, which were below 5% of all trials (ASD: 4.9%; TD: 2.1%), was very low so that we considered the direct comparison of the neural correlates of correct trials in the different conditions across both groups as justified. The observation of very high accuracy scores in both conditions and among both diagnostic groups of at least 94%, might be due to our design that directly assessed grouping in a categorical design, rather than using a paradigm indirectly measuring grouping strength in a parametric design. Discrepancies in findings of intact or impaired performance in tasks employing Gestalt principles in ASD might be due to direct versus indirect measurement (Farran & Brosnan,
2011). Future studies should focus on “critical” tasks that have been shown in the past to lead to group-difference results–in comparison to tasks that have been shown to show equivalent performance—in the same sample.
A limitation of our study was that only a part of our participants agreed to have their MRI scans taken (n = 13). Instead of excluding all participants for whom MRI scans were not available and to avoid the risk of having less representative sample sizes, we decided to include theses data into the analysis. For these participants, a template scan (as a substitute for an individual MRI) was used as provided by FreeSurfer package (Fischl et al.,
2001). The average localisation error (distances between the centre of masses of correctly coregistered areas using the individual MRI and a registration based on a template) due to missing MRIs was small (estimated at 1.15 cm). Our ROI definition was based on relatively large regions as defined by the Desikan-Killiany atlas. The mean surface area of our ROIs defined in Table
2 was estimated at about 30.70 cm
2, which would result in a diameter of about 6.26 cm for circular areas. In other words, a displacement of the centre of mass of significantly active vertices in the range of the coregistration error is very likely to be located in the same anatomical area.
A limiting factor of a priori hypothesis setting was the state of art of neuroscientific (in particular MEG) data on Gestalt processing at the time of study design. The current study should be considered exploratory. Given the limited sample size included and the unconstrained analysis, the interpretation of results should be treated with caution and submitted to scientific scrutiny in future research studies Thus, there remain several potential explanations for the group differences in MEG findings. We cannot entirely rule out that (i) these could be based on chance, (ii) the group differences might indicate different processing strategies related to slightly decreased performance in the ASD group, or (iii) the group differences might reflect compensatory processing in the ASD group. Finally, concerning any group differences in recordings of brain activity one cannot entirely rule out that these may be due to uncontrollable confounds that are not accessible by any behavioural assessment.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.