Posted comment on ´Expectation cues and false percepts generate stimulus-specific activity in distinct layers of the early visual cortex` written by J. Haarsma, N.Deveci, N.Corbin, M.F.Callaghan and P.Kok and published in Journal of Neuroscience 2023 vol 43(47) p. 7946 doi 10.1523/JNEUROSCI.0998-23.2023
SUMMARY
Haarsma and colleagues discuss in their article their findings that deep layers of the V2 visual cortex reflect perceptual expectations in a visual stimulus task but high confidence false percepts (ie. absence of grating pattern) was reflected in the stimulus-like feedforward activity of the middle input layers only and were unaffected by perceptual expectation cues. These findings are contrary to other studies on feedforward/feedback prediction theories of false percepts and may put a stronger emphasis on feedforward signals than generally thought in the case of hallucination models which have been largely based on top-down expectations being the driving force.
The experimental set-up involved 25 participants (average approx. 25 years old) performing a laboratory based perceptual orientation discrimination task and 100 participants taking part in an online study. In the case of the online study, questionnaire data was collected for the Peter et al Delusions Inventory test (PID) and the Cardiff Anomalous Perceptions Scale (CAPS). Subscales were added for both so that total scores could be calculated and then correlated with the online study behavioural measures. For the laboratory-based perceptual orientation discrimination task, layer-specific fMRI was carried out in order to assess the feedforward and feedback influences on perception. Auditory cues were given to implicitly signal the most likely upcoming orientation (termed the expectation). Activity in the chosen cortical layers represented the false perceptions of the oriented grating visual stimulus with the premise that orientation-specific activity in the agranular layers of the visual cortex indicate feedback activity driving the false percepts and activity in the middle layers of the visual cortex indicate feedforward activity.
The visual stimuli used was a computer-generated grey scale luminance-defined sinusoidal Gabor grating which was presented differently according to the investigation taking place. For example: fMRI study behavioural session – PC presented; fMRI scanning session – projected onto rear screen and viewed with a mirror. Trials were either with (grating-present) or without presentation of the grating stimulus (grating-absent). On the grating-present trials (50% of the total), an auditory cue was presented and 750ms later a grating which was displayed in an annulus surrounding a fixation bull`s eye. The stimuli were combined in one of four noise patches (created and selected with a spatial frequency matching the gratings), which led to finally to a 4% contrast grating embedded in 20% contrast noise during the fMRI session. For the grating-absent trials only one of the four noise patches was presented.
The experimental procedure relating to the laminar fMRI study began on Day 1 with the participants taking part in a behavioural practice session. This consisted of an instruction phase with 7 blocks of 16 trials with the task made progressively more difficult and auditory cues predicting orientation of the grating stimulus and presentation of the grating stimulus, ie. 45 degrees, 135 degrees or no grating (ie. grating-absent). The participants then carried out 4 runs of 2 blocks, each of 64 trials. The expectation cues (the auditory sound) were 100% valid for the first 2 runs to ensure that the association had been learnt. The other 2 runs had the expectation cues at 75% valid (unexpected grating orientation for 25% of trials) to rule out response bias. The grating contrast was decreased over the four runs (7.5, 6, 5, 4%) with no grating-absent trials and the noise patch contrast remaining constant at 20%.
On Day 2, basically the same tasks of Day 1 were repeated, but the participants were in the fMRI scanner. Half of the trials were gratings-absent meaning that there were no gratings (cue definition – invalid), had only noise patches and the cues always predicted the orientation of the grating with 100% validity. The other half of the trials were grating-present. These trials had an auditory expectation cue followed 750ms later by a grating stimulus embedded in the noise. The grating contrast was fixed at 4% and presented for 33ms. It then disappeared and an orientation response prompt appeared (left/right-pointing arrow on either side of fixation dot). The participants had to select the arrow corresponding to their answer (ie. left arrow – anticlockwise or 135 degrees; right arrow – clockwise or 45degrees) which was then followed by a judgment on their confidence in their decision (ie. letters CONF? appeared on the screen and a decision had to be made between 1 and 4 where 1 meant grating not seen to 4 where there was certainty on seeing the grating). This was then followed by a functional localiser task in blocks of 4 (4TRs – flickering gratings at 2HZ, 100% contrast, block of approx. 14.3s; gratings fixed orientations of 45 or 135degrees) where the two orientations were presented in pseudorandom order followed by a fixation bull`s eye on a blank screen (approx. 14.3s). The participants were asked to respond when the black fixation dot briefly dimmed to ensure central fixation. The online study followed a similar procedure with only minor differences.
The MRI scanning carried out in the experiments followed established procedures. The fMRI data obtained was preprocessed (two volumes of each run discarded; cropped to cover only the occipital lobe) and the methods for segmenting and coregistering the cortical surfaces followed previously published studies. The regions of interest (ROIs) used by Haarsma and team were the primary visual cortex (V1) and secondary visual cortex (V2) and to discount overlap the ROIs were constrained to the voxels responsive to the localiser gratings. A selection of 500 voxels which most strongly preferred 45 degrees over 135 degrees and vice versa were finally selected. The time course of each voxel was normalised (z-scored) and multiplied by the absolute t value of the orientation contrast to weight the data by the most robust orientation preference.
With regards to the definition of the cortical layers, the grey matter (GM) was divided into 3 equally volumed layers (deep, middle, superficial) by established methods. Based on the layers I to III, layer IV and layers V to VI, four signed distance functions (SDFs) were calculated and used for the calculation of the distribution of the volume of each selected voxel over the five compartments (WM, CSF and 3 GM layers). Spatial GLM was used to decompose the layers in order to counteract the ´partial volume problem` so that following mathematical analysis BOLD time courses for both the 45degree and 135degree preferring voxels for the superficial middle and deep layers of the early visual cortex could be obtained. The authors also used a temporal GLM to estimate effects of interest (ie. expecting and falsely perceiving specific orientations) in each of the three GM layers. The methods used followed established criteria and BOLD responses were subjected to two-way repeated measures ANOVA.
Behavioural analyses were also carried out. In the case of the online study, repeated-measures ANOVAs were used to compare the accuracy and confidence scores across the different contrast levels. Confidence levels relating to accuracy at grating orientation identification were also compared as well as whether the effects of the cues were mediated by the participants having awareness of their meaning. A logistic regression model was used to investigate which factors predicted orientation response on separate grating-present trials (predictors – current stimulus orientation, current stimulus contrast, orientation predicted by the cue, orientation response on the previous trial, interaction between present stimulus contrast and orientation) and grating-absent trials (predictors – previous orientation response and orientation). Correlations between abnormal perceptual experiences measured by the CAPS questionnaire and the cue effects were also investigated. In the case of the fMRI study, several behavioural analyses were carried out. For example: modulation of accuracy by confidence; the proportion of high confidence false percepts on grating-absent trials; proportion of cue congruent responses; and participant`s orientation responses.
The first set of results reported related to the experiencing of false percepts independent of perceptual expectation cues. Haarsma and colleagues found that the participants accurately identified the grating orientation on grating-present trials more often than chance (mean accuracy – 0.83). The participants were more accurate when they were confident they had seen a grating than when not (high – approx. 0.9; low – approx. 0.74) which showed that they were able to perform the task and were able to accurately judge their level of confidence. The participants also reported the perceived orientation more quickly on grating-present than grating-absent and were more confident on the grating-present trials (present – approx. 2.5; absent – approx. 2.2). Underestimation of the frequency of grating-absent trials was also reported (average reported 0.14; true frequency 0.5). Participants also reported perceiving a grating with confidence values of 3,4 or higher (high) on 36% of the grating-absent trials plus the perceptual expectation cues did not significantly bias which orientation participants perceived on grating-absent trials (0.53 to chance level of 0.5). The small numerical trend towards false percepts being congruent with the expectation cues arose from a few participants who had become aware of the meaning of the cues indicating a concomitant response bias (values of approx. 0.7 to 0.76). High-confidence false percepts were found to be not more affected by the cues than the low-confidence percepts.
Haarsma and colleagues also found that certain predictors dictated responses. For example, orientation responses on grating-present trials were predominantly driven by the presented stimulus. They were also driven by which orientation was perceived on the previous trial (previous response – 0.7 significant). This also occurred for grating-absent trials whereas cues did not (cues – approx. 0.3; outlier group – 1.5 to 1.7). Again, the trend was driven by a few participants who had become aware of the meaning of the cues (the outliers). However, in general Haarsma and team reported that the participants reported false percepts, but these were not significantly driven by the perceptual expectation cues in the present experiment. This indicated to them that possibly the false percepts had arisen from spontaneous fluctuations instead.
The next set of results reported by Haarsma and colleagues related to the orientation-specific activity seen in the V1 and V2 layers reflecting the false percepts and expectation cues experienced by the participants. The first observation was that the laminar profiles of the V2 for the false percepts with both high and low confidence (the guesses) and perceptual expectations were significantly different from each other. A significant difference was also observed in the three laminar profiles when high- and low-confidence false percepts were directly compared. The observation was driven by differences in the layers, ie. increased orientation-specific activity for high-confidence compared to low-confidence false percepts in the middle layers (approx. 2.2 for high-confidence) but not in the superficial or deep (approx. 0 for both). This indicated that the participants reported perceiving a specific-orientation grating with a high degree of confidence in the absence of the stimulus and this report was accompanied by observed activity in the V2 middle layers. Activity associated with the high-confidence false percept was significantly higher in the middle layer than in the superficial and deep layers (approx. 2.2 versus 0). When the percept was reported with low confidence (a guess) orientation specific activity was not observed (approx. 0). Therefore, high confidence false percepts and perceptual expectations were found to have different laminar profiles with the difference driven primarily by the middle layer which was activated by high-confidence false percepts, but not expectations.
Further investigation showed that perceptual expectations gave rise to significant orientation specific activity in the deep layers and superficial layers (approx. 1 and 0.8 respectively) whereas false percepts did not activate either. However, the differences were not found to be significant between the false percepts and expectation-induced activity in these superficial and deep layers. A repeat of the repeated-measures ANOVAs with the different ROI gave no interaction of the ROI with the effects of interest indicating that two orientation-specific ROIs did not respond differently under the conditions of interest. Therefore, the authors concluded that high-confidence false percepts were associated with activity specific to the perceived orientation in the middle layers and perceptual expectations were related to activity specific to cued orientation in the deep and superficial layers. The effect of perceptual expectations in the deep layers was independent of whether there was awareness of the meaning of the cues or not. Therefore, it was concluded that false percepts can arise from feedforward activity in a cortical circuit and is different to that signalling perceptual expectations.
The investigation of the activity in the V1 showed that the effects of the V2 did not extend to it. Interactions were observed between layers (superficial, middle and deep), stimulus condition (high- and low-confidence false percepts, perceptual expectations) and ROI (V1 and V2) in line with the effects being specific to V2. The authors explained this by saying that relatively low spatial frequency gratings used in their study were more effective at activating V2 than V1. This was supported by the cross validated analysis of the orientation-specific BOLD signals within the functional localiser which showed stronger orientation-specific effects in the V2 than the V1 across all the layers (V1 – deep, middle, superficial as 3, 5, 6; V2 – deep, middle, superficial as 4, 8, 9). However, the authors stated that it was difficult to make separate inferences about V1 and V2 because of the overlap of the two areas. However, Haarsma and colleagues concluded that their investigation of the effects of the presented orientation on grating-present trials showed that low-contrast gratings embedded in noise evoked significant orientation specific activity in the superficial layers of the V2 but not in the other layers.
The third set of results described by the authors related to the noise patch control analyses. In order to establish that the noise patches themselves did not contain orientation signals in the V2, Haarsma and team generated four noise patches with a flat orientation energy spectrum. The authors found that the participants were still significantly biased towards specific orientations for some of the noise patches (noise patch 2 and 3 significantly identified more often as 45 degrees (63%) and 135 degrees (69%); noise patch 4 – vice versa). Noise patch 1 showed no significant bias (54% identified as 135 degrees).
In order to discount the explanation that stimulus-specific activity reported in the V2 middle layers on the high confidence false percept trials could be driven by an unspecified signal present in the noise patches themselves then two control analyses were carried out. In the first analysis, the authors used the premise that if the noise patches were driving the results then the effects would disappear if the noise patches contributed to estimated BOLD activity for the 45degree and 135 degree percepts equally independent of the level of perception. Haarsma and colleagues found that there was no confounding effect of the noise patches because high-confidence false percepts were reflected by increased middle layer activity compared with low-confidence false percepts and this was not observed in the other layers. The other control analysis carried out involved looking at whether the different noise patches produced any differences in stimulus-specific activity. This was found not to be the case and there was no evidence that the noise patches alone caused stimulus-specific activity. Therefore, Haarsma and team concluded that their control analyses confirmed that the effects observed reflected internally generated percepts and were not stimulus-driven signals in the noise patches themselves.
The final set of results described by Haarsma and colleagues related to the prediction of everyday hallucination severity by high confidence false percepts and reduced sensory precision. This involved the online experiment results where high-confidence false percepts with fMRI detected middle layer activity was correlated to the prevalence and severity of hallucinatory percepts described in daily life by the participants and measured by the CAPS questionnaire. The results showed that 22 of the 100 participants became aware of the meaning of the cue. There was increased confidence in having seen the grating with grating contrast, but there was no effect of cue validity on confidence and no interaction between cue validity and stimulus contrast. Accuracy increased with contrast and was lower when the expectation cue was invalid and participants who became aware of the meaning of the cue showed stronger cues effects on accuracy as shown by a group interaction of cue effect. The expectation cues influenced participants choice behaviour on grating-absent trials and this was driven by participants who became aware of the cue meaning who were significantly more influenced by the cues than those who were unaware of the their meaning. These participants only showed trend-level responses in line with the cue whereas those aware showed a significant effect of cue. The findings were said by Haarsma to be similar to the fMRI study where the effect of the cue was also driven by those aware of the cue purpose. Accuracy and confidence increased with grating contrast. Modeling by the authors of choice behaviour on grating-absent and grating-present trials showed that responses on grating-present trials were driven by the interaction of the current stimulus and contrast (termed sensory precision), previous response and current stimulus. The responses of the grating-absent trials were driven by previous responses.
With regards to the abnormal perceptual experiences in daily life (total CAPS scores), Haarsma and colleagues showed with their online study that the prevalence of the false percepts was positively correlated with the average confidence that participants reported on grating-absent trials, ie. prevalence of high-confidence false percepts. They also found that the sensory precision term, the influence of grating contrast on choice behaviour correlated negatively with abnormal perceptual experience scores. This indicated that the less sensitive participants were to stimulus contrast, the more likely they were to experience abnormal perceptual experiences in real life. Both confidence on grating-absent trials and sensory precision were found to be separate predictors of abnormal perceptual experience severity. The authors found no relation between average confidence on grating-absent trials and delusion ideation but there was a correlation with the sensory precision term. Therefore, Haarsma and colleagues concluded that high-confidence false percepts of oriented gratings were related to the severity of everyday abnormal perceptual experiences and these since they were reflected by stimulus like signals in the middle input layers of the early visual cortex could indicate that abnormal perception in everyday life may partly result from similar stimulus-like sensory fluctuations.
Having described their results, Haarsma and team then went on to discuss their findings. They firstly discussed the feedforward/feedback prediction theories of false percepts in light of their findings. They found that deep layers of the V2 visual cortex reflected perceptual expectations (feedback-induced stimulus templates) but the high confidence false percepts (ie. perception of a grating that was not present) were reflected in the stimulus-like feedforward activity of the middle input layers only and were unaffected by the perceptual expectation cues. The authors also found from their online study that these high-confidence false percepts correlated with everyday hallucination severity in their study participants. Therefore, Haarsma and colleagues concluded that false percepts could come from stimulus-like feedforward activity in the V2 and does not necessarily require feedback-induced stimulus templates in the deep layers of the visual cortex. This observation was said to put a stronger emphasis on feedforward signals than generally thought in the case of hallucination models which have been largely based on top-down expectations being the driving force. Haarsma and colleagues then give examples of other research that support this view and the view that false percepts can arise from spontaneous activity that resembles sensory input (Burke) rather than feedback from higher order regions. This forms the basis of circular inference models of hallucinations (ie. ascending loops of feedforward activity from weak sensory signals triggering perceptual hypotheses that are counted as sensory evidence and result in ´runaway overcounting loops`) and can be correlated to hallucinations and delusions observed in schizophrenia patients.
From a behavioural perspective, Haarsma and colleagues continued the discussion by saying that their findings showed that the prevalence of high-confidence false percepts correlated to everyday abnormal perceptual experiences and supported the results of other studies that showed that subjects that hallucinate are more prone to perceive stimuli in noise detection tasks. The authors also showed that there was a negative correlation to the sensory precision term which indicates that the less the reliance on contrast of the sensory stimulus in making the perceptual decision then the greater the level of hallucinations in everyday life. This finding was said to support that of others, ie. a reduction in sensory precision increases the influence of prior expectations possibly leading to hallucinations. However, the reduced reliance on sensory contrast and the confidence on grating-absent trials were found to be separate predictors of everyday hallucination severity which suggested to the authors that there are separate underlying mechanisms contributing to hallucinations.
Haarsma and colleagues then continued their discussion by looking at the role or non-role of perceptual expectations in the generation of hallucinations. They stated that their findings should not be considered as evidence against the role of top-down perceptual expectations in hallucinations especially in light of the large amount of indirect evidence available. They instead discussed how perceptual expectations could be involved in hallucinations of a ´different nature` to those they had seen in their study. It was said that their fMRI findings brought forward the suggestion that false percepts could occur from feedforward activity in early sensory regions mapping onto different forms of hallucinations such as minor phenomena and complex visual hallucinations. However, hallucinations induced by perceptual expectations could instead be reflected by signals in the agranular layers, mimicking the expectation effects in their study. The authors then proposed the question whether perceptual experiences could be generated by agranular feedback signals alone or whether additional activity of the input layers was required. It was also speculated that feedback signals may need to override activity in the middle input layers in order to experience expectation-induced hallucinations. Another suggestion was that top-down expectation effects could have played a role in their study as there was a strong expectation of stimulus presence regardless of content. By this, the authors meant that their participants strongly expected to see a grating on every trial independent of the grating orientation. This led on to them commenting that in fact expectations about stimulus presence versus absence and expectations about stimulus content have already been suggested as being supported by different neural processes. Haarsma and team therefore concluded that further investigation would be needed to elucidate this matter.
The next topics discussed related to associated aspects to the experiment. The first was feedback source into the V2 middle layer itself. Haarsma and colleagues said that in theory the V2 middle layer signals reflecting false percepts in their study could have resulted from feedback signals to the V1 or thalamus being sent downstream to the middle layers of the V2. However, the absence of stimulus-specific effects reflecting false percepts in the V1 and the lack of effect of the cued orientations on behaviour led to the conclusion that this type of indirect feedback was improbable although further proof would be needed. The second aspect was that in their study expectations was shown not to affect perception. This was put forward because both the normative sample and the expectations in their study were implicit and therefore, this could affect interpretation of the results. Others were reported to have found that conscious expectations exert stronger effects on perception than unconscious expectations and studies that report effects of implicit cues on perception typically reveal biased perception of existing stimuli rather than eliciting percepts de novo. In Haarsma`s experiments even though expectation cues did not affect perception, they did induce orientation-specific templates in the early visual cortex deep layers as seen by others. There was, however, unlike with others, significant expectation-evoked activity in the superficial layers which Haarsma explained could be due to the concurrent presentation of the noisy stimuli which could unlock modulatory effects of feedback in those superficial layers. Alternatively, the representation of expected orientations in the deep layers was reliable even in those participants who were not aware of the cue-stimulus relationship. This led to the suggestion that the brain could generate sensory expectations based on statistical relationships learned outside of conscious awareness.
The third aspect relating to Haarsma`s study was the role of the V1. Haarsma and colleagues` findings showed that there were no reliable effects of either perceptual expectations or false percepts in the V1. They explained this on for example lower spatial frequency stimuli using favoured V2 neurons versus higher 1 to 1.5cycles in other studies reporting V1 effects, but also stated that they do not locate their effects entirely on V2 activity. The reason given by Haarsma was because V1 neurons are also known to respond to low frequencies especially in the periphery and that the V1 and V2 ROIs used in the authors` experiments had a degree of overlap.
The final aspect discussed was the nature of the visual stimulus itself. Haarsma and colleagues used oriented gratings which they said likely caused stimulus-specific effects from the inhomogeneities in the spatial distribution of the orientation preferences across the visual field (eg. well-known radial bias). The authors questioned whether different stimulus types, eg. abstract shapes, complex objects would produce the same effects.
Therefore, in conclusion, Haarsma and colleagues stated that the orientations of falsely perceived stimuli with high confidence were reflected by activity in the middle input layers of the V2 and this indicated that a feedforward signal contributed to false percepts. In contrast, expectations (the most likely upcoming orientation) induced orientation specific activities in the deep and superficial layers of the V2 with no effect on perception. Therefore, false percepts were said to likely arise from low level content-specific fluctuations in the input layer of the visual cortex and was stated to put into question the view that false percepts are driven by top-down expectations. Haarsma and colleagues said that this explanation may have possible implications for disorders known to be associated with false perceptual inferences and/or hallucinations, eg. psychiatric disorders such as psychosis and schizophrenia and neurological disorders such as Parkinsons disease. Therefore, Haarsma and colleagues said that further research into the nature of the low-level sensory fluctuations is necessary as well as investigating whether false percepts can also be driven purely by top-down signals.
COMMENT
What makes this topic interesting is that Haarsma and collegues` work shows that in certain circumstances (ie. Haarsma`s simple grating image) visual event processing of individuals with schizophrenia can take two forms: the first is the visual pathway involving deep, middle and superficial layers of the V2 visual cortex resulting in correct perception of the presented visual event; and the second, a visual pathway involving the middle layers only of the V2 visual cortex resulting in false perception and incorrect identification of the presented image. The second form was found to be related to the severity of visual hallucinations experienced by the study subjects. Therefore, the work by Haarsma and team shows that false percepts arise from specific involvement of the middle layer of the V2 compared to other layers and hence, visual hallucinations (at least of a simple visual event nature) have the same roots. This comment looks at how the perception of simple visual hallucinations/false percepts may come to be ´skewed` in some circumstances.
In the case of Haarsma and colleagues` experiments, the study participants, who were known to suffer from schizophrenia and experience visual hallucinations to differing degrees, were presented with a simple visual event, ie. a circle of marked gratings (dark) that changed direction against a lighter noise background. The participants were given an auditory cue before presentation and then asked to indicate the direction of the grating. FMRI was simultaneously carried out on the V2 area so that decisions could be matched to neuronal firing of the different layers found in that area. The authors found that there was a level of accurate responses (ie. correct perception was carried out with regards to the presented image) and a level of inaccurate responses. These inaccurate responses were termed by Haarsma as false percepts and the experimental set-up was adjusted so that their frequency was high. This was done so that the fMRI data was of the highest quality possible.
If we look at the correct responses made, we can assume that visual input for the real-time presented event (ie. the gratings) follows the same physiological route as any other visual material and that perception follows normal principles. In general, perception of external events within the internally-generated event representation can be said to arise from four scenarios dependent on event characteristic source and modulation of the active characteristics. The first scenario would be where perception is based on input from the real-time sensory input only and follows ´rules` of event characteristic perception according to the physical visual system (eg. colour, shape, distance pathways). The second scenario would be the same as the first scenario but with additional modulation from top-down areas relating to the event and using learnt physical visual event ´rules`. This top-down influence would arise from knowledge from previous experiences of characteristics of real-time events and would result in either confirming or adapting the real-time input of the event`s physical characteristics. The third scenario involves even more top-down modulation and therefore, consists of first and second scenarios and also modulation coming from more personal knowledge recalled relating to events experienced such as emotional tags, interpretations, reactivated specific event details. The final scenario is purely internally generated and relates to input from top-down sources only (eg. during eyes shut, sleep, imagination) and as expected, perception of the event representation relies on knowledge reactivated from experienced personal events and again includes information about emotional status, interpretations and recalled event features.
In Haarsma`s experiment, the visual event requiring perception (the grating) consists of the physical characteristics of colour contrast of stripes against a ´noise` background (adjusted to the lowest contrast possible that would still be discernible) plus shifts of the locations of the stripes (ends – left to right, 45 or 135 degrees). Therefore, the visual pathway involved would begin like any other sensory event with the retinal cells (the rods – lines, edges, shadows) and with input going up the visual pathway hierarchy in order to manage the complexity of the image. For example, bipolar cell firing with their on-off, centre-surround specialised firing leads to the retinal ganglion cells capable of discerning movement (M-type), shape (P-type) and colour (nonM, nonP). This is followed by the appropriate firing of the lateral geniculate nucleus cells (LGN) where a cellular laminar organisation is first observed. This type of organisation allows the different visual characteristics to be separated for specific feedforward signalling and feedback signalling (80% of the LGN firing comes from feedback from V1 firing). It also allows firing of core features to be strengthened and lateral inhibition and priority to the unattended allows natural refractory periods of cells to be overcome and other new input (perhaps widening the number of event features inputted) to be added to the final event representation. The signalling then continues up the hierarchy to the visual cortical areas with the lower levels (V1 and V2) areas carrying all information to the higher cortical levels where event characteristics are dealt with separately (eg. movement – V3, V5, MST; object shape – V4, IT). This comment will restrict the discussion of the physical visual pathway to the V2 (with mention of the lower V1 level – primary visual cortex) since this is the area investigated with fMRI by Haarsma and team. Just like the LGN, both cortical areas have a laminar-type organisation which supports the discernment of distinct visual features and allows specialised input and output of the area.
In the case of the correct responses made in Haarsma`s experiments, activities in the V1 layers would reflect the ´grating` pattern of the dark/noise stripes presented. This is the first of the cortical regions to receive and process visual information and it responds to simple visual components such as orientations and direction. In general, it also contains a retinotopic map which organises spatial information from the visual field. Neuronal firing in this area early in time after visual stimulus presentation (up to 40ms) gives strong tuning to a small set of stimuli so that there can be discrimination of small changes in visual orientations, spatial frequencies and colours. Later in time from the visual onset, the area response involves firing in cells linked to a more global organisation of the event from feedback processing from higher areas and lateral connections from pyramidal neurons. The laminar and columnar organisation allows the visual event to be accurately represented by the firing patterns.
With regards to the laminar organisation, the V1 consists of 6 distinct layers but primarily layer IV and its sub-layers deals with information from the LGN (Huff) relating to shape and contrast (edge information) which is relevant to the presented event of Haarsma`s experiments. The largest number of LGN axons goes to layer IVC. The shape-important parvo-interblob pathway, with axons from the P-type ganglion cells of the LGN, terminate in layer IVC beta compared to the magnocellular pathway of M-type ganglion cells activation of the LGN terminating in layer IVCalpha. There is also a large amount of connectivity with activation of other higher cortical areas by the projections of the pyramidal cells of the V1 layers II, III and IVB, output projections such as layer V to the pons and superior colliculus and feedback from layer VI back to the LGN. There is also feedback signalling into the V1 which is thought to originate from the V2 and the higher levels V4, IT and MT. Therefore, it would be expected that the visual event characteristics presented in Haarsma`s experiments would activate normal visual pathway signalling in the V1 when the perception of stripe direction reported by the study participants was correct.
Just like the V1, all three visual pathways go from the V1 to the primary area investigated by Haarsma and colleagues that of the V2. The visual cortical area V2 (termed the secondary visual cortex, pre-striate cortex and one of the extrastriate areas) is also associated with the visual features of shape, colour, and movement. The three repeating striped substructure is specific for the pathways involved: thick stripes represent continuation of the magnocellular pathway; thin stripes the continuation of the blob pathway; and interstripes the extension of parvo-interblob pathway. The types of cells found in the area also help towards maintaining the complexities of the image, eg. complex cells and hypercomplex cells. The complex cells have larger receptive fields and are excited by bars of light in a particular orientation that have moved, irrespective of the precise location of image on retina. This would correspond directly to the grating type of event characteristic presented in Haarsma`s experiments. The other cells, the hypercomplex cells, receive input from complex cells and respond best to moving corners and ´stopped ends´ (i.e. produce stronger response to a line ending within the field than to one crossing it). These are probably not in demand in Haarsma`s experiments although they might cover directional change.
The striped structure of the V2 described above correlates to the different layers and different cells observed and the various functionality apportioned to them. As said by Haarsma, the V2 consists of basically 6 layers determined by differences in cell density between the layers. According to Haarsma and combining other studies (Balaram), layers I to III are termed the ´deep` layers. Layer I consists of sparse and densely myelinated cells (similar to that of V1) whereas Layer II has small, closely packed cells, is sparsely myelinated and has more punctate labelling of VGLUT2 and PV in chimpanzees than macaque monkeys (ie. demonstrates a species difference). Layer III is split into two with layer IIIA (the superficial layer) being moderately myelinated, containing small densely distributed cells and is only weakly labelled for PV whereas Layer IIIB (termed the deep III layer) is more densely myelinated and contains scattered distributions of variably sized cells that label strongly for PV. Layer IV (Haarsma and Balaram`s middle layer) is a wide, dense band of small granule cells, demonstrates moderate VGLUT2 activity and weak PV activity and whereas in macaques the cells are densely myelinated, in chimpanzees and humans they are moderately myelinated. The final group of layers as defined by Haarsma and Balaram are the superficial layers and these are Layers V and VI. Layer V is again split into two: a thin upper layer, the VA consisting of densely packed cells, separated from VB by a dense myelin band in macaques which is not seen in humans or chimpanzees; and the Layer VB which is wider with cells more distributed. The final layer, Layer VI, consists of a wide superficial band of distributed medium and large cells with dense myelination and weak PV reactivity plus it has a deeper band of scattered cells which diffuse into the white matter below the V2. This simple description shows how complex the physiology of the V2 is and how appropriate it is to the intricacies of the input it receives.
Just like with the V1, the activity of the V2 area is related not only to feedforward signalling from the real-time visual input but also feedback signalling from other areas and both types of signalling rely on the connectivity between the V2 layers and different regions. For example, the V1/V2 border demonstrates connectivity at layer IV of the V2 and IVC of V1 and there is connectivity to the V4 from the V3, which links the WHAT and WHERE visual pathways. There are also strong feedforward connections from the V1 as expected (direct and via the pulvinar) and feedforward connections to the V3, V4, V5 areas for the more complex visual processing relating to motion, form and figure-ground segregation. The feedforward signalling from the V1 is associated with the hierarchical processing of the visual stimuli with V2 extracting more information (eg. texture, depth, colour). However, there is also feedback signalling from the V2 to the V1 area and this has a modulatory role on the V1 neurons relating to attention, binding of features for perception and figure-ground segregation.
With regards to visual information, the firing of the V2 area shows excitatory characteristics with AChE reactivity of the neuronal cells indicative of cholinergic input. This is said to show species differences ( adult monkeys – largely absent; specific layers of chimpanzees and humans ie. 1 weak, 5 some, 2 and 4 none, 3 and 6 dense – Balaram). The information portrayed relates to the integrated information related from the V1 and is associated with the increased level of complexity and response patterns to objects. Therefore, the layers respond to differences in shape, colour, spatial frequency, moderately complex patterns and object orientation (depth and movement) (Huff). There are two interesting points about the way the layers deal with information and which are relevant to Haarsma`s study. The first is the presence of certain cells in the V2 that demonstrate dual selectivity for colour and spatiotemporal characteristics and these are 50% more common in the V2 superficial and deep layers receiving feedback connections from the higher areas compared to layer IV (the middle layer) that does feedforward signals (Shipp). Feedback as said before is known to be for modulating attention that will aid feature binding by prioritising certain characteristics/areas and filtering out others. This benefits the firing and functionality of the higher areas. Therefore, it has been suggested that dual selective neurons of the deep and superficial layers perform a ´bridging function` mediating the transfer of feedback induced bias for specific features to higher levels of specialised processing. Hence, there is promotion of the selection of the target objects representation and unification of the outcome of parallel, object-selective processes being carried out along the segregated visual pathways.
The second point concerning modular organisation of the V2 confirms this view. The V2 is described as basically having a modular organisation with a cyclic series of stripes (staining for cytochrome oxidase) that segregates the modular inputs from V1 and relays them up the visual pathway (De Yoe). The association of colour and spatiotemporal properties appears to be related to the layer organisation as described above, but there is also a modular structure within the layer which allows either cue invariance (eg. contours defined by motion, colour or contrast) or binding of attributes. This is achieved by the dual selective neurons present that combine selectivity for colour and direction of motion (Shipp). The dual selective neurons appear to be randomly organised within the layers and are probably more frequent in the feedback layers. Modular organisation (seen with the CO staining) appears to be more distinct in the middle layers (III and IV) of the V2 than the superficial and deep layers (I, II, V and V1). De-emphasis of the modular organisation occurs in the case of feedback to the V2. This is reported as feedback to the V2 from the higher areas V4, V5, MT and is found to be concentrated within the stripes responsible for feedforward firing but also into the intervening areas between the source stripes (Shipp).
Therefore, the visual characteristics observed and discerned from the presented visual grating image of Haarsma`s experiments go up the visual pathway from the V2 layers. In general and in short, the visual information is split into the WHERE pathway (dorsal pathway, perception-action model – areas V1, V2, V3, V5, MST) and the WHAT pathway (ventral pathway, object pathway, areas V1, V2, V4, IT) which are streams involved in specific forms of information processing (Mishkin). The areas have accordingly specific information determination strengths, eg. V3 – cells responsive to form, especially shapes of objects in motion; V4 – majority of cells responsive to colour, many to line orientation; and V5 (also known as MT middle temporal) responsible for motion. The streams have two distinct purposes: the WHAT pathway – object recognition, relatively simple stimuli lead to excitation of cells early en route, more complex stimuli make later cells fire where the physical location of object in visual field is less important; and the WHERE pathway – recognition of spatial relationships with spatial skills related to perception – localising points in space, depth perception, line orientation and geometric relations (judge angles), motion and rotation. In terms of Haarsma`s visual image both are required, ie. object recognition – the grating and contrast; spatial orientation – the shift in grating direction. Therefore, V2 firing in the case of correct perception requires both WHAT and WHERE pathway activations.
However, the perception of the visual input corresponding to Haarsma`s experimental visual image eliciting correct responses requires more than just visual input and the perception is likely to be as described in Scenario 1 and 2 at the start of this comment where there is application of learnt ´perceptual rules`. There are many ´principles/rules of perception` and these are mainly followed subconsciously having been obtained from life-long experience. A more detailed description will be given of some later when false percepts are discussed, but for now some examples applied to the perception in correct responses are: Gibson`s theory of direct perception, contour similarity, Biederman`s connectionist theory. Therefore, independent of the participants` schizophrenia and hallucination frequency, in the case of the correct answers relating to Haarsma`s presented image, visual information from that presented image follows the same visual pathway and perception rules for other objects in other conditions.
This then leads on to what is happening in the case of false percepts described by the same participants of Haarsma`s experiments. Visual hallucinations are defined as ´seeing images when there is nothing in the environment to account it for` with simple visual hallucinations including flashes or geometric shapes and complex ones that could show faces, animals or scenes (Teeple). Therefore, with regards to Haarsma`s experiments the false percepts could be described at best as simple hallucinations, ie. a failure to see lines or seeing lines in places they are not. Haarsma`s experiments involved participants suffering from schizophrenia where the participants recorded their personal levels of abnormal perceptual experiences in daily life (total CAPS scores). The team showed with their online study that the prevalence of the false percepts (ie. visual hallucinations, guesses) was positively correlated with the average confidence that participants reported on the grating-absent trials, ie. the prevalence of high-confidence false percepts. Haarsma and team also found that the sensory precision term, the influence of grating contrast on choice behaviour correlated negatively with abnormal perceptual experience scores. This indicated to the team that the less sensitive participants were to stimulus contrast, the more likely they were to experience abnormal perceptual experiences in real life. Both confidence on grating-absent trials and sensory precision were found to be separate predictors of abnormal perceptual experience severity and no association was found between average confidence on grating-absent trials and delusion ideation. However, there was a correlation with the sensory precision term. Therefore, Haarsma and colleagues concluded that high-confidence false percepts of their gratings image were related to the severity of everyday abnormal perceptual experiences and these, since they were reflected by stimulus-like signals in the middle input layers of the V2 visual cortex, were suggested as indicating that abnormal perception in everyday life may partly result from similar stimulus-like sensory fluctuations.
Schizophrenia is normally associated with auditory hallucinations (Chaudhury) but visual hallucinations are also reported with the nature of the hallucination being dominated by denatured people, parts of bodies, unidentifiable things and superimposed things (Chaudhury). The physiology behind visual hallucinations in general has been studied (eg. Boksa, Asaad, Manford) and a number of different physiological causes have been found, some specific for those experienced in schizophrenia. For example:
1)Disturbances in firing patterns from reduced grey matter volume, eg. in the superior temporal gyrus and primary auditory cortex observed with auditory hallucinations in schizophrenia sufferers but it is assumed it also applies to visual hallucinations with the V1 area being affected instead (Manford). There is also a reduction in volume of the dorsolateral prefrontal cortex reported in schizophrenia and this has been said to be linked to faulty frontotemporal connectivity (Manford).
2)Disturbances in firing patterns from lesions, eg. psychologically normal individuals with visual hallucinations have shown that these are associated generally with lesions of the visual pathway or any other sensory modality affected. Another example of firing pattern dysfunction are sufferers of Charles Bonnet syndrome which is linked to damage to the visual system such as macular degeneration and/or lesions between the retinal cells and the visual cortex.
3)Disturbances in firing patterns from other causes, eg. seizure activity/ ´irritation` of cortical areas responsible for visual processing such as primary visual cortex leading to simple visual hallucinations with higher areas leading to more complex hallucinations (Manford).
4)Disturbance in firing patterns from removal of inhibitory firing leading to hyperexcitability of an area, eg. same as that observed in phantom limb syndrome (Manford).
5)Disturbance of a particular area or connected areas, eg. there are reports in schizophrenia of impaired modulation of thalamocortical gamma activity which is known to be involved in perception and is normally constrained by sensory input and prefrontal and limbic attentional systems (Behrendt). Attentional mechanisms can then play a dominant role in the absence of sensory input leading to hallucinations. The uncoupling of the thalamocortical activity and sensory input may be influenced by factors such as acetylcholine nicotinic receptor abnormalities and dopaminergic hyperactivity (Behrendt).
6)Disturbance of the action of the reticular activating system`s action, eg. the reticular activating system dysfunction and lesions of the brainstem may lead to visual hallucinations (Manford). This is supported by visual hallucinations linked to certain sleep disorders and occur more frequently with drowsiness.
7)Disturbance of the action of the anterior cingulate cortex (Szechtman). It was found that the ACC is activated in hypnotizable people with auditory hallucinations and this is deemed likely to be linked to hypnotic suggestibility itself rather than the hallucination characteristics.
8)Disturbances of neurotransmitters (Asaad). For example, the uncoupling of the thalamocortical activity and sensory input in schizophrenia may be influenced by factors such as acetylcholine nicotinic receptor abnormalities and dopaminergic hyperactivity (Behrendt).
9)Disturbances of the processes involved in shifting information from the unconscious to the conscious (Asaad).
One particular cause of visual hallucinations in schizophrenia linked to perception rather than physiological changes of the relevant brain areas is suggested as a misattribution of internally generated stimuli (sensory – visual and auditory) as coming from an external source (Blakemore). This misattribution is suggested as linked to incorrect placement of sensory stimuli in place and time. Support for misjudgement of time comes from studies where in schizophrenia there are deficits in judging time intervals (Blakemore).
And so, how can we explain with regards to the visual input system and perception, Haarsma`s results of the visual hallucinations (the false percepts) by the same participants who report correctly when presented with the same grating image but at a greater contrast? Haarsma and team report that in the case of the false percepts, there is a feedforward signal representing the externally placed image that is reflected by firing in the middle layers of the V2. The orientations perceived of this image are incorrect (false percept). At the same time there is no stimulus of the deep and superficial layers of the V2 where expectations (in this case, the most likely upcoming orientation) from top-down modulation would likely to cause activation. Therefore, it may be argued that ´perception` of the real-time visual events experienced by Haarsma`s participants is ´skewed`. The explanations suggested here to explain this problem are based on the reported specific firing responses of the different V2 layers and therefore, they are divided into three groups: the first, reflects the lack of top-down modulation on perception; the second, a possible ´weak` feedforward signal that is misinterpreted (the strongest explanation); and thirdly, the possibility of reinforcement of incorrect perceptual rules brought about by the repetition of the trials.
The first possible explanation to the reporting of false percepts in Haarsma`s experiments relates to the lack of top-down modulation of the visual input. Top-down modulation can mean many things, eg. increasing attention on prioritised features due to emotional value, working memory focus, strengthened firing due to recall, but in the case of the V2 it is unlikely that these would be necessary because of the simple nature of Haarsma`s visual image. This indicates that the Scenarios 3 and 4 given above for interpretation of visual images are ruled out. The reasoning capability of the participant is seen not to be required even though the perception of the presented visual image was incorrect. This is because the prevalence of the false percepts is positively correlated with the average confidence the participants. Therefore, the participants feel no need for further reasoning or questioning of their judgements. This situation would have maybe changed if feedback of performance success had been given to the participants during the sessions. The other top-down modulation that could possibly occur could come from the visual areas higher up the hierarchy of the WHAT and WHERE pathways involved in the perception of more complex visual images. This is supported by the findings that the de-emphasis of the modular organisation occurs in the case of feedback (ie. from the higher areas V4, V5 IT) and these are observed in the superficial and deep layers of the V2. Again, lack of firing of these V2 layers suggests that top-down modulation within the visual pathways themselves also does not occur in Haarsma`s experiments.
The second explanation for the occurrence of the false percepts in Haarsma`s experiments with sufferers of schizophrenia is that visual input is a possible ´weak` feedforward signal that is then misinterpreted, ie. perception is ´skewed`. This is probably the strongest explanation and one favoured by Haarsma and colleagues with regards to the appearance of hallucinations in general. For example, disturbance in firing activity in areas responsible for visual processing such as the primary visual cortex V1 by means of seizure activity/ ´irritation` of cortical areas may lead to simple visual hallucinations (Manford). Haarsma and others suggest that the potential feedforward mechanism underlying their reported false percepts is due to spontaneous stimulus-like activity in the visual cortex and hence, resembling visual input, independent of top-down signals. This may be valid in the case of hallucinations in the absence of visual input but for Haarsma`s experiments, the false percepts arise on presentation of a real visual image. Therefore, an adaptation to the possible explanation is that the feedforward signal is weak and therefore, subject to a number of different subconscious perceptual processes that may be correct or not. There are several reasons why the feedforward signal may be weak. The first is that there may be possible lesions or damage of the visual pathway (or areas of it). This cause is suggested above for hallucinations in sufferers of Charles Bonnet syndrome which is linked to damage to the visual system such as macular degeneration and/or lesions between the retinal cells and the visual cortex. Alternatively, there may be a reduction in volume of the dorsolateral prefrontal cortex which is reported in schizophrenia and which has been said to be linked to faulty frontotemporal connectivity (Manford). These are thought to be unlikely in Haarsma`s participants since there is correct perception for some presentations.
The second possible explanation relates to a problem with visual attention as described above. During natural vision, attention leads to perception optimization of attended locations and objects (Ruff) by dynamically altering the visual representation of the behaviourally relevant objects ( Lukor). It does this in monkeys by simultaneously recording information from V1 neurons (Andersen) and the MT area (Ruff). It is known that V1 role creates a saliency map to guide shifts of attention (gaze shifts) so that visual location attracting attention is the highest firing neuron and also it can sustain activation in the absence of stimulus (Offen). For these, strengthening of firing feedback arising from the higher levels which includes the V2 (layers 2, 3A, 5,6 ie deep and superficial layers) is required. Therefore, a weak signal could mean that there is a lack of attention on the relevant features and more on the irrelevant. In the case of Haarsma`s visual image, attention may not be on the most important features of the grating, ie. the ends where the directional change would be most apparent and instead more on the noise rather than the stripe. With reference to the V2 a lack of feedback from this area proven by the absence of firing in the deep and superficial layers may mean that attentional modulation does not occur in these participants and certain relevant features are not given priority.
In a similar vein, with a weak signal or lack of signal the brain attempts to ´fill-in` missing details or ´bind` together features that should normally not occur if the event characteristics were rich. Neural representations of internally and externally generated signals are normally highly similar (Dijkstra) and therefore, in order to optimise perception or recognition when there is a weak signal details may be ´added` (´filling-in`) from past experiences. This adjustment or adaptation to what is actually being seen could be said to be one of the possible explanations on the same lines as ´skewed` perception and that makes up the third possible reason why false percepts are reported in Haarsma`s experiments.
With regards to ´skewed` perception, Haarsma found that the less sensitive participants were to stimulus contrast, the more likely they were to experience abnormal perceptual experiences in real life. This could be interpreted as ´skewed perception` for the visual features and suggests that Scenario 2 (input plus top-down influence relating to the real-time event using learnt physical visual event rules) again applies. The event requiring perception of its visual characteristics provided by Haarsma and colleagues consisted of colour contrast (bands of stripes against background) plus location of stripes ends (ie. gives angles of direction) and the V2 in particular is known to perceive certain characteristics (Bradley) such as segmentation (grouping and separating image segments), seeing very dim lines in absence of or with broken colinear flanking lines, and perception of edges (illusory contours) when the object and background have similar brightness (no contrast to define edge) (Bradley). Therefore, the perception theory applicable to Haarsma`s visual event and likely not to be affected in individuals having visual hallucinations is the theory of perceptual segregation based on Gestalt theorists and their Laws of Prägnanz, Proximity, Similarity, Continuance, Closure and Common fate (summarised as objects moving together are grouped together). This is the ability to work out which parts of presented visual information belong together and hence, form separate objects. Therefore, with a visual event based on a grating then the important figure-ground segregation would come from the darker stripe as the ´figure` and the background noise as the ´ground`. The success at seeing the grating means that the participants of Haarsma`s study have no problem with determining the segregation only the direction of the figure and therefore, the middle layer firing of the V2 at least can be attributed to this function.
However, there are also perception theories applicable to Haarsma`s visual event which are possibly affected in individuals having visual hallucinations and could be linked to the reported lack of activity in the V2 deep and superficial layers. These are given below:
1) Gibson`s theory of direct perception(Gibson, 1950) – Information that remains constant as the observer moves is important for perception (invariant information) and this includes consideration of texture gradient, flow pattern and horizon ratios. In Haarsma`s example, the false percept could come from a discrepancy in that the participant is regarded as having a ´moving position` and hence, the visual image is altered. An example of this is the visual illusion of the moving train with the individual misjudging who/what is moving.
2) Gestalt laws on perceptual organisation –Gestalt theorists tried to explain perceptual organisation on isomorphism,which means that the experience of visual organisation is mirrored by a precisely corresponding process in the brain. They assumed that the event representation formed from group neuronal activity was the same as what is in the visual environment. This is known not to be strictly true since as given above the event representation is more than just visual system activation patterns. It has to be assumed that in Haarsma`s experiments the event representation is an untrue image of the external event (false percept) and this comes about by a number of different mechanisms.
3) contour similarity – Perception theory follows two key principles: adjacent segmentsof any contour typically have very similar orientations; and that segments of any given contour that are further apart generally have somewhat different orientations (Geisler). In Haarsma`s experiments participants do not follow this perception theory since orientation errors of the contours of the grating are observed.
4)Marr`s representational theory (1982) – Marr described objects as a series of representations with increasing complexity beginning with the raw primal sketch leading onto the 21/2 sketch and 3D model. The simplicity of the visual image in Haarsma`s experiments probably means that the latter two are not relevant. However, the raw primal sketch probably does. This sketch contains information about light-intensity changes of the pixels leading to distortions of the grey-level representations of retinal image. To counteract this, light intensity values of neighbouring pixels are averaged so that the ´noise` is eliminated by the smoothing process. Since this smoothing process can also lead to valuable information being lost, several representations of the image are formed varying in degree of blurring. Information from these image representations is then combined to form the ´raw primal sketch`. Marr (1980) proposed that the raw primal sketch consisted of four different event characteristics: edge-segments, bars, terminators and blobs, each of which is based on a different pattern of light-intensity change in the blurred representations. The determination of false percepts in Haarsma`s experiments means that these event characteristics are misinterpreted by the individuals. This could mean that the ´smoothing process` leads to event characteristics being ´filled in` as described above. Under normal conditions, the filling-in of data is known to lead to object recognition.
5) Biederman`s recognition-by-components theory (1987) – This theory has the central assumption that objects consist of basic shapes or components known as geons (eg. as blocks, cylinders, arcs, spheres) and determination of geons is accomplished by early edge extraction using luminance and texture colour. Segmentation to parts or components where concave parts of objects contour is then used, followed by a decision as to which edge information from an object possesses the important characteristic of remaining invariant across different viewing angles. Biederman reported that there are five such invariant properties of edges: curvature, parallel, co-termination (edges terminating at a common point), symmetry, and co-linearity (points sharing a common line). The relevance to Haarsma`s study is that the Biederman theory says that even in sub-optimal conditions (ie. a weak signal) objects can still be recognised. Providing the concavities of a contour are visible, there are mechanisms allowing the missing parts of the contour to be restored since normally redundant information is still present. This is however, unlikely to be the case with the image and participants visual capabilities of the Haarsma experiment.
8) connectionist model (Hummel and Biederman 1992) – This theory highlights the ´binding problem` and relates to how different kinds of information are integrated to produce object recognition, eg. presentation of several objects at the same time and deciding which geons belong to which object. It consists of a seven-layer connectionist network taking as its input a line drawing of an object and producing as its output a unit representing its identity. According to Ellis and Humphreys (1999), the binding mechanism they employ depends on synchrony in the activation of the units in the network. Those units whose activation vary together are bound together with fast links to help ensure they are activated at the same time and therefore, so are the features they represent. According to the connectionist model, object recognition depends on edge information rather than on surface information such as colour. Therefore, with regards to Haarsma`s image, lack of event features and blurred contrast and noise edges would mean that incorrect binding of stripes could occur. This would result in the reporting of the false percepts.
Therefore, as given above a weak signal whether from a lack of attentional modulation from the V2, or not can lead to ´skewed perception` of the presented image in Haarsma`s experiments and the reporting of false percepts. The only proviso here to this explanation is that this might not be the case with Haarsma`s participants since visual illusions such as the Ponzo Illusion and Mueller-Lyer Illusion exist even if event details are adequate. The determinations are due to incorrect perception or changing perception due to the operating visual system at the time. However, there is also the Kanizsa`s illusory square which occurs due to filling-in of details and the Neckar cube which is due to lateral inhibition and priority to the unattended forcing a change in perception based on different event features. This leads onto another possible explanation which although unlikely should be considered and that is that incorrect perceptual decisions occur because they are reinforced by repetition to the point that the participant has a disconnect between actual visual input and the internally-generated visual representation. It is known that visual attention may be allocated to stimulus features that have been associated with reward on previous trials (Arvaidya) and therefore, as the experimental session continues the participants of Haarsma`s experiment filter only self-rewarded ´features` and wrongly use these to make their incorrect decisions. This is supported by the observation that perceptual history can modify bottom-up stimulus processing in the early visual cortex so that responses are biased in the light of previous perceptual decisions (De Jong) plus faced with inconclusive or conflicting perceptions there is a dominance of whatever perceptual interpretation has been commonly reported on a previous encounter (Brascamp). There also may be an element of visual imagery. In general, visual imagery adds sensory evidence to perception (Dijkstra) and there are reports of an actual experiment using gratings like in Haarsma`s studies that gradually appear in the noise whilst imagining a grating in a different direction (Dijkstra). In this case, it was found that imagery increased the perceptual stimuli (Dijkstra). It has also been reported that neural overlap between imagery and perception in the visual system correlates to experienced imagery vividness (Dijkstra). This vividness fluctuates with individuals and within individuals and depends on activity of large brain area networks including visual, frontal and parietal areas and therefore, this supports Haarsma`s finding that the frequency of false percepts correlates with the occurrence of hallucinations in daily life for the study participants in question. Other supporting evidence is that visual imagery can also affect future decisions since it can bias perception of subsequent ambiguous stimuli towards imagined stimulus (Pearson). Early studies found that simultaneous imagery decreased the likelihood that participants detected external stimuli showing that external input was incorrectly being attributed to imagery (Perky). In the case of Haarsma`s experiment then visual imagery could have an effect because of the weak signal and number of trials including those discarded due to high levels of contrast and correct responses. However, there is also evidence against previous experiences dominating real-time decisions and visual imagery since the ventromedial frontal lobe is seen to be important for the reward-based bias (Arvaidya) plus for visual imagery, no top-down modulation is observed and also visual imagery and perception only overlap and are not totally the same since visual imagery and perception of external events fire separate brain areas (Brogaard). This is also supported by visual working memory and perceptual expectations requiring representations in the deep and superficial layers but not the middle layers (Aitken) and Haarsma`s experiments say this is not the case with their grating image.
Therefore, what can be concluded from Haarsma and colleagues` investigations about visual hallucinations occurring in participants that have a prevalence to abnormal perceptions in their daily lives? It is assumed that a visual hallucination, defined as ´seeing something that is not there`, and manifesting as a neuronally-generated visual representation, has a large input from higher brain areas just as in expectation and visual imagery rather than purely from an externally generated signal. However, Haarsma and team showed that in the case of their simple visual image, activity in the V2, the brain area responsible for feedforward signalling of visual characteristics and feedback signalling onto the V1 to prioritise those relevant visual characteristics, that the activity was confined to those layers responsible for the externally sourced information. The deep and superficial V2 layers responsible for higher area order input showed little firing. Therefore, the incorrect responses by Haarsma`s study participants came from false perception of the externally sourced visual event. Even with an event consisting of contrast and direction of stripes, the participants demonstrated a high level of inaccuracy in their interpretation of what was being presented. The level of inaccuracy was comparable to the level of inaccurate perception of events in the participant`s daily lives and the confidence they had in their decisions. By looking at what could cause the false percepts of such a simple external visual event this comment shows that it is likely due to a weak signal from the V1. This comes about probably from a lack of top-down modulation into this area from higher areas (including the V2) relating to improved visual attention on high-priority event characteristics. This means that the level of details being transmitted up the hierarchy (including the V2) is impoverished. This results in ´skewed` perception using standard, learnt perceptual rules and based on poor or absent real-time characteristics. It is likely that visual events of greater complexity would elicit active neuronal groupings (the visual representation) that not only exhibit these same bottom-up problems but also have the added components of activated recall of past experiences and emotional tags. If this is the case with visual hallucinations in schizophrenia it is hard to suggest viable ways of improving perception. Increasing conscious visual attention on the real-time source of a protagonist for a hallucination in order to enrich the visual experience may be an answer but a lot more research is required on the physiology of hallucinations.
Since we`re talking about the topic…………………….
………slow repetitive transcranial magnetic stimulation (rTMS) has been reported to reduce auditory hallucinations in schizophrenia by reducing brain excitability (Zhang, Cole). If Haarsma`s experiments were repeated but with participants exposed to slow rTMS, would a change in visual hallucination frequency be observed and also be attributable to decreased visual pathway excitability?
……….. could a change in perception as cause for false percepts and a link to V2 middle layer processing be investigated if Haarsma`s experiments were repeated but instead of the gratings image a Gollin picture test used?
…………sensory deprivation is said to cause visual hallucinations (Waters). If Haarsma`s experiments were carried out directly after a period of sensory deprivation would changes in V2 layer activity further elucidate the mechanism of how false percepts occur?
…………. when two streams of visual stimuli are rapidly presented left and right, the second target in the left visual field is better identified than that in the right (Verleger). If Haarsma`s experiments are repeated but the image adjusted so that specific features are presented in the left and right visual fields, could the firing patterns of the V2 elucidate where any visual pathway differences in schizophrenic sufferers lie?
neurochitchat 