Compensatory shifts in visual perception are associated with hallucinations in Lewy body disorders

Visual hallucinations are a common symptom of Lewy body disorders (LBD) and are associated with increased distress in both patients and carers. There is a need to develop an understanding of the mechanisms underpinning visual hallucinations to inform the management and treatment of this symptom. We report the first investigation into the effects of visual priming on a complex visual illusion that is posited to be a useful analogue of visual hallucinations. We show that visual hallucinations are associated with a tendency to accept illusory perceptions as real, in order to avoid missing true perceptions. This finding suggests that the underlying mechanism of visual hallucinations in Lewy body disorders may be one of compensation, rather than deficit. Furthermore, the results suggest that environmental cues can influence a person’s tendency to perceive visual information as real. The findings support a novel mechanism for explaining the interaction between environment and hallucinatory episodes in these disorders. An implication for interactive models of visual perception is that in the face of perceptual (bottom-up) impairment, top-down factors act to compensate for this impairment. By extension, providing patients with LBD with rich visual environments that eliminate visual ambiguity, and providing environmental cues to aid their perception, may reduce hallucination frequency in this population.

Visual hallucinations—the involuntary experience of seeing something that is not veridically present while awake (Collerton, Perry, & McKeith, 2005)—are a common occurrence in Lewy body disorders (LBD) such as Parkinson’s disease, Parkinson’s disease dementia, and dementia with Lewy bodies (Fénelon & Alves, 2010; Ferman et al., 2013; Kitayama, Wada-Isoe, Nakaso, Irizawa, & Nakashima, 2007) as well as a number of other disorders and normal states. The psychosocial impact of visual hallucinations can be high in LBD and has been associated with increased distress (Mosimann et al., 2006), co-morbid mental health problems (Bjoerke-Betheussen, Ehrt, Rongve, Ballard, & Aarsland, 2012), and increased carer distress (Lee, McKeith, Mosimann, Ghosh-Noydal, & Thomas, 2013).

Current theoretical explanations of visual hallucinations, while differing in emphasis, broadly agree that impairments in both visual perception and attention are implicated (Collerton & Mosimann, 2010). Models such as the attentional network (Shine, Halliday, Naismith, & Lewis, 2011), Attention, Input, Modulation (Diederich, Goetz, & Stebbins, 2005), and Perception Attention Deficit (Collerton et al., 2005) all propose to differing degrees that combined impairments in both “bottom-up” perceptual functions and “top-down” attentional processes lead to the intrusion of erroneous object representations—hallucinations—into the visual scene.

Consistent with these models, combined impairments in top-down and bottom-up processes are risk factors for visual hallucinations in LBD (Barnes & Boubert, 2008; Bronnick, Emr, Tekin, Haughen, & Aarsland, 2011; Gallagher, Parkkinen, O’Sullivan, Spratt, Shah et al., 2011; Imamura, Wada-Isoe, Kitayama, & Nakashima, 2008; Meppelink, De Jong, Teune, & Van Laar, 2009; Ozer, Meral, Hanoglu, Ozturk, Cetin et al., 2007; Ramírez-Ruiz, Junqué, Martí, Valldeoriola, & Tolosa, 2006, 2007; Straughan, Collerton, & Bruce, 2016). This focus on factors within participants, however, neglects potential interactions with the environment and does not provide an explanation of why hallucinatory episodes tend to occur at specific times and locations and to be tightly integrated with the visual environment (Collerton, Taylor, Tsuda, Fujii, Nara et al., 2016).

Investigating individual hallucinations, given their unpredictable episodic nature, is difficult. The recent development of the pareidolia task as an experimental analogue of visual hallucinations has addressed this gap (Uchiyama, Nishio, Yokoi, Kirayama, Imamura et al., 2012). Pareidolia are a class of visual illusion whereby a visual object is perceived in visual information which is itself meaningless, ambiguous, or bears no veridical relationship to what is perceived by the viewer. “Seeing faces in clouds” is a naturally occurring example (see Fig. 1 for examples of naturally occurring pareidolia). Studies have demonstrated considerable overlap in the phenomenology of pareidolia and visual hallucinations (Uchiyama, Nishio, Yokoi, Hosokai, Takeda et al., 2015). Moreover, hallucination severity is positively correlated with experimental pareidolia frequency and both respond to cholinesterase inhibitors (Yokoi, Nishio, Uchiyama, Shimomura, Iizuka et al., 2014). Uchiyama et al. (2012) found that the number of illusory perceptions in pareidolia images was able to discriminate between participants with dementia with Lewy bodies and Alzheimer’s disease with 100% sensitivity and a specificity of 88%.

Signal detection theory offers a potentially fruitful avenue for testing interactional models, because it provides a way to measure bottom-up perceptual processes (i.e., sensitivity) and higher order “top-down” processes (i.e., response bias) at the same time (Dolgov & McBeath, 2005). Yokoi et al. (2014) applied signal detection theory to their pareidolia test by varying the absence/presence of real visual objects among ambiguous visual noise. They demonstrated that, compared with participants with Alzheimer’s disease, participants with dementia with Lewy bodies had significantly reduced ability to discriminate visual “signal” from “noise,” and were more prone to report the presence of a visual object when one was not present. These perceptual effects were suggested to be related to the risk of visual hallucinations.

A recent study from our group (Straughan et al., 2016) examined the role of priming of visually ambiguous images (e.g., silhouettes, fragmented images) in hallucinating and non-hallucinating participants with Parkinson’s disease (this was a separate sample from the current study). Significant priming effects emerged across both groups, demonstrating the preserved priming in participants with LBD. Hallucinating participants were significantly more impaired in resolving visually ambiguous images. Because all trials in this study involved the presence of an object (i.e., there was always something to “see”), the question remains as to what happens when hallucinating participants are primed in the absence of a veridical percept. Pareidolia offer a way to investigate this, as the absence/presence of an object within visual noise can be manipulated.

The current study aimed to extend the findings of Straughan et al. (2016) and Yokoi et al. (2014) by assessing differences between patients with and without hallucinations and by adding associative priming to experimentally examine the effects of environmental factors.

The results corroborate previous findings that pareidolia are a good analogue to visual hallucinations (Uchiyama et al., 2015; Yokoi et al., 2014). In the present study, there was considerable overlap between the phenomenology of both and hallucinating participants were more prone to pareidolia than their non-hallucinating counterparts. Thus, the previously established difference between groups prone to hallucinations (LBD) and those not (Alzheimer’s disease or controls) is specific to those participants who do hallucinate.

Non-hallucinating LBD participants were indistinguishable from healthy controls on both signal detection measures. Consistent with the hypothesis that hallucinations are associated with an impaired ability to separate visual signal from noise, hallucinating LBD participants displayed significantly lower perceptual sensitivity than the non-hallucinating groups.

Hallucinating LBD participants exhibited a significantly lower response criterion during unprimed trials, in comparison to non-hallucinating LBD participants. The presentation of a prime item eliminated this difference. On the other hand, no significant priming effects emerged when unprimed versus primed trials were compared within groups.

During the experiment, over half (9/16) hallucinating participants reported multiple pareidolia within a single trial. Traditional signal detection measures are unable to account for this fully, as they measure absence/presence of illusory perception, rather than absolute number of perceptions within a trial. Using signal detection analysis may therefore be overly conservative in this context. When the absolute number of illusory perceptions (pareidolia rate) was considered, it emerged that the presentation of a prime item reduced the number of pareidolia experienced by hallucinating participants, although relative number of pareidolia in this group remained high in comparison to non-hallucinators.

The lower criterion score in hallucinating participants during unprimed trials may be indicative of a downward shift in response bias in order to compensate for poor perceptual sensitivity. In other words, hallucinating LBD participants were more “liberal” in their response bias to avoid missing veridical percepts, at the cost of occasionally accepting visual noise as a percept. This is consistent with the proposal of Dolgov and McBeath (2005) that hallucinations are underpinned by a more relaxed response bias in order to maximize the number of perceptual “hits” at the expense of an increase of “false alarms.”

Perceptual systems are viewed as inherently self-optimizing and will adjust sensitivity/response bias thresholds within the context of the environment and limitations imposed on the system (Lynn & Barrett, 2014). It has recently been proposed that neural systems will reorganize themselves when limits are imposed upon them, in order to maximize their evolutionary “fitness” (Yamaguti & Tsuda, 2015). Thus, the observed differences in hallucinating participants in the current study could be conceptualized as a compensatory response in the face of Lewy body pathology.

Introducing a prime stimulus had only a small and unreliable (in the case of VH–/VH+ participants) effect on sensitivity. This suggests that environmental cues do not significantly improve sensitivity to ambiguous information. However, the introduction of a prime item did eliminate significant differences in response bias between the groups. In addition, examination of pareidolia rate indicated that the presence of a prime reduced the number of illusory perceptions experienced by hallucinating participants, although their absolute number remained higher than the other groups. These converging findings provide some evidence for the ameliorative effects of environmental cues on frequency of illusory perceptions, although given the limited nature of these findings, this conclusion can only be tentative.

Priming visual environments to reduce illusory perceptions might have practical implications for the care of LBD patients; reducing visual noise and increasing environmental cues may lead to a reduction in hallucination frequency. Guidance on the design of “dementia-friendly” environments is already in existence (Fuggle, 2013) and further research into this area may offer new insights to develop these strategies further in order to discover what kind of visual cues are most effective. It should be noted, however, that in the present study, even after priming, hallucinating participants still had a lower sensitivity score than their non-hallucinating counterparts, as well as a higher number of pareidolia. While environmental cues may therefore play a role in reducing the frequency of illusory perceptions, it is unlikely that they will go as far as to eliminate them.

There are some limitations in the study that indicate a need for caution. First, the sample size was modest (although comparable to similar studies), meaning that the study might have lacked sufficient power to detect some significant results. In addition, while the groups were matched on age, education level, and number of medications taken (in the case of the patient groups), they differed on cognitive ability, visual acuity, disease duration, and diagnosis (there was a higher proportion of patients with Parkinson’s dementia and dementia with Lewy bodies in the hallucinating group). There were also significantly more men in the hallucinating group. The influence of these variables cannot be ruled out.

The differences in visual acuity between hallucinating and non-hallucinating groups is of note. The significant drop in sensitivity score in hallucinating patients may have been related to the significantly poorer visual acuity in this group, rather than due to the neurocognitive changes that are associated with LBD. However, the visual acuity of “low acuity” non-hallucinating patients was indistinguishable from acuity scores of hallucinating participants, yet sensitivity scores remained significantly lower in the hallucinating group. We would contend that this evidence goes some way to ruling out the confounding variable of visual acuity.

In conclusion, pareidolia offer a useful experimental tool to investigate the mechanisms of visual hallucinations. The current study incorporated this class of stimulus with a priming paradigm in order to assess top-down and bottom-up factors in visual hallucinations in LBD. Extending previous research, the findings suggest that only people who hallucinate are poor at discerning visual signals from visual noise. In addition, this deficit in perception appears to be compensated for by a more liberal detection threshold, enhancing the likelihood of perception at the cost of simultaneously enhancing the rate of hallucinatory perception. Environmental factors such as visual cues may therefore potentially modulate the chance of hallucinations.