Use this URL to cite or link to this record in EThOS:
Title: Perceptual Bayesian inference in autism and schizophrenia
Author: Karvelis, Povilas
ISNI:       0000 0004 9349 4686
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2020
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Recent theories in the field of computational psychiatry regard schizophrenia (SCZ) and autistic spectrum disorders (ASD) as impairments in Bayesian inference performed by the brain. In Bayesian terms, perception is a result of optimal real-time integration of sensory information (’likelihood’), which is intrinsically noisy and ambiguous, and prior expectations about the states of the world (‘prior’), which serve to disambiguate the meaning of the sensory information. Priors capture statistical regularities in the environment and are constantly updated to keep up with any changes in these regularities. The extent to which prior or likelihood dominate perception depends on the uncertainty with which they are represented, with less uncertainty resulting in more influence. Individuals with ASD and SCZ might show impairments in how they update their priors and/or how much uncertainty there is ascribed to prior and likelihood representations, leading to differences in inference. While this Bayesian account can be argued to be consistent with many previous experimental findings and symptoms of SCZ and ASD, recent experimental work inspired by these ideas has produced mixed results. In this work, we investigated possible Bayesian impairments in SCZ and ASD experimentally by addressing some of the methodological limitations of the previous work. Most notably, we used an experimental design that allows to disentangle and quantify separate influences of priors and likelihoods, and we tested both SCZ and ASD patient groups as well as autistic and schizotypy traits in the general population. We administered a visual motion perception task that rapidly induces prior expectations about the stimulus motion direction, leading to biases and occasional hallucinations that can be well described by a Bayesian model. In this task, autistic traits were found to be associated with reduced biases, which was underlied by more precise sensory representations, while the acquired priors were not affected by autistic traits. Patients with ASD, however, showed no evidence of increased sensory precision, while there also were no impairments in the acquisition of priors. We also found no effects in the acquisition of priors or sensory representations along schizotypy traits and in patients with SCZ. However, under conditions of high ambiguity SCZ patients were less likely to hallucinate the stimulus than controls. The second part of the thesis is focused on further exploratory analyses conducted using these same datasets. First, we investigated post-perceptual repulsion effects in our task and whether they were related to trait or group differences. We found clear evidence of repulsion from the cardinal directions. In addition to that, we found evidence for a repulsion from the central reference angle, which was randomly selected for each participant and which could only be inferred from the stimulus statistics. Furthermore, we found the repulsion from the central reference angle to be reduced along schizotypy traits. Interestingly, in both SCZ and ASD groups this repulsion was also found to be negligible. While these results are exploratory, they might point to a trans-diagnostic features of ASD and SCZ. Second, we investigated within-trial dynamics of evidence accumulation by constructing a Continuous Choice Drift Diffusion Model (CDM) – an extension of the classical binary choice drift diffusion model. The results of this model showed that increased sensory precision along AQ found in a Bayesian model was underlied by faster drift rates, while slower responses and reduced hallucinations in SCZ were explained by a larger decision threshold. In addition, this model provided a more complete characterization of the performance in this task (by including reaction times) and it serves to emphasize the importance of accounting for exposure to stimulus duration and judgement time in future studies investigating Bayesian inference. Together, this work provides novel experimental evidence that speaks to the hypothesis of impaired Bayesian inference in ASD and SCZ. Furthermore, the analysis of reference repulsion effects and within-trial dynamics provide additional insight related to SCZ and ASD differences that extend beyond the Bayesian framework.
Supervisor: Series, Peggy ; Lucas, Christopher Sponsor: Engineering and Physical Sciences Research Council (EPSRC)
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
Keywords: computational psychiatry ; statistical regularities in the environment ; autism spectrum disorders ; schizophrenia ; computerized visual motion estimation tasks ; biases in perception