Albert Powers1, Christoph Mathys2, Philip Corlett1; 1Yale University, 2SISSA, International Advanced Studies Institute, Trieste, Italy
Perception is an active process, characterized by the building of an internal model of our environment, blending incoming sensory evidence with prior beliefs. Within this framework, hallucinations may result from increased weighting of these prior beliefs during perception. To test this, we adapted a classic sensory conditioning paradigm to functional MRI: participants were exposed to repeated pairings of a visual stimulus with an auditory stimulus. They subsequently reported the perception of the auditory stimulus even when it was absent, contingent on the presence of the visual—a conditioned hallucination. We recruited four groups: those with psychosis, both with hallucinations and without, healthy voice-hearers, and healthy controls. Conditioned hallucinations occurred with markedly increased frequency in those who had hallucinations outside the laboratory, regardless of psychosis status. They activated tone-responsive regions of auditory cortex and other regions active during clinical hallucinations. Computational modeling demonstrated an increased reliance on strong priors in voice-hearers (encoded by insula and superior temporal gyrus), regardless of psychosis status. By contrast, those with psychosis, regardless of hallucination status, were slow to update beliefs (cerebellum and hippocampus). Recently published in Science, these results may signify a means of distinguishing people with hallucinations from those without, and, orthogonally, people with a need for treatment from those without. Ongoing work focuses on applying these measures to risk stratification of people at clinical high risk of psychosis (CHR). Preliminary data suggest this approach holds promise for early detection of illness in CHR. Ongoing work aims to use these computational approaches hallucination-specific treatment selection.
