A PREDICTIVE CODING ACCOUNT OF AUDITORY FALSE PERCEPTION

Abstract

The Predictive Coding theory posits that the brain actively predicts upcoming sensory input through a hierarchical neural architecture and optimises the internal prediction model through prediction errors (PEs). This framework provides a valuable fundamental neural architecture to dissect the neurophysiology of auditory perception and auditory false perception. Thus, this thesis aims to 1) The interaction between different predictive coding components and factors that contribute to prior strength in perceptual inference that is more likely to drive auditory false percept, 2) The difference in neural signatures between input-driven and model-driven auditory false perception and 3) The dynamic pattern of maladaptive connectivity related to an aberrant predictive coding system in patients with chronic auditory false percept (i.e., tinnitus). Chapter 2 investigates interaction between different predictive coding components and the role of perceptual confidence in reinforcing the perceptual prior belief through the local-global oddball paradigm and the Conditioned Hallucination (CH) paradigm, which is induced by the associative learning of a visual cue with changing auditory stimulus. Those with higher proneness to CH also showed higher neural responses to the context-driven prediction error (cPE). The results suggest that high CH perceivers rely more on the internal model and perform increased responses to cPE than the low CH perceivers. Further, there was a significant positive correlation between relative confidence (confidence in "YES" relative to "NO") and response preference rate (answering "YES" relative to "NO") across all tone thresholds. And high perceivers displayed significantly greater relative confidence. These findings not only provide consistent evidence of the interaction among different predictive coding components but also extend the current mechanistic explanation of false perception. Chapter 3 dissects and distinguishes the neural signatures of two auditory illusion models, namely Zwicker Tone (ZT) induced by manipulating the input stimulus (input-driven illusion) and CH (model-driven illusion). During the perception of the ZT, the results showed altered activities in various sensory and high-order regions, such as frontal regions, insula and posterior cingulate cortex (PCC). In addition, a combined bottom-up and top-down connectivity network was mainly centred at the inferior temporal gyrus (ITG), and parahippocampus (PHC). By contrast, CH induced the decreased activation of the entorhinal cortex (EC), ITG and PHC but increased activation of the insula. The CH network was more top-down modulated, especially from the EC and PCC. These findings provide a more nuanced perspective to directly compare and tease apart the neural mechanism of auditory false perception. Chapter 4 investigated the aberrant hierarchical PE processing in patients with chronic tinnitus by analysing the time-varying connectivity of the auditory predictive coding networks. We observed a decreased connectivity strength to the stimulus-induced PE. By contrast, the results demonstrated an increase in the connectivity of the DMN-dominant network for processing cPE in the tinnitus group, suggesting the orientation to an internal state in tinnitus. Also, tinnitus patients showed stronger connectivity to the PHC and ventromedial prefrontal cortex (vmPFC) for establishing prediction during the cPE condition. Taken together, this thesis improves our understanding of the involvement of the predictive coding system in the mechanism of auditory false perception, physiologically and pathologically, providing valuable insights for theoretical models of perception. The findings could prompt potential clinical applications, offering clearer targets and pathways for improved diagnostics and therapies for auditory disorders, thereby bridging auditory neuroscience and practical healthcare solutions.