Predictive Processing of Multiple Acoustic Features in the Auditory Cortex

聽覺皮層中多種聲學特徵的預測處理

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date29 Jun 2021

Abstract

Brain responses that differentiate the processing of unexpected deviant stimuli from expected standard stimuli are often quantified as mismatch negativity (MMN). The MMN (a difference in event-related potentials based on non-invasive recordings evoked by sensory deviants and standards) has long been recognized to reflect novel stimulus processing. Previous studies in humans (Phillips et al., 2015) suggest that MMN is related to deviance detection based on sensory prediction violations. However, it has also been suggested that predictions of stimulus contents (“what”) vs. stimulus timing (“when”) have different putative mechanisms (Auksztulewicz et al., 2018b). While the latter study raises the possibility that violating predictions of different stimulus attributes can be dissociable, it is unknown whether this functional specialization extends to different content-based predictions, such as acoustic pitch, location, duration, and spectral composition. Furthermore, since deviant detection rests on extracting a memory representation of recent stimulus statistics, the MMN can be used as an indirect measurement of sensory memory formation. However, other more direct measurements, such as intracranial recordings from animal models, are needed to elucidate the neural mechanisms of auditory memory formation.

In this thesis, I investigated whether violations of acoustic predictions based on different stimulus features (pitch, duration, location, or formant) are reflected in differences in neural activity in human and animal models. The first two experiments used a roving oddball experimental paradigm, adapted from a previous study (Garrido et al., 2008). In the first experiment, I investigated whether feature-specific differences in the spatial distribution of MMN responses can be mapped onto different cortical regions using electrocorticography (ECoG) recording in the rat auditory cortex. We demonstrated that MMN could be observed following the violation of four independent acoustic features, but that these MMN responses show a large degree of heterogeneity not only across different acoustic features but also across individual animals.

In the second experiment, I applied the same paradigm to humans using electroencephalography (EEG). As in the rat study, we found significant mismatch responses following prediction violations of four independent acoustic features. While no significant differences were observed between MMN signals corresponding to different stimulus features in a traditional univariate analysis, acoustic feature information could be decoded from the fine-grained topography of mismatch responses based on multivariate analysis. Consistent with previous studies, the results indicated that deviant detection along different stimulus features could be linked to differences in the spatial distribution of neural responses.

In the third experiment, I used a direct manipulation of auditory memory formation and recorded auditory cortical activity in anesthetized rats to infer the neural correlates of auditory pattern learning, differentiating between repeated sequences and novel sequences. While no significant differences between re-occurring and new sequences were observed in event-related potentials (ERPs), when comparing responses in the time-frequency domain, we found robust differences between re-occurring and new sequences. This result was most prominent in the beta frequency band, indicating a neural correlate of learning formation in passively listening and anesthetized animal models.

In conclusion, the thesis shows that the neural correlates of sensory deviance detection depend on the type of sensory feature whose violation underlies deviant stimulus presentation and that the auditory cortex is susceptible to auditory memory formation both through indirect measurements (mismatch responses) and more direct measurements (repeated presentation of the same stimulus sequence).