Testing the predictive coding account of auditory processing
Bayesian theories of brain functions formulate perception as an active process making sense of the environment in a context dependent fashion. Predictive coding (PC) is a mathematical learning framework that the brain could entertain to that aim and whose implementation appears biologically plausible. In this project, we used PC to characterize the dynamics of perception.
Precisely, we focused on the mismatch negativity (MMN), a well-known auditory component that could reflect a context-sensitive prediction error under PC. We tested this hypothesis using simultaneous EEG-MEG and ECoG recordings. Trial-by-trial modeling enabled to reveal perceptual learning at play, within a fronto-temporal network. EEG-MEG data further supported an optimal adaptation of this learning in line with self-organization principles, as the perceptual reorganization induced by surprising events was found modulated by the global informative value of the sound sequence.
The ECoG study was performed in collaboration with Gerwin Schalk's group (Wadsworth Center, Albany, NY)
Automatic sensory adaptation in Autism
Predictive coding provides a potentially unifying framework of the autism spectrum (AS), linking its heterogeneous symptoms to the accumulating evidence of atypical perception. (.
In previous studies by our group, atypical learning dynamics in AS could be revealed in the visual and tactile domains (Sapey-Triomphe et al., 2016, 2018). In this project, we designed an MEG auditory oddball paradigm (yielding the MMN) to characterize implicit perceptual learning in AS.
This work is conducted in collaboration with Laurent Mottron (Montreal)
Infering cortical generators from surface signals
EEG and MEG signals provide an excellent temporal resolution for characterizing brain dynamics, but suffer from an unavoidable drawback: they remain surface measurements.
Infering the cortical sources at the origin of these signals involves solving an inverse problem and since the 1980s, numerous methods have been proposed to improve spatial resolution.
In this project, we could quantify the gain of combining EEG and MEG signals in the inverse problem based on simultaneous EEG-MEG empirical data. We found a finer model separation (disambiguation) with multimodal fusion as compared to unimodal EEG or MEG approaches.
Simultaneous EEG-MEG acquisitions are now routinely performed at the Lyon MEG platform (Cermep).
Evaluating a new generation of MEG sensors
Magnetoencephalography is a powerful non-invasive and millisecond-resolved imaging system to directly measure brain activity in humans.
Current sensor technology (SQUIDS) is constrained by technical specificities that limit potential applications in children and in clinical settings. They could be soon replaced by a new generation of sensors (OPMs) promising highly informed signals and exciting applications.
The MEG platform in Lyon contributes actively to this research in collaboration with the MAG4Health company and our team. And with already promising results in the visual and somatosensory modalities (Gutteling et al., 2023).
In this project, we evaluate the two technologies in the case of auditory responses.
​
SQUIDS: Superconducting Quantum Interference Device
OPMs: Optically Pumped Magnetometers