The Spatiotemporal Hierarchical Structure of the Brain in Auditory Memory Recognition and Predictive Coding

Background

This study aims to explore the hierarchical brain mechanisms involved in human identification of previously memorized music sequences and their systematic changes. While extensive research has been conducted on neural processing of visual-spatial patterns, our understanding of conscious recognition of auditory sequences and associated predictive errors remains insufficient. The auditory system extracts information from patterns and sequences that evolve over time, providing a unique opportunity to understand the brain’s temporal hierarchy. Related research suggests that the brain continuously updates internal models to predict external information and stimuli through the Predictive Coding Theory (PCT).

Research Origin

The authors of this paper include L. Bonetti, G. Fernández-Rubio, F. Carlomagno, M. Dietz, D. Pantazis, P. Vuust, and M. L. Kringelbach. They are affiliated with Aarhus University, Oxford University, University of Bologna, University of Bari, and the Massachusetts Institute of Technology, and the research was published in the journal Nature Communications in 2024.

Research Process

Experimental Design

In this experiment, 83 participants were asked to complete an old/new auditory recognition task while magnetic brain imaging (MEG) recordings were taken. First, participants memorized a short musical segment, then during the test phase, they were randomly presented with 135 five-note music sequences (27 memory sequences (M) and 108 new sequences (N)), and they needed to judge whether these sequences were previously memorized music fragments (“M”) or new variant sequences (“N”).

Data Collection

MEG data were recorded in a magnetically shielded room at a sampling rate of 1000Hz, with typical parameters selected for preprocessing. The head shape and head position indicators for each participant were recorded and used to calibrate the MEG data with individual MRI anatomical scans.

Data Preprocessing and Analysis

Independent Component Analysis (ICA) was used to remove eye movement and electrocardiogram artifacts, and the data were segmented and baseline-corrected. Support Vector Machine (SVM) was then used to decode different neural activities, evaluating the decoding accuracy of memory sequences (M) versus new sequences (N).

Univariate and Multivariate Pattern Analysis

Multivariate pattern analysis and temporal generalization analysis were conducted to decode different neural activities between memory and new sequences and assess their stability. The comprehensive results showed significant neural activity differences between M and N sequences.

Source Reconstruction and Functional Hierarchical Analysis

Source reconstruction techniques were used to identify the temporal and spatial distribution characteristics of brain activity. Dynamic Causal Modeling (DCM) was then used to evaluate neural hierarchical structures. By comparing evidence from different models, the validity of the hypothesized model was confirmed.

Induced and Event-Related Potential Analysis

Event-related potential analysis using complex Morlet wavelet transform revealed power changes and time-frequency analysis results in different regions of interest (ROIs).

Main Research Findings

Differences Between Memory and New Sequences

The results indicated that the auditory cortex could not distinguish different intensities of errors at the first deviant note, while bilateral hippocampal, anterior cingulate cortex, and medial cingulate cortex responses to the first deviant note were significantly stronger than to subsequent notes. This suggests that these areas are closely related to the brain signals of change awareness.

Brain Functional Hierarchical Structure

The study confirmed feedforward connections from the auditory cortex to the hippocampus, anterior cingulate cortex, and medial cingulate cortex, as well as feedback connections in the opposite direction. This hierarchical structure remained consistent throughout the sequence, with the medial cingulate cortex occupying the top position in this hierarchy at the final note. This suggests that when hearing the final note, the brain may be preparing for sequence classification (“memorized” or “new”), with the medial cingulate cortex playing a crucial role in decision-making and evaluation processes.

Functional Hierarchical Similarities and Differences

The hierarchical structure for sequence recognition remained generally consistent whether for memory sequences or variant sequences, but the temporal dynamics, intensity, and polarity significantly differed between the two. When predicted notes matched previously stored memories, the auditory cortex first, followed by the hippocampus, anterior cingulate cortex, and medial cingulate cortex, showed a positive response. Conversely, when predicted notes did not match, the same brain networks showed a negative response with a faster pathway.

Brain Wave Spectrum Analysis

Induced corresponding analysis indicated that after the sequence ends, the power in the alpha and beta frequency bands was significantly stronger in variant sequences than in memory sequences, consistent with previous research findings.

Conclusion and Significance

This study expands the application of Predictive Coding Theory (PCT) by providing quantitative evidence of hierarchical brain mechanisms during long-term memory recognition and predictive processing. The study suggests significant differences in the temporal dynamics, intensity, and polarity of the brain’s functional hierarchical structure during complex cognitive tasks. This research aids in understanding the neural basis of human perception and cognition and provides new insights for building more complex cognitive models.

Research Highlights

1. Provides quantitative evidence for the hierarchical mechanism of the brain during long-term memory recognition and predictive processing.
2. Finds significant differences in temporal dynamics, intensity, and polarity between memory sequences and variant sequences.
3. Confirms feedforward connections from the auditory cortex to the hippocampus, anterior cingulate cortex, and medial cingulate cortex, and feedback connections in the opposite direction.

This study demonstrates the brain mechanisms and hierarchical structure in complex cognitive tasks, providing strong support for a deeper understanding of Predictive Coding Theory.