Close
The brain’s sensory regions receive prior cortical information over the longest timescales
Cooper, G., Blackburne, G., Das, R. K., & Skipper, J. I. (2023). The brain’s sensory regions receive prior cortical information over the longest timescales. Paper presented at the 29th Annual Meeting of the Organization for Human Brain Mapping, Montréal, Canada.
Introduction
Predictive accounts of brain functioning posit a mechanism through which cortical representations are arranged in a hierarchical manner, whereby higher-order representations constrain those beneath them. Under ecological conditions, ambiguous sensory information unfolds across multiple timescales simultaneously. For example, in language perception, phonemic disambiguation is facilitated via the integration of word and sentence-level information that unfold over longer timescales. As such, early sensory regions of the brain are likely to be situated at the end of a hierarchy of receptivity to prior cortical information over increasing timescales (Skipper, 2014). Conventional synchronous functional connectivity measures preclude the detection of such temporally extended functional architectures due to reliance on near instantaneous correlation between regions.
Methods
Here, we introduce whole-brain estimates of the temporal delay at which communication between regions takes place while processing ecologically valid stimuli (movies; Figure 1). We apply this technique to the Naturalistic Neuroimaging Database (NNDb), where 86 participants watch one of 10 feature length movies during fMRI in a 1.5T scanner. In practice, the whole-brain timeseries of each participant are segmented into three even chunks representing the beginning, middle, and end of each movie. In each segment, signals are extracted from a parcellation of 1024 regions of interest via the DiFumo functional parcellation technique. The delayed between a given ROI and a given voxel is estimated using the 3dDelay program within AFNI, which solves the Hilbert transformation of the envelope of the cross correlation function between these two signals; resulting in a sub-TR estimation of the delay at which cross-correlation is maximal. This yields one delayed connectivity map for each region of interest. All ROI-wise delay maps are concatenated within participants for subsequent aggregate statistical analyses. Additionally, the global average ‘delayed output’ of a given region is computed as whole-brain average delay of each map. Affinity propagationclustering is then used to define maximally separable topographies of delayed connectivity across delay maps.
Results
We find that primary sensory regions exhibit the highest median delayed connectivity received from regions across the cortex, as well as the highest variability across delays (Figure 2). Primary sensory regions also exhibit the highest delayed outputs to the rest of the cortex. These observations were stable across participants and conditions.
Further, we show that the delay architecture of the human brain can be meaningfully decomposed into 26 separable clusters, originating from separable sensory modalities, subcortical networks, and previously defined resting state networks, wherein delay maps originating from subcortical regions and higher order regions exhibit shorter global average delayed connectivities than those originating from sensory regions.
Conclusions
Our results provide evidence that under naturalistic conditions, the activity of putatively ‘unimodal’ sensory regions receive from, and transmit information to the cortex over the most temporally extended range of timescales. Within a predictive coding framework this may be facilitative of a mechanisms through which contextual representations encoded in the activity of higher-order cortical regions can constrain the activity of earlier members of sensory processing hierarchies. This has significant implications for contemporary models of temporal processing, prediction and consciousness as well as the use of canonical synchronous functional connectivity methods in neuroscience.
References
Dadi K., (2020) Fine-grain atlases of functional modes for fMRI analysis. NeuroImage 221: 117126.
Saad Z. S., (2003) Estimation of FMRI response delays. NeuroImage 18(2): 494–504.
