Analyzing movie-watching functional magnetic resonance imaging (fMRI) data from 176 young adults in the Human Connectome Project, researchers conducted a detailed analysis to outline the functional clusters within the human cerebral cortex. The study applied hierarchical clustering on time-averaged fMRI data acquired during naturalistic movie viewing, which resulted in a highly organized spatial distribution of functional activity across cortical regions. This approach provided insights into sensory, category-selective, and cognitive processing networks in the brain, offering a nuanced understanding of cortical parcellation.
Authors: Reza Rajimehr, Haoran Xu, Asa Farahani, Simon Kornblith, John Duncan, Robert Desimone
Key points
- The study analyzed fMRI data from 176 participants watching movies to explore the functional organization of the cerebral cortex.
- Hierarchical clustering identified 24 functional clusters, revealing sensory, category-selective, and cognitive networks.
- These clusters align with known brain regions and networks, ensuring neurobiological accuracy.
- Naturalistic movie-watching stimuli offered a dynamic perspective on brain activity compared to traditional static approaches.
- The findings provide a detailed cortical map, enhancing our understanding of sensory processing and cognitive functions.
Introduction
Human Cerebral Cortex has rich, functional complexity with its participation in sensory processing, cognitive tasks, and even social behaviours. Yet after so many decades of neuroimaging studies, the goal to successfully capture this naturalistic condition remains challenging. Using the high-resolution fMRI movie-watching paradigm, hierarchical clustering in Cortex 1 was evaluated by uncovering the functional clusters2 that proved robust and neurobiologically valid.
Methods
Participants listened to audiovisual movie clips, receiving 7T fMRI scans. Clips included a range of stimuli, such as people, animals, scenes, objects, and narratives, so that diverse cortical activity could be captured. Preprocessed data were transformed to a standardized cortical surface and the left and right hemispheres aligned. Averaged fMRI time courses were classified into ∼60,000 cortical vertices3, and with hierarchical clustering, these spatially coherent functional clusters were identified.
Key Findings:
- Clustering uncovered neurobiologically meaningful cortical divisions. Hemispheric symmetry and consistency across runs guaranteed reliability.
- Clusters mirrored known functional regions and networks, making it a holistic cortical map.
Study Participants and Methodology:
176 healthy young adults from the Human Connectome Project (HCP) database were utilized. Scanned in the movie clips using a 7T fMRI scanner while they watched them. It encompasses film and movie scenes of independent films and Hollywood movies that range from 1 to 4.3 minutes, all running for four cycles over the full scanning time of 60 minutes. These films presented a range of visual stimuli from people, animals, scenes, objects, sounds, music, speech, and storyline.
Results:
Preprocessing and Data Transformation:
Functional data were preprocessed and registered on a standard cortical surface where the left and right hemispheres were aligned. Data matrices were averaged across subjects to increase reliability and low-frequency fluctuations.
Time courses were de-meaned and concatenated across functional runs, thus enabling robust inter-subject synchrony.
Hierarchical Clustering:
Hierarchical clustering was used to determine the vertex distinct clusters based on the geometric distance in the activity space. The clustering result obtained is presented in an excellent fit with a correlation coefficient of 0.7436 as an outcome of the study.
This is a multi-scale hierarchical approach towards clustering which permits recursive subdivisions of any part of the cerebral cortex with the visual cortex. Spatial maps of clusters on the surface of the cortex revealed an apparent spatial organization of functionally defined clusters.
Reliability and Reproducibility:
High similarity between the clustering of vertices in the left and right hemispheres was found using metrics like the Fowlkes-Mallows index and adjusted Rand index. The clustering similarity was significantly higher for real data than for simulated/random data.
The clustering similarity decreased as the number of clusters increased but remained stable after reaching 24 clusters.
Functional Parcellation:
At the 24-cluster level, the maps of functional parcellation of both hemispheres were largely similar.
The four categories of clusters were found to be:
- Sensory cortices.
- Category-selective areas.
- Major cognitive networks.
A cluster corresponding to the anterior temporal cortex and regions with low fMRI responses during movie watching. These clusters corresponded to various cortical regions associated with sensory processing, social cognition, language, and executive functions.
Cluster Distribution:
Six clusters in the early visual cortex (V1–V4) lie along the dorsal-ventral axis and correspond to eccentricities of the visual field. Two clusters correspond to the auditory cortex: A1, belt, parable, A4, and A5. One big cluster corresponds to the somatomotor cortex. Seven category-selective clusters were identified for the following functions: Animacy (face) areas, animacy (body and motion) areas, object/tool areas, posterior-lateral scene areas, anterior-medial scene areas, extended scene network, and action perception network.
Seven cognitive processing clusters were identified for attention, language, social cognition, default mode network, and executive control networks.
Clustering Tree Analysis:
- The tree of hierarchical clustering had an interesting branching pattern between the visual cortex and animacy areas versus the peripheral visual cortex and scene areas.
- It shares the default mode, social cognition, and language processing networks in a tree, indicating an association of semantic, social, and linguistic representations in the cortex.
Response Variability and Eye Position Correlation:
Clusters were examined for response variability and correlation with eye position variability while movie-watching. The anterior temporal cortex showed the smallest response variability, whereas the correlation systematically increased from foveal to peripheral regions in early visual cortex, which is consistent with eye movements being driven by events in the visual periphery.
Topographic Relationship with Category-Selective Areas:
- The functional parcellation clusters were compared with classically defined category-selective areas using localizer maps.
- Animacy (face) and animacy (body and motion) clusters corresponded to the locations of the well-known face-selective and body-selective areas, including occipital face area and extrastriate body area, respectively.
- Object/tool clusters were overlapping with vertices of tool-selective neurons, and scene clusters were overlapping with scene-selective areas like the occipital place area and parahippocampal place area.
Activation and Responsiveness:
- Face, body, object/tool, and scene clusters showed high activity for frames that contained the respective categories, such as faces, bodies, or scenes.
- The “extended scene network” cluster was strongly scene selective and was significantly correlated with both posterior-lateral and anterior-medial scene clusters.
- This extended scene network, with an inordinate component in lateral prefrontal cortex, may process the high-level semantic features of scenes under a more naturalistic condition.
Cluster Locations in Lateral Temporal Cortex:
There were five clusters identified in lateral temporal cortex:
- Auditory cortex.
- High-level auditory cortex.
- Language processing network.
- Social cognition network.
- Default mode network.
These results showed that the cerebral cortex functional clusters were well identified and characterized during the process of naturalistic movie-watching, and thus demonstrated both sensory and cognitive processing networks in the brain.
Discussion:
The authors took a data-driven approach to functionally parcellating the cerebral cortex by using fMRI driven by audiovisual movie stimuli. Results were marked by the notable inter-subject synchrony of brain responses under naturalistic viewing conditions and application of clustering algorithms in partitioning the cortex into distinct functional regions. Hierarchical clustering was scale-independent, so it was possible to extract the clusters at different resolutions, and it turned out that the resulting 24-cluster map was identical with known cortical areas and networks, such as category-selective regions. The identified clusters were functionally annotated using anatomical location, stimulus response, and cognitive knowledge about large-scale cortical networks.
One of the main challenges in clustering is to determine an optimal number of clusters; the cortex is multi-scale organized. The research focused on finding functional networks at a macroscopic level but with a recognition of sub-areas at the mesoscale. This was in contrast to resting-state fMRI studies, since this approach identified dynamic interactions across distributed brain areas and offered better explanations for perceptual and behavioral phenomena. The parcellation revealed that some regions, such as those involved in face and body perception, showed clear distinctions when using dynamic movie data, which would have been harder to identify using static stimuli in localizer experiments.
Its approach has the benefits of detecting subtle functional selectivity in less-studied cortical regions and exploring functional interactions between networks. It also helped in identifying the role of stimulus correlations, such as those between faces, bodies, and motion, in shaping cortical responses. Despite the low statistical power due to the lack of repeated stimuli in movies, extensive signal averaging helped in revealing robust clusters with key roles in cortical processing.
The research also Identified certain clusters among which the action perception was associated with the lateral parietal and premotor cortex responsible for the mirror-neuron system of action understanding. Executive control networks responded at times when there was an unexpected transition from movie to rest, which indicates a surprise signal associated with unpredictability. These findings illustrate the utility of naturalistic stimuli for functional brain mapping and give insights into the cortical organization of perception and behavior.
Conclusion:
Overall, the study demonstrated how a movie-driven, data-driven parcellation approach could open new avenues for understanding the dynamic interactions of cortical areas with relevance to understanding the functional organization of the brain in more natural, ecologically valid contexts.
- outer layer of the brain that is involved in higher-level functions such as thinking, memory, decision-making, and sensory processing. ↩︎
- clusters refer to groups of functionally similar regions in the brain that exhibit coordinated or similar patterns of activity. ↩︎
- individual points or locations on the surface of the cortex ↩︎
