Abstract
Our daily life is realized by the complex orchestrations of diverse brain functions, including perception, decision-making, and action. The essential goal of cognitive neuroscience is to reveal the complete representations underlying these functions. Recent studies have characterised perceptual experiences using encoding models. However, few attempts have been made to build a quantitative model describing the cortical organization of multiple active, cognitive processes. Here, we measure brain activity using fMRI, while subjects perform 103 cognitive tasks, and examine cortical representations with two voxel-wise encoding models. A sparse task-type model reveals a hierarchical organization of cognitive tasks, together with their representation in cognitive space and cortical mapping. A cognitive factor model utilizing continuous, metadata-based intermediate features predicts brain activity and decodes tasks, even under novel conditions. Collectively, our results show the usability of quantitative models of cognitive processes, thus providing a framework for the comprehensive cortical organization of human cognition.
Introduction
While cognitive neuroscience has made significant progress in mapping sensory perception—such as vision or audition—using quantitative encoding models, mapping "active" cognition remains a challenge. Complex processes like mental calculation, memorization, and decision-making involve intricate orchestrations across the brain. Most previous research has either relied on passive stimuli or focused on a narrow set of tasks, leaving us without a unified map of how the brain organizes the vast array of cognitive functions used in daily life. This study addresses this gap by building a quantitative model that describes the underlying latent features linking diverse cognitive tasks.
Methods and Experimental Design
The researchers recorded whole-brain fMRI data from six healthy subjects while they performed 103 distinct naturalistic tasks. These tasks spanned a wide range of cognitive domains, from simple motor responses to complex logical inferences and memory challenges.
To analyze the resulting brain activity, the team employed two types of voxel-wise encoding models:
- Task-Type Model: A sparse model representing tasks as binary labels, used for Hierarchical Clustering Analysis (HCA) and Principal Component Analysis (PCA) to visualize task relationships.
- Cognitive Factor Model: The core innovation of this work. Instead of simple labels, each task was transformed into a 715-dimensional continuous feature space using metadata from Neurosynth (a large-scale automated synthesis of functional neuroimaging data). This allowed the model to characterize tasks based on "cognitive factors" like "working memory" or "spatial processing."
Key Findings
The study revealed several key insights into the organization of human cognition:
- The Geometry of Thought: Hierarchical clustering identified six major task clusters: Visual, Auditory, Motor, Language, Introspection, and Memory. Within these, tasks were further organized by specific properties, such as the emotional valence of images.
- Continuous Cognitive Gradients: Rather than being organized into discrete, isolated modules, the brain represents tasks along a continuous gradient. PCA showed a gradual shift from simple sensory-motor processing to abstract, complex cognitive functions.
- Topographic Entanglement: Multiple cognitive dimensions were found to be "entangled" within specific brain regions. For instance, the Inferior Parietal Lobule (IPL) exhibited a fine-grained topographic shift from motor-related activity to calculation and logical inference.
- Zero-Shot Generalization: Most significantly, the cognitive factor model demonstrated high predictive power for entirely novel tasks. By training on 80% of the tasks, the model could predict brain activity for the remaining 20% of unseen tasks in approximately 86.2% of cortical voxels, with a mean decoding accuracy of 96.0%.
Discussion
This framework provides a powerful tool for quantifying the cortical organization of human cognition. While the study was limited by a small subject cohort and the temporal resolution of fMRI, it demonstrates that a model trained on a sufficiently broad battery of tasks can generalize to new cognitive "recipes." Future work may apply this approach to explore individual differences, clinical conditions, or to develop more sophisticated brain-machine interfaces capable of decoding complex human intent.
Summary updated