Abstract
The human brain is characterized by spatially distinguishable cytoarchitecture, structure, function, connectivity, and morphology, allowing its parcellation into distinct regions. Delineating structurally and functionally homogeneous brain regions through parcellation is crucial for advancing our understanding of brain organization and function. Functional parcellation leverages resting-state or task-based fMRI data to map regions with coherent activity or connectivity patterns to uncover brain network architecture changes across the lifespan and in disease. Recent advances in artificial intelligence (AI) have transformed this field by enabling data-driven and individualized mapping of functional brain organization. This review covers current methodologies across supervised, unsupervised, and self-supervised learning frameworks in functional parcellation using resting-state functional MRI (rs-fMRI), highlighting their applications in spatial and temporal feature extraction as well as individual parcellations. We compared traditional approaches such as independent component analysis with AI-based methods such as graph neural networks, convolutional neural networks, and transformer networks, emphasizing their distinctive methodological basis and performance. We elaborated validation strategies including test-retest reproducibility, functional homogeneity, alignment with task-based fMRI or electrophysiology, and cross-modality validation. We also discussed limitations of AI-based approaches, such as data requirements, generalizability, and interpretability. Furthermore, we proposed future directions including multimodal integration, foundation models, and explainable AI. Collectively, this review outlines the current strategies of functional parcellation using AI, ultimately supporting its usage for understanding brain organization across the lifespan and in disease.