Processing fMRI Brain Signals Using Latents from Natural Image Autoencoders

Modeling long-range spatiotemporal dynamics in functional Magnetic Resonance Imaging (fMRI) remains a key challenge due to the high dimensionality of the four-dimensional signals. Prior voxel-based models, although demonstrating excellent performance and interpretation capabilities, are constrained by prohibitive memory demands and thus can only capture limited temporal windows. To address this, we propose TABLeT (Two-dimensionally Autoencoded Brain Latent Transformer), a novel approach that tokenizes fMRI volumes using a pre-trained 2D natural image autoencoder. Each 3D fMRI volume is compressed into a compact set of continuous tokens, enabling efficient long-sequence modeling with a simple transformer encoder. Across large-scale benchmarks including the Human Connectome Project (HCP) and ADHD-200 datasets, TABLeT consistently outperforms existing models in multiple tasks, while demonstrating substantial gains in computational and memory efficiency over the state-of-the-art voxel-based method. Our findings highlight a new paradigm for scalable spatiotemporal modeling of brain activity.