Minimal Fusion Gate (MFG) Model for Functional MRI Classification

Deep learning methods have advanced the classification of functional MRI (fMRI) time series, yet complex architectures such as transformers and graph-based networks often incur high computational costs and reduced interpretability. In this paper, we propose a Minimal Fusion Gate (MFG) model designed to efficiently handle the high dimensionality of BOLD signals. Our approach enhances a simple MLP framework with a gating mechanism that adaptively captures temporal dynamics without adding extra learnable parameters. Experimental results in gender classification demonstrate that integrating gating yields higher accuracy while preserving low computational overhead and simplicity. We suggest that the Minimal Fusion Gate model offers a balanced trade-off between complexity and performance for large-scale fMRI classification.