کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Dynamic Simplicial Coherence Networks for Resting-State fMRI Gender Classification

نویسندگان :

Mansooreh Pakravan¹

1- تربیت

کلمات کلیدی :

resting-state fMRI،gender classification،functional connectivity،simplicial complex،higher order networks،deep learning

چکیده :

Gender differences in functional brain organization are consistently reported in resting-state functional magnetic resonance imaging (rs-fMRI), yet the structure of these differences at the level of higher-order network interactions remains only partially understood. While recent deep learning models can directly decode gender from multivariate fMRI time series, most existing architectures treat brain regions as independent channels and largely ignore higher-order connectivity motifs beyond pairwise correlations. In this work, we propose the Dynamic Simplicial Coherence Network, a neural architecture that explicitly integrates temporal dynamics with simplicial representations of functional connectivity. The model takes only regional rs-fMRI time series as input and internally constructs a dynamic correlation graph. Edgelevel and triangle-level interactions are summarized through a coherence rule inspired by simplicial complex theory, embedded via dedicated multilayer perceptrons, and fused with temporal convolutional features to form a joint representation of regional dynamics and higher-order network structure. We evaluate the proposed model on the Human Connectome Project rs-fMRI dataset using the Brainnetome parcellation and a subject-level five-fold cross-validation protocol. The proposed method achieves a subject-level accuracy of 0.926 ± 0.013, F1-score of 0.944 ± 0.009, and balanced accuracy of 0.909 ± 0.021, outperforming a strong temporal baseline. These results demonstrate that explicit modeling of simplicial structures provides complementary and discriminative information beyond purely temporal filters.