Review of: "Deep Learning Modeling for Prediction of Cognitive Task Related Features from Resting-state fMRI Data"

This study presents a technically sophisticated and methodologically innovative attempt to predict cognitive task performance (Kohs block-design scores) from resting-state fMRI using a modi ed connectivity-based graph convolutional network combined with LSTM (cGCN-LSTM).Strengths include the relatively large sample size (n=615), careful prevention of data leakage (group FC computed from training folds), comparison with a classical baseline (LASSO), and the introduction of a leave-two-out occlusion approach to probe network-level interactions.The nding that cGCN-LSTM retains predictive signal from rs-FC even after removing age-while LASSO performance collapses-suggests that temporal and graph-structured modeling captures higher-order information not represented by simple FCS features.The network-level interpretations (DMN, cerebellum, cross-network connectivity) are neurobiologically plausible and consistent with aging and visuospatial processing literature.