A Neuroinformatics Framework for Evaluating Functional Connectivity Metrics in Small-Sample Resting-State fMRI: An...

Functional connectivity (FC) analysis using resting-state fMRI is widely applied to study brain organization in autism spectrum disorder (ASD). However, variability in FC estimation, limited sample sizes, and high-dimensional feature spaces often produce unstable and over-optimistic machine-learning results, highlighting the need for rigorous methodological benchmarking. We propose a neuroinformatics framework for systematic evaluation of multiple FC measures under small-sample rs-fMRI conditions. Data were analyzed across children, adolescents, and adults. Group independent component analysis followed by dual regression extracted subject-specific network time series. Five FC measures-full correlation, partial correlation, bivariate Granger causality, coherence, and mutual information-captured linear, nonlinear, time-, and frequency-domain interactions. To ensure leakage-aware evaluation, feature selection was performed strictly within training folds using leave-one-out cross-validation, and multiple machine-learning classifiers were used as standardized evaluation tools. Performance of FC measures varied across developmental stages. Linear connectivity measures showed more stable behavior in childhood, nonlinear information-theoretic measures were most informative in adolescence, and frequency-domain measures demonstrated stronger performance in adulthood. Unlike studies focused primarily on diagnostic accuracy, this work emphasizes comparative methodological evaluation and explicitly addresses data leakage and overfitting through strict cross-validation and training-only feature selection. FC-based machine-learning outcomes are strongly influenced by developmental stage and methodological choice. The proposed leakage-aware framework provides a practical benchmark for evaluating FC measures in small-sample rs-fMRI Studies and supports more reliable neuroimaging machine-learning research.