Determining four confounding factors in individual cognitive traits prediction with functional connectivity: an...

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Resting-state functional connectivity (RSFC) is utilized for predicting individual traits.
• Confounding factors can influence the accuracy of brain-behavior relationship predictions.

Purpose of the Study:

• Investigate the impact of time series length, functional connectivity (FC) type, brain parcellation, and target variance on RSFC-based prediction.
• Identify optimal parameters for robust and efficient individualized trait prediction.

Main Methods:

• Utilized Human Connectome Project data (1,206 subjects).
• Assessed prediction performance for fluid intelligence, working memory, and picture vocabulary using partial least square regression.
• Compared various settings for time series length, FC calculation methods (Pearson, Spearman, Partial correlation, Mutual Information, Coherence), and brain parcellations (ICA100/200).

Main Results:

• Optimal time series length (300 points) and ICA parcellations (ICA100/200) improved prediction performance and efficiency.
• Pearson, Spearman, and Partial correlation yielded higher accuracy and lower computational cost compared to mutual information and coherence for FC calculation.
• Cognitive traits with greater inter-subject variance were more predictable.
• Increased scan duration improved prediction, partly due to enhanced RSFC test-retest reliability.

Conclusions:

• Determining optimal parameters for time series length, FC type, and brain parcellation is crucial for accurate RSFC-based prediction.
• These findings can guide the standardization of RSFC-based prediction pipelines for improved reliability and comparability.