General psychopathology factor (p-factor) prediction using resting-state functional connectivity and a...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• The general psychopathology factor (p-factor) aggregates shared variance across mental disorders.
• Adolescent Brain Cognitive Development (ABCD) Study provides large-scale neuroimaging data for investigating psychopathology.
• Scanner heterogeneity in multi-site fMRI studies poses challenges for robust prediction models.

Purpose of the Study:

• To develop and validate a scanner-generalization neural network (SGNN) for predicting the individual p-factor from resting-state functional connectivity (RSFC).
• To overcome scanner effects in fMRI data from the ABCD Study.
• To identify functional networks (FNs) crucial for p-factor prediction and scanner generalization.

Main Methods:

• Utilized fMRI data from 6905 adolescents across 18 sites in the ABCD Study.
• Estimated the p-factor using Child Behavior Checklist (CBCL) scores via exploratory factor analysis.
• Developed a novel SGNN model to reduce scanner effects and predict the p-factor, evaluated using leave-one/two-site-out cross-validation (LOSOCV/LTSOCV).

Main Results:

• The SGNN model achieved significantly higher prediction accuracy (Pearson's correlation coefficients) compared to kernel ridge regression (KRR) in both LOSOCV and LTSOCV.
• Identified default-mode and dorsal attention FNs as important for p-factor prediction.
• The intra-visual FN was found to be critical for achieving scanner generalization.

Conclusions:

• The proposed SGNN model effectively predicts the adolescent p-factor from RSFC while mitigating scanner-related confounds.
• This approach enhances the reliability of fMRI-based prediction of general psychopathology in large, multi-site studies.
• The findings highlight specific functional brain networks relevant to general psychopathology and scanner-invariant prediction.