Predicting PTSD severity using longitudinal magnetoencephalography with a multi-step learning framework.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging

Background:

• Post-traumatic stress disorder (PTSD) is a debilitating mental health condition.
• Accurate prediction of PTSD severity is crucial for effective treatment.
• Magnetoencephalography (MEG) offers a non-invasive method to study brain activity.

Purpose of the Study:

• To evaluate the effectiveness of incorporating temporal information from longitudinal MEG data in predicting PTSD severity.
• To identify neural oscillatory frequencies and functional connections predictive of PTSD severity.
• To develop and validate an informatics approach for longitudinal brain state tracking.

Main Methods:

• Utilized a two-step informatics approach: pre-learn feature selection (CV-SVR-rRF-FS) and deep learning (LSTM-RNN).
• Analyzed longitudinal MEG functional connectome data from two timepoints (Phase I and II).
• Correlated functional connections with PTSD severity (PTSD CheckList score) at Phase II.

Main Results:

• Pre-learn feature selection identified key functional connections from Phase I MEG data associated with Phase II PTSD severity.
• Longitudinal models incorporating both Phase I and II MEG data demonstrated enhanced predictive performance compared to single timepoint models.
• Alpha and high gamma frequency bands showed improved prediction, with lower frequencies outperforming higher frequencies.

Conclusions:

• Temporal information from longitudinal MEG functional connectome data significantly improves PTSD severity prediction.
• Identified specific neural oscillatory signatures beneficial for outcome prediction in PTSD.
• The developed informatics framework is applicable to other mental health conditions for longitudinal tracking and outcome prediction.