A deep learning approach for mental health quality prediction using functional network connectivity and assessment...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Current mental health assessments rely on questionnaires, lacking direct biological insights.
• Resting-state functional MRI (rs-fMRI) offers potential for identifying brain connectivity biomarkers.
• Individualized prediction models can bridge biological and behavioral aspects of brain health.

Purpose of the Study:

• To develop and validate a deep learning model for estimating mental health quality using rs-fMRI data.
• To identify specific static functional network connectivity (sFNC) patterns associated with different mental health categories.
• To explore the potential of advanced neuroimaging and AI for mental health evaluation and personalized interventions.

Main Methods:

• Utilized static functional network connectivity (sFNC) derived from rs-fMRI as input for a deep learning model.
• Employed guided gradient class activation maps (guided Grad-CAM) to visualize discriminative sFNC patterns.
• Validated the model on the UK Biobank dataset, comparing its performance against four alternative models.

Main Results:

• The deep learning model achieved classification accuracies of 76% (excellent), 78% (good), 88% (fair), and 98% (poor) for mental health categories.
• Demonstrated superior performance over alternative models, with accuracy improvements ranging from 4-18%.
• Identified distinct sFNC patterns: cerebellar-subcortical for excellent mental health and sensorimotor/visual domains for poor mental health.

Conclusions:

• The integration of rs-fMRI and deep learning provides a promising framework for objective mental health assessment.
• Distinct neural connectivity patterns correlate with varying levels of mental health quality.
• This approach holds potential for guiding personalized mental health interventions and monitoring treatment efficacy.