Nonsuicidal self-injury prediction with pain-processing neural circuits using interpretable graph neural network.

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Nonsuicidal self-injury (NSSI) is characterized by intentional tissue damage without suicidal intent.
• Individuals with NSSI often display altered pain perception, but the underlying neural circuits are not well understood.
• This study investigates the neural underpinnings of NSSI using advanced computational methods.

Purpose of the Study:

• To predict the risk of NSSI using multimodal neuroimaging data.
• To identify specific brain network connectivity patterns associated with NSSI.
• To explore the role of pain processing abnormalities in NSSI.

Main Methods:

• Utilized resting-state fMRI and diffusion tensor imaging from NSSI patients, healthy controls, and disease controls.
• Constructed pain-related brain networks for each participant.
• Developed an interpretable graph attention networks (GAT) model incorporating demographic and self-reported pain data.

Main Results:

• The GAT model achieved 80% accuracy in distinguishing NSSI patients from healthy controls using imaging data alone.
• Accuracy increased to 88% when self-reported pain scales were included.
• Identified critical connectivity between amygdala-parahippocampus and IFG-insula in NSSI-related pain processing.

Conclusions:

• The GAT model effectively predicts NSSI risk and highlights the importance of limbic and IFG functional integration in NSSI pain processing.
• Altered pain processing appears to be a significant mechanism in NSSI.
• Findings offer insights for developing novel neural modulation interventions for NSSI.