An explainable multimodal artificial intelligence model for classifying suicide attempters with borderline...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Borderline Personality Disorder (BPD) is characterized by emotional dysregulation.
• Suicide attempts are prevalent in BPD, with significant mortality rates.
• Current tools for identifying suicide risk factors in BPD are insufficient.

Purpose of the Study:

• To investigate the potential of Artificial Intelligence (AI) in identifying suicide attempts among individuals with BPD.
• To develop and validate a Machine Learning (ML) model for predicting suicide attempts in BPD.
• To explore multimodal data signatures for differentiating suicide attempters within the BPD population.

Main Methods:

• Development of DRAMA-BPD (Detecting Retrospective suicide Attempts with Machine learning Approaches in Borderline Personality Disorder), an explainable, multimodal ML model.
• Training the ensemble classifier on sociodemographic, clinical, and MRI data from 104 individuals with BPD.
• Utilizing SHapley Additive exPlanations (SHAP) for model interpretability and feature identification.

Main Results:

• DRAMA-BPD achieved a balanced accuracy of 0.68, sensitivity of 0.58, specificity of 0.77, and AUC of 0.68.
• SHAP analysis highlighted cortical volumes and thickness from T1-weighted MRI scans as key predictors.
• The Symptoms Checklist 90 Revised (SCL-90-R) also emerged as a significant contributor to the model's classification.

Conclusions:

• Preliminary evidence suggests that a multimodal AI approach can differentiate suicide attempts in individuals with BPD.
• The DRAMA-BPD model shows promise for prospective cohort validation and clinical applications in suicide prevention.
• Explainable AI methods can identify neuroimaging and clinical factors associated with suicide attempts in BPD.