Identification of suicidality in adolescent major depressive disorder patients using sMRI: A machine learning approach

Background: Suicidal behavior is a major concern for patients who suffer from major depressive disorder (MDD), especially among adolescents and young adults. Machine learning models with the capability of suicide risk identification at an individual level could improve suicide prevention among high-risk patient population. Methods: A cross-sectional assessment was conducted on a sample of 66 adolescents/young adults diagnosed with MDD. The structural T1-weighted MRI scan of each subject was processed using the FreeSurfer software. The classification model was conducted using the Support Vector Machine - Recursive Feature Elimination (SVM-RFE) algorithm to distinguish suicide attempters and patients with suicidal ideation but without attempts. Results: The SVM model was able to correctly identify suicide attempters and patients with suicidal ideation but without attempts with a cross-validated prediction balanced accuracy of 78.59%, the sensitivity was 73.17% and the specificity was 84.0%. The positive predictive value of suicide attempt was 88.24%, and the negative predictive value was 65.63%. Right lateral orbitofrontal thickness, left caudal anterior cingulate thickness, left fusiform thickness, left temporal pole volume, right rostral anterior cingulate volume, left lateral orbitofrontal thickness, left posterior cingulate thickness, right pars orbitalis thickness, right posterior cingulate thickness, and left medial orbitofrontal thickness were the 10 top-ranked classifiers for suicide attempt. Conclusions: The findings indicated that structural MRI data can be useful for the classification of suicide risk. The algorithm developed in current study may lead to identify suicide attempt risk among MDD patients.

Automated summary

The study utilized machine learning models to analyze suicide risk in MDD patients, and found that structural MRI data can be effectively used for suicide risk classification, leading to the development of an algorithm with high predictive accuracy.