Related Experiment Video
Updated: Jul 19, 2025
Diffusion Tensor Magnetic Resonance Imaging in the Analysis of Neurodegenerative Diseases
Published on: July 28, 2013
Can diffusion tensor imaging have a diagnostic utility to differentiate early-onset forms of bipolar disorder and
Yesim Saglam1, Ahmet Oz2, Gokcen Yildiz2
1Department of Child and Adolescent Psychiatry, University of Health Sciences, Bakirkoy Prof Dr Mazhar Osman Research and Training Hospital for Psychiatry, Neurology and Neurosurgery, Istanbul, Turkey.
Background/Aim:
Accurate diagnosis of early-onset psychotic disorders is crucial to improve clinical outcomes. This study aimed to differentiate patients with early-onset schizophrenia (EOS) from early-onset bipolar disorder (EBD) with machine learning (ML) algorithms using white matter tracts (WMT).
Method:
Diffusion tensor imaging was obtained from adolescents with either EOS (n = 43) or EBD (n = 32). Global probabilistic tractography using an automated tract-based TRACULA software was performed to analyze the fractional anisotropy (FA) of forty-two WMT. The nested cross-validation was performed in feature selection and model construction. EXtreme Gradient Boosting (XGBoost) was applied to select the features that can give the best performance in the ML model. The interpretability of the model was explored with the SHApley Additive exPlanations (SHAP).
Findings:
The XGBoost algorithm identified nine out of the 42 major WMTs with significant predictive power. Among ML models, Support Vector Machine-Linear showed the best performance. Higher SHAP values of left acoustic radiation, bilateral anterior thalamic radiation, and the corpus callosum were associated with a higher likelihood of EOS.
Conclusions:
Our findings suggested that ML models based on the FA values of major WMT reconstructed by global probabilistic tractography can unveil hidden microstructural aberrations to distinguish EOS from EBD.
