Interpretable Machine Learning Model for Characterizing Magnetic Susceptibility based Biomarkers in First Episode Psychosis
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Altered neurochemicals in deep-brain nuclei, especially dopamine dysfunction, arelinked to psychosis. Quantitative Susceptibility Mapping (QSM) measures brainmagnetic susceptibility changes, including iron concentration, which affects dopaminepathways. This study used machine learning (ML) to analyze MRI data and build aclassifier distinguishing healthy individuals from First-Episode Psychosis (FEP)patients while predicting their response to antipsychotic treatment. A random forestmodel was trained, with the SHAP framework assessing feature importance andinterpretability. Hierarchical clustering identified relationships among features. Themodel achieved performance, with 76.48 ± 10.73% accuracy for classifying FEPpatients (based on R2* values in the nucleus accumbens and amygdala, and QSM inthe thalamus) and 76.43 ± 12.57% accuracy for predicting treatment response (basedon R2* values in the hippocampus, caudate, and putamen, and QSM in the amygdala).MRI-based biomarkers and ML could help tailor personalized treatments for FEPpatients, especially those not responding to standard therapies.
Description
Keywords
Interpretable Machine Learning, Magnetic Resonance Imaging, Biomarkers, Artificial Intelligence for Medical Imaging, Quantitative Susceptibility Mapping