The Contribution of Explainable Machine Learning Algorithms Using ROI-based Brain Surface Morphology Parameters in...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Differentiating early-onset schizophrenia (EOS) from early-onset bipolar disorder (EBD) is clinically challenging.
• Distinct neurobiological underpinnings may exist between EOS and EBD.

Purpose of the Study:

• To develop and validate machine learning (ML) models for differentiating EOS from EBD.
• To utilize surface-based morphometry and brain volume measurements for diagnostic classification.

Main Methods:

• High-resolution T1-weighted MRI scans were analyzed for cortical thickness, gyrification, sulcal depth, fractal dimension, and brain volumes.
• Machine learning classifiers, including Adaptive Boosting (AdaBoost), K-nearest neighbors (KNN), and Support Vector Machine (SVM), were applied to feature subsets and combined datasets.
• The SHapley Additive exPlanations (SHAP) technique was used for feature interpretability.

Main Results:

• The AdaBoost algorithm achieved the highest accuracy of 82.75% using features from the Destrieux atlas.
• Specific ML models showed high performance for individual feature subsets: KNN for fractal dimension (79.31%), SVM for sulcal depth, and AdaBoost for gyrification index.
• The KNN algorithm demonstrated the highest accuracy (79.31%) on the entire dataset from the Desikan-Killiany-Tourville atlas.

Conclusions:

• Machine learning models effectively differentiate between EOS and EBD using surface-based morphometry.
• This approach shows promise for improving the differential diagnosis of these early-onset psychiatric disorders.
• Future multicenter studies are recommended for validating these findings.