Machine learning classifier using abnormal brain network topological metrics in major depressive disorder.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Network Science

Background:

• Resting-state functional brain networks are crucial in understanding brain diseases.
• The utility of abnormal network metrics for machine learning-based disease classification remains unclear.

Purpose of the Study:

• To investigate if abnormal resting-state functional brain network metrics can be used for classifying major depressive disorder (MDD).
• To identify specific brain regions with abnormal network topology in MDD patients.

Main Methods:

• Constructed resting-state functional brain networks from 90 brain regions for 38 MDD patients and 28 healthy controls.
• Calculated graph theory-based nodal metrics and used nonparametric permutation tests for group comparisons.
• Employed six machine learning algorithms, with statistically significant metrics as features, to classify MDD.

Main Results:

• Significant abnormalities in nodal centralities were observed in the limbic system, basal ganglia, medial temporal, and prefrontal regions.
• Support vector machine and neural network algorithms achieved the highest classification accuracy (79.27% and 78.22%) using 28 features.
• A strong positive correlation was found between feature importance and statistical significance of network metrics.

Conclusions:

• Major depressive disorder is associated with altered functional brain network topology.
• Statistically significant nodal metrics are effective features for machine learning-based classification of MDD.