Machine-Learning Classifier for Patients with Major Depressive Disorder: Multifeature Approach Based on a High-Order...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• High-order functional connectivity networks capture dynamic brain activity crucial for disease classification.
• Traditional methods face challenges in neurological interpretability and computational cost.
• Existing approaches often reduce dimensionality, losing valuable network information.

Purpose of the Study:

• To develop a novel method for generating high-order minimum spanning tree functional connectivity networks.
• To enhance neurological significance and reduce computational expense in network analysis.
• To improve the classification accuracy of brain diseases using advanced network features.

Main Methods:

• Generation of high-order minimum spanning tree functional connectivity networks.
• Application of frequent subgraph mining for feature extraction.
• Integration of quantifiable local network features and multikernel learning for classification.

Main Results:

• The proposed method significantly increases neurological significance and reduces computational load.
• Frequent subgraph mining effectively captures discriminative subnetworks as features.
• Multikernel learning achieved a high classification accuracy of 97.54% for major depressive disorder.

Conclusions:

• The novel high-order minimum spanning tree functional connectivity network method offers a neurologically interpretable and computationally efficient approach.
• This method enhances the accuracy of brain disease classification, demonstrated by its effectiveness in identifying major depressive disorder.
• The findings suggest a promising direction for analyzing complex brain network data in clinical neuroscience.