Multi-Level Functional Connectivity Fusion Classification Framework for Brain Disease Diagnosis.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Functional magnetic resonance imaging (fMRI) analysis, particularly functional connectivity (FC), is crucial for brain disease diagnosis.
• Existing methods often overlook high-order functional relationships, focusing primarily on low-order FC features.
• This limitation hinders the comprehensive understanding and classification of complex brain disorders.

Purpose of the Study:

• To propose a novel multi-level functional connectivity (FC) fusion classification framework (MFC) for improved brain disease diagnosis.
• To integrate both low-order and high-order FC patterns for more robust biomarker discovery.
• To enhance the accuracy and generalizability of brain disease classification using advanced machine learning techniques.

Main Methods:

• A deep neural network (DNN) was designed to extract abstract feature representations from low-order and high-order FC patterns.
• The DNN model incorporated unsupervised and supervised learning, including prototype learning for improved feature discrimination.
• An ensemble classifier with a hierarchical stacking strategy was trained using the fused multi-level FC features.

Main Results:

• The proposed MFC model demonstrated robust classification performance across various datasets and experimental conditions.
• The framework effectively combined low-order and high-order FC patterns, yielding significant diagnostic improvements.
• Results confirmed the model's effectiveness and generality, outperforming traditional classification approaches.

Conclusions:

• The multi-level FC fusion framework offers a promising approach for brain disease diagnosis.
• Integrating high-order functional relationships alongside low-order patterns enhances classification accuracy.
• This study advances the application of AI in neuroscience for clinical diagnosis and biomarker identification.