Convolutional Neural Network With Graphical Lasso to Extract Sparse Topological Features for Brain Disease...

Area of Science:

• Neuroscience
• Machine Learning
• Medical Image Analysis

Background:

• Functional connectivity analysis offers network-level insights into brain mechanisms and serves as a biomarker for brain disease classification.
• Machine learning, particularly convolutional neural networks (CNNs), is increasingly used for functional connectivity classification due to their ability to extract topological features.
• Conventional CNN methods overlook sparse connectivity patterns (SCPs), potentially leading to feature redundancy and limiting performance.

Purpose of the Study:

• To propose a novel CNN-based model incorporating graphical Lasso (CNNGLasso) for extracting sparse topological features.
• To address the limitations of conventional CNNs by considering SCPs for improved brain disease classification.
• To enhance the performance and generalization of brain network analysis in disease classification.

Main Methods:

• Development of a novel graphical Lasso model to identify group-level SCPs.
• Utilizing SCPs to guide the extraction of topological features within brain networks.
• Application of the extracted sparse topological features for classifying patients from normal controls using a CNN framework.

Main Results:

• The CNNGLasso model demonstrated superior performance compared to existing methods on the ABIDE dataset.
• The model successfully extracted sparse topological features, improving classification accuracy.
• Identified abnormal brain regions aligned with previous research findings, validating the model's clinical relevance.

Conclusions:

• The proposed CNNGLasso model effectively extracts sparse topological features for brain disease classification.
• Incorporating SCPs enhances the performance and generalization of CNN-based brain network analysis.
• The CNNGLasso model shows significant promise for clinical applications in diagnosing brain disorders.