SCALABLE FUSED LASSO SVM FOR CONNECTOME-BASED DISEASE PREDICTION.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Resting-state functional magnetic resonance imaging (fMRI) generates functional connectomes (FCs) to study brain connectivity.
• High-dimensional FC data poses challenges for machine learning models aiming to distinguish patient groups from healthy controls.
• Existing methods often use feature selection that ignores the inherent spatial structure of FC data.

Purpose of the Study:

• To develop a novel machine learning approach that incorporates the 6-dimensional spatial structure of functional connectomes.
• To improve the accuracy and neuroscientific interpretability of machine-based classification using resting-state fMRI data.

Main Methods:

• Proposed a fused Lasso regularized support vector machine (SVM) to analyze the 6-D structure of FCs.
• Introduced a scalable alternating direction method algorithm to solve the complex optimization problem associated with the fused Lasso SVM.
• Applied the method to real resting-state fMRI scans.

Main Results:

• The proposed method effectively utilizes the spatial information within functional connectomes.
• Achieved more neuroscientifically informative results compared to traditional feature selection techniques.
• Demonstrated the scalability and effectiveness of the novel optimization algorithm.

Conclusions:

• The fused Lasso SVM with a novel optimization algorithm offers a powerful tool for analyzing high-dimensional FC data.
• This approach enhances the ability to distinguish between patient groups and healthy controls using resting-state fMRI.
• The method provides more meaningful insights into brain connectivity patterns relevant to neurological conditions.