A Tensor-Based Framework for rs-fMRI Classification and Functional Connectivity Construction.

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Machine learning analysis of Resting-State Functional Magnetic Resonance Imaging (rs-fMRI) is crucial for understanding brain disorders like Alzheimer's disease.
• Current methods often involve calculating individual functional connectivity (FC) matrices, which can be inefficient and limit analytical depth.
• Identifying common disease-related patterns and discriminating between patients and controls are key challenges in rs-fMRI research.

Purpose of the Study:

• To develop a novel tensor framework for analyzing rs-fMRI data that bypasses the need for individual FC matrix construction.
• To improve the accuracy of brain disorder classification using rs-fMRI.
• To identify early-stage connectivity patterns associated with Alzheimer's disease.

Main Methods:

• A novel tensor framework was developed to obtain a general FC matrix without constructing individual matrices per sample.
• The framework reduces data dimensionality and creates a novel discriminant function that operates directly on samples.
• The method avoids data vectorization and incorporates test data into training without prior label information.

Main Results:

• The proposed tensor framework significantly boosts fMRI classification performance.
• The method successfully identified novel connectivity patterns in early-stage Alzheimer's disease.
• Experiments on the ADNI dataset validated the framework's effectiveness.

Conclusions:

• The novel tensor framework offers a more efficient and effective approach to analyzing rs-fMRI data for disease classification.
• This method enhances the understanding of brain connectivity in neurodegenerative diseases like Alzheimer's.
• The framework's ability to reveal early-stage patterns holds promise for improved diagnostic capabilities.