Multiview Feature Learning With Multiatlas-Based Functional Connectivity Networks for MCI Diagnosis

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Diagnostics

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) derived functional connectivity (FC) networks show potential for diagnosing Alzheimer's disease and mild cognitive impairment (MCI).
• Traditional FC network construction relies on predefined brain atlases, potentially overlooking subject-specific anatomical variations and limiting diagnostic sensitivity.
• Single-atlas approaches may introduce bias, hindering the detection of subtle differences between healthy individuals and those with MCI.

Purpose of the Study:

• To propose a novel multiview feature learning method using multi-atlas-based FC networks to enhance MCI diagnosis.
• To address the limitations of single-atlas methods by incorporating subject-specific brain parcellations.
• To improve the accuracy and reliability of brain connectome-based individualized diagnosis for neurological disorders.

Main Methods:

• A three-step transformation process generates multiple subject-specific atlases from a standard template.
• Multiple FC networks are constructed using selected atlas exemplars, creating diverse 'views' of brain connectivity for each subject.
• A multitask learning algorithm performs joint feature selection across these multiple FC networks, feeding selected features into a support vector machine classifier.

Main Results:

• The proposed multi-atlas FC network method significantly improved MCI classification accuracy compared to traditional single-atlas approaches.
• The integration of subject-specific parcellations and multiview learning effectively captured complex brain connectivity differences.
• Experimental comparisons validated the superior performance of the multi-atlas method in distinguishing MCI patients from controls.

Conclusions:

• Multi-atlas based functional connectivity network analysis offers a promising avenue for the individualized diagnosis of mild cognitive impairment.
• This approach mitigates the bias associated with single-atlas methods, leading to more robust diagnostic performance.
• The findings highlight the potential of advanced neuroimaging analysis techniques for early and accurate detection of neurodegenerative diseases.