Multimodal MRI classification in vascular mild cognitive impairment.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Vascular mild cognitive impairment (VMCI) presents diagnostic challenges.
• Multimodal magnetic resonance imaging (MRI), integrating diffusion tensor imaging (DTI) and brain morphometry, shows promise for VMCI assessment.
• Machine learning (ML) techniques offer advanced analytical capabilities for complex neuroimaging data.

Purpose of the Study:

• To implement and compare ML techniques for multimodal classification of VMCI patients versus healthy subjects.
• To evaluate ML for discriminating cognitive performance within the VMCI group.
• To assess the added value of multimodal neuroimaging over monomodal approaches.

Main Methods:

• Constructed feature vectors from cortical parcellation, volumetric segmentation, and DTI statistical metrics.
• Employed Sequential Minimal Optimization (SMO) and Functional Tree classifiers.
• Utilized feature selection and 10-fold cross-validation for robust performance evaluation.

Main Results:

• Monomodal classification achieved satisfactory performance.
• Multimodal classification significantly improved discrimination between VMCI patients with low cognitive performance and healthy subjects, increasing sensitivity by up to 10% with unchanged specificity.
• ML techniques demonstrated effectiveness in neuroimaging-based disease state discrimination.

Conclusions:

• Multimodal MRI data, analyzed with ML, enhances the diagnostic accuracy for VMCI.
• ML-driven analysis of neuroimaging data is valuable for identifying VMCI and predicting cognitive decline.
• The study confirms the utility of ML in discriminating diseased states using multimodal neuroimaging.