Brain connectivity and novel network measures for Alzheimer's disease classification

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biomedical Engineering

Background:

• Alzheimer's disease (AD) diagnosis relies on clinical assessments and neuroimaging, but distinguishing early-stage AD and mild cognitive impairment (MCI) remains challenging.
• Anatomical brain connectivity, reflecting structural connections between brain regions, is altered in AD and may offer valuable diagnostic biomarkers.
• Existing connectivity measures may not fully capture the complex network alterations in AD.

Purpose of the Study:

• To evaluate novel and existing anatomical connectivity measures for their efficacy in discriminating Alzheimer's disease (AD) patients from healthy controls.
• To explore the utility of different network-based measures derived from diffusion MRI for identifying individuals with early or late mild cognitive impairment (MCI) or AD.
• To assess the diagnostic performance of these connectivity measures using machine learning classification techniques.

Main Methods:

• Diffusion-weighted magnetic resonance imaging (dMRI) data from 200 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) were analyzed.
• Connectivity matrices were constructed using whole-brain tractography and a novel flow-based connectivity measure on a 3D lattice.
• Support vector machines (SVM) with 10-fold cross-validation were employed to classify subjects into diagnostic groups (healthy, MCI, AD).

Main Results:

• Several novel and existing anatomical connectivity measures demonstrated significant ability to discriminate between Alzheimer's disease (AD) patients and healthy controls.
• The classification accuracy, sensitivity, and specificity varied across different combinations of connectivity features, highlighting the importance of feature selection.
• Feature ranking analysis identified key network properties that are most informative for distinguishing between different stages of cognitive impairment and AD.

Conclusions:

• Novel flow-based and tractography-derived anatomical connectivity measures show potential as sensitive biomarkers for Alzheimer's disease (AD) detection.
• Machine learning classifiers utilizing these connectivity features can effectively differentiate individuals with AD and MCI from healthy controls.
• These findings suggest that advanced neuroimaging-based connectivity analysis could improve the early diagnosis and management of Alzheimer's disease.