Structural connectome and machine learning for Alzheimer’s disease detection

Background Alzheimer's disease (AD) is hypothesised as a disconnection syndrome where degenerating white matter fibre bundles leads to deterioration in the integration and communication between brain regions. Connectomics allows study of vivo connectivity elucidates how changes network. Although some studies have shown evidence alteration structural AD cognitively normal individuals (CN), large proportion research focused on functional connectomics AD. Emerging explored use machine learning (ML) distinguish differences with promising results. This project aims identify connectome using novel image processing techniques generate network metrics utilise ML classify CN. Method We examined data from 143 age-matched subjects (AD mean: 71.1 ± 2.79 CN 71.09 2.72) Disease Neuroimaging Initiative cohort 2 (ADNI2). used magnetic resonance images (T1-weighted diffusion-weighted images) combined latest state-of-the-art imaging tools connectomes. Relevant were measure compare connectivity, while algorithms Result found significant clustering coefficient (p < 0.05), normalised degree variance 0.0001), hierarchical complexity 0.005) rich club 0.0001) (table 1). also established compared classification performances within our model. Random forest yielded sensitivity 53.06% specificity 82.98% 2) for imbalanced (AD=49, CN=94). On balanced (AD=CN=49), model was 81.63% specific 69.39% sensitive 3) detecting Conclusion The results show feasibility analysis combining detection. While previous achieved data, we reported both models. As real-world are more likely be imbalanced, lower performance models suggests that further improvement needed clinical implementation.