Diagnosis and Prognosis Using Machine Learning Trained on Brain Morphometry and White Matter Connectomes

Accurate, reliable prediction of risk for Alzheimers disease (AD) is essential for early, disease-modifying therapeutics. Multimodal MRI, such as structural and diffusion MRI, is likely to contain complementary information of neurodegenerative processes in AD. Here we tested the utility of commonly available multimodal MRI (T1-weighted structure and diffusion MRI), combined with high-throughput brain phenotyping--morphometry and connectomics--and machine learning, as a diagnostic tool for AD. We used, firstly, a clinical cohort at a dementia clinic (study 1: Ilsan Dementia Cohort; N=211; 110 AD, 64 mild cognitive impairment [MCI], and 37 subjective memory complaints [SMC]) to test and validate the diagnostic models; and, secondly, Alzheimers Disease Neuroimaging Initiative (ADNI)-2 (study 2) to test the generalizability of the approach and the prognostic models with longitudinal follow up data. Our machine learning models trained on the morphometric and connectome estimates (number of features=34,646) showed optimal classification accuracy (AD/SMC: 97% accuracy, MCI/SMC: 83% accuracy; AD/MCI: 97% accuracy) with iterative nested cross-validation in a single-site study, outperforming the benchmark model (FLAIR-based white matter hyperintensity volumes). In a generalizability study using ADNI-2, the combined connectome and morphometry model showed similar or superior accuracies (AD/HC: 96%; MCI/HC: 70%; AD/MCI: 75% accuracy) as CSF biomarker model (t-tau, p-tau, and Amyloid {beta}, and ratios). We also predicted MCI to AD progression with 69% accuracy, compared with the 70% accuracy using CSF biomarker model. The optimal classification accuracy in a single-site dataset and the reproduced results in multi-site dataset show the feasibility of the high-throughput imaging analysis of multimodal MRI and data-driven machine learning for predictive modeling in AD.