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

Yun Wang, Chenxiao Xu, Ji-Hwan Park, Seonjoo Lee, Yaakov Stern, Jong Hun Kim, Shinjae Yoo, Hyoung Seop Kim, Jiook Cha
2018 bioRxiv   pre-print
Accurate, reliable prediction of risk for Alzheimer's disease (AD) is essential for early, disease-modifying therapeutics. Multimodal MRI, such as structural and diffusion MRI, may contain multi-dimensional information neurodegenerative processes in AD. Here we tested the utility of structural MRI and diffusion MRI as imaging markers of AD using high-throughput brain phenotyping including morphometry and white-matter structural connectome (whole-brain tractography), and machine learning
more » ... s for classification. We used a retrospective cohort collected at a dementia clinic (Ilsan Dementia Cohort; N=211; 110 AD, 64 mild cognitive impairment [MCI], and 37 subjective memory complaints [SMC]). Multi-modal MRI was collected (T1, T2-FLAIR, and diffusion MRI) and was used for morphometry, structural connectome, and white matter hyperintensity (WHM) segmentation. Our machine learning model trained on the large-scale brain phenotypes (n=34,646) classified AD, MCI, and SMC with unprecedented accuracy (AD/SMC: 97% accuracy, MCI/SMC: 83% accuracy; AD/MCI: 98% accuracy) with strict iterative nested ten-fold cross-validation. Model comparison revealed that white-matter structural connectome was the primary contributor compared with conventional volumetric features (e.g., WHM or hippocampal volume). This study indicates promising utility of multimodal MRI, particularly structural connectome, combined with high-throughput brain phenotyping and machine learning analytics to extract salient features enabling accurate diagnostic prediction.
doi:10.1101/255141 fatcat:cfvsbvqmxnf27keulvxyt5d6pm