Machine learning applied to brain imaging data

Abstract

Multivariate data‐driven analysis may identify subtle anomalies in retinal and brain structure in the absence of visible pathology. The modelling of healthy ageing critically requires methodology that can detect subtle changes in this process and that can also be applied to pathological conditions. In the last few years multiple machine learning models have been proposed that learn age patterns from magnetic resonance images.In recent years, classification approaches using imaging data have shown that multimodal classification methods may perform better than the use of a single imaging modality for the diagnosis of neurodegenerative conditions of the eye and the brain. The currently used clinical approach often emphasizes the use of qualitative structural, molecular and/or functional imaging data for clinical diagnosis. Based on the hypothesis that classification of isolated imaging modalities is not predictive of their respective value in combined approaches, we investigate the potential of multimodal quantitative integration approaches in neurodevelopmental and neurodegenerative conditions. We also review imaging evidence based on machine learning reflecting critical interactions between ageing, systemic disorders (diabetes) and neurodegenerative diseases.