Machine learning based imaging biomarkers in large scale population studies: A neuroimaging perspective

Abstract Neuroimaging data is both very large and complex, as it encompasses many dimensions related to structure, function, and pathology. A growing interest in neuroimaging research has been focused on the use of machine learning methods for the analysis of complex brain imaging data. The success of machine learning approaches depends greatly on the existence of large-scale studies that are complemented with a broad and deep phenotyping. In recent years, studies started to collect imaging data with large samples and mega-analyses started to emerge by pooling thousands of samples together. These studies present opportunities, as well as new challenges, for neuroimaging research, enabling the application of techniques that typically require very large sample sizes for model training and advanced and exploratory analytic techniques that go beyond classical machine learning methods. We present a multifaceted sample of these methods and studies involving machine learning principles applied to large scale population studies.