查看文章

Background

The dominant case-control paradigm, for which a group of healthy individuals is compared to a group of patients, occludes individual-level differences contributing to disorder or disease. Therefore, we developed normative modelling which is inspired by approaches used in paediatrics for which a child’s height is charted against age. Here, I present novel results on mapping the heterogeneity of mental disorders against population norms which were estimated for imaging and cognitive data using models of varying complexity.

Methods

We employed machine learning algorithms within the normative modelling framework. These were based on different implementations of deep neural networks and Bayesian regressions and trained on a large population. The estimated cognitive and imaging-based models were applied to clinical cohorts including patients with Schizophrenia, Bipolar Disorder and …