Using multimodal normative modelling to predict clinical outcomes in early psychosis

Individuals at the early stages of psychosis already suffer from neuroanatomical, neurofunctional and neurocognitive alterations.

However, it is not clear how these alterations interact with each other and, more importantly, how this interaction may help predict who will become unwell.

The aim of the proposed study is to uncover previously hidden relationships between brain anatomy, function and cognition that can improve our understanding of the first signs of the illness and help predict illness trajectory at the individual level.

This will be achieved by triangulating, for the first time, data fusion, machine learning and a unique longitudinal dataset of individuals at the prodromal stage or with a recent first episode of psychosis.

This combination of methods will allow me to address three main objectives: 1) identify previously unknown relationships between anatomy, function and cognition in healthy individuals; 2) develop a model that captures the normative pattern of these hidden relationships and determine by how much each patient deviates from this pattern; 3) use these deviations to predict each patient’s longitudinal clinical outcomes.

I will also develop an open access tool that allows non-expert researchers to perform data fusion on their own data. Keywords: psychosis, data fusion, machine learning, outcome prediction.