Skip to navigation Skip to main content Skip to footer

Approved research

Studying mental health via research domain criteria, neuroimaging and convolutional neural networks

Principal Investigator: Dr Kerstin Ritter
Approved Research ID: 33073
Approval date: March 27th 2018

Lay summary

We want to employ state-of-the-art convolutional neural networks (CNNs) for characterizing mental health based on structural and functional MRI data. Main research questions: How well do CNN-derived MRI representations characterize mental health? And how can we visualize and interpret these CNN-derived MRI representations? - Proposed research supports characterization of individual subjects with respect to mental health - Link of imaging data and behavioral measures - Improvement of diagnostics for derived mental illnesses Based on UK Biobank data, we aim to establish a relationship between mental health measures on the one hand and MRI data on the other hand. To do so, we will employ state-of-the-art machine learning algorithms, in particular convolutional neural networks, which are very powerful in detecting hidden connections in data. Novel visualization techniques will then be used to understand what image features have been used and to what extent they fit into respective models of mental health. Subset of all currently available patients with MRI data and additional data as it becomes available (especially GP data)