Approved research

Using deep learning technology to make individualised inferences in brain-based disorders

Lay summary

Brain-based disorders, including psychiatric and neurological illnesses, represent 10.4% of global disease. At present, we do not have objective tools for detecting these disorders, monitoring their progression over time and optimising treatment. In this project, which will last 24 months, we will use an innovative approach that capitalises on the latest developments in an emerging area of artificial intelligence known as deep learning technology. Inspired by how the human brain processes information, this technology allows detection of complex and distributed patterns in the data that are difficult to capture using existing approaches. First, we will assemble a very large dataset comprising neuroimaging data from eighteen thousand plus disease-free individuals and two thousand plus patients with psychosis. We will then use deep learning technology to develop a model of the disease-free brain across the different ages and genders. During this phase we will use clinically-relevant data fields to distinguish people with a psychiatric or neurological disorder from disease-free individuals, since only the latter will be included in the normative model. Finally, we will illustrate how this model can be used to detect neuroanatomical alterations and inform clinical assessment in individual patients. In this second phase, we plan to use the MRI of the individuals excluded during the first phase to test the normative model. In addition, we will model non-clinical data fields as possible confounding variables when building and testing the normative model in order to increase the generalizability of the model. This project will lead to the development of a practical and flexible web-based tool for measuring neuroanatomical alterations in any brain-based disorders. Indeed, we are developing an outlier detection model that should test for differences in brain structure at the level of the individual. This model could be applied to any brain-based disorder (either psychiatric or neurological), and therefore we are not focusing on any specific disorders. This tool could help clinicians assess the presence of a disease, monitor its progression over time and optimise treatment in individual patients.