The applicant will work as part of a collaborative team, led by Dr Sarah Morgan at the School of Biomedical Engineering and Imaging Sciences. The School is a world leading centre of expertise in AI for healthcare, providing an outstanding environment in which to develop machine learning tools and engage with an interdisciplinary community of researchers with an interest in AI for healthcare. The post holder will have opportunities to learn from colleagues across the department through regular seminars and tutorials, and benefit from close links to industry through the London Institute for Healthcare Engineering. They will also collaborate closely with researchers and clinicians at the King’s IoPPN, which is a world leading centre for Psychiatric research. 

 

This is a 2 year postdoctoral research post, with the possibility of extension to 4 years, working on the UKRI funded project ‘PROSPECT: Predicting psychosis outcomes from speech and brain connectivity’. The overall aim of the role is to develop innovative machine learning approaches to predict longitudinal symptom changes for patients with psychotic illnesses, using patterns of brain connectivity derived from MRI.

 

The post holder will work at the intersection of machine learning, neuroimaging and Psychiatry, developing methods with the potential to improve our ability to predict clinical outcomes. They will have the opportunity to work with rich brain MRI datasets from patients with psychotic illnesses, curate and process these datasets, and derive both functional and structural brain networks. This will include using our group’s new Morphometric Inverse Divergence (MIND) approach for estimating structural similarity networks, which enables robust structural brain networks to be derived from T1-weighted images alone and has already been shown to be sensitive to schizophrenia.

 

A key focus of the project will be on testing novel approaches to improve the accuracy of psychosis outcome prediction from structural and functional brain networks. To that end, we will explore uncertainty-aware machine learning approaches, to assess whether prediction accuracy can be improved by focussing on particular subsets of patients. We will also investigate different ways to define longitudinal outcomes, and test whether combining functional and structural networks in novel ways can help improve prediction accuracy. For example, we will explore whether functional networks generated from structural brain networks can be used to capture additional predictive power. If the role holder desires, there may also be scope to relate brain imaging to speech and language data.