Developing a neuroscience-led basis for diagnosis, prognosis, management and therapy for aphasia post-stroke.

There are approximately 100,000 stroke incidents per year in the UK. Impairments in language (and other higher cognitive functions like memory, multi-tasking and problem solving) are common symptoms post-stroke, effecting about 33% of patients in the early stages after a stroke and 20% in the long term. The inability to communicate effectively and complete simple daily tasks has a negative impact on quality of life (i.e., unable to work, withdrawing from social activities, etc.) and mental health leading to higher dependence on social and assisted care. It is critical, therefore, to understand the nature of the problems experienced after stroke to help manage long term care and plan for effective rehabilitation. Our current understanding of language problems after stroke and how they are treated comes from careful study of behaviour from a wide range of disciplines (e.g., neuropsychology, linguistics, and sociology). In contrast, the field of cognitive neuroscience has vastly improved our understanding of language as new ways of scanning the brain have been developed but this has largely focused on intact language function, with only minimal progress in understanding how to use this technology for diagnosis and prognosis. These technologies are exciting because not only can we use them to identify areas of brain damage after a stroke, we can also determine how connecting fibres or the way areas communicate is disrupted and importantly, how the brain might recover over time. My Patient and Public Involvement groups have indicated that uncertainty around recovery causes great distress. In addition, time constraints imposed on rehabilitation mandate efficient treatment plans; something that can be improved by understanding recovery. This Fellowship has two goals: 1) create a unique database of long-term recovery after stroke; and 2) apply machine learning to behavioural and neuroimaging data. This will allow me to: (a) improve diagnosis by identifying important brain regions for specific aspects of language and executive functions; (b) understand prognosis by investigating how recovery after brain damage changes the way our brain operates and how connectivity between different areas change over time; and (c) build and test models of recovery using machine learning (i.e., use a brain scans to determine the expected severity and type of behavioural problems at one-year and the likelihood of recovery across different behavioural domains). This will be achieved using a combination of existing chronic stroke data from three sources (each with samples larger than >80) and through a new recovery cohort collected throughout this project. Stroke survivors will be assessed within one-month of having a stroke with detailed neuroimaging and neuropsychology testing, in collaboration with clinical partners at Cambridge University Hospitals NHS Trust: approximately 900 stroke admissions per year. Each case will be re-assessed one-year later with the same assessment battery (managed by a dedicated research assistant). The outcome of the project will improve our basic scientific knowledge around behavioural problems after stroke and how they change over time, while also being able to inform management and therapy strategies for patients - bridging the currently large gap between technical brain imaging, computer science and the clinical arena. Importantly, it is widely known that damage due to a stroke differs greatly between people. The model predictions that can be achieved in this Fellowship will provide behavioural estimations at the individual level rather than the group level, allowing for a pathway towards precision medicine.