Neuroscience Approaches to Stroke Rehabilitation and Predictions of Brain Health

Up to 80% of people who experience a right-hemisphere stroke suffer from hemispatial neglect, a severe visuospatial impairment, where they fail to perceive/misperceive items presented in the contralesional part of space. This syndrome is debilitating, impedes rehabilitation, and is a strong independent predictor of poor recovery. In my talk I will present data showing that immediate on-line actions are relatively unimpaired in these neglect patients. Building on this insight, I will discuss two recent rehabilitation studies where we assessed whether these spared actions could be exploited for neglect rehabilitation a) via behavioural visuomotor feedback training or b) in a clinical trial using non-invasive brain stimulation. I will also discuss the feasibility of using virtual and augmented reality for cognitive training and stroke rehabilitation. I will finish by presenting a first AI foundation model we have started to build to estimate cognitive impairments from brain (MR) images. Cognitive performances are often evaluated using medical assessments, while brain imaging would offer a more reliable and reproducible estimate of cognitive function, and its decline over time. With recent advances in artificial intelligence, as well as the increasing availability of open datasets like the UK Biobank, I will introduce a novel way to interpret brain images, training an AI model from a multitude of cognition data.