Using Gradients of Connectivity to Predict Cognitive Impairment After Stroke

2026-07-22

Many stroke survivors have chronic problems with language, attention, cognitive control and motor function. There is an urgent need to predict these sequelae so that rehabilitation can be better focussed, but it has proved challenging to anticipate the effects of stroke from brain lesions alone. We will use recent advances from our ERC-funded FLEXSEM project to predict the long-term consequences of stroke. Our work decomposes whole-brain connectivity patterns derived from resting-state functional magnetic resonance imaging (fMRI) into components, known as ‘gradients’ that capture substantial variance in function, and links these gradients to individual differences in cognitive ability. Pilot work suggests that (i) connectivity gradients can predict the chronic effects of stroke on language comprehension and (ii) that changes to these gradients can be anticipated from the lesion alone. This suggests our approach has high clinical utility, since stroke survivors are typically assessed with structural but not functional MRI. This PoC project will validate our approach across a range of domains, including speech fluency, comprehension, phonology and cognitive control. We will use resting-state fMRI and structural scans along with high-quality neuropsychology. We will parcellate the fMRI data from patients, compute the connectivity between pairs of parcels, perform dimension reduction on the connectivity matrix to generate gradients, and identify associations with cognitive problems. We will then examine the extent to which post-stroke cognitive changes can be predicted from structural scans by estimating lesioned gradients using data from aged-matched controls. Focus groups with speech therapists and stroke survivors, co-production of the validation work and dissemination activities for clinicians will ensure the tools we develop are suited to European healthcare systems to identify targets for rehabilitation.

United Kingdom (UK)

2026-01-01