Changing gradients: Predicting cognition and wellbeing following stroke using whole-brain intrinsic connectivity gradients and machine learning

Background:

Stroke is a leading cause of cognitive impairment, which often leads to difficulties in mobility, communication, attention and working memory, alongside poor wellbeing. While research has investigated the functional consequences of brain injury, it is hard to predict post-stroke recovery, since patients with similar lesions can have very different outcomes. This might be because stroke disrupts whole-brain patterns of connectivity in addition to local effects. Resting-state fMRI can characterise the impact of stroke on large-scale connectivity gradients, and this may provide a better way of understanding post-stroke recovery than analysis of specific sites or networks. 

Objectives:

Using analytical approaches recently advanced by the supervisors, the student will decompose whole-brain patterns of resting-state connectivity into ‘gradients’ which capture key dimensions of connectivity across the cortex. We will establish (i) how stroke disrupts dimensions of connectivity both locally and at a distance from the lesion, (ii) how changes in connectivity gradients predict cognition and wellbeing following stroke and (iii) the extent to which a structural scan, delineating a patient’s lesion, is sufficient for predicting changes in connectivity gradients (important since resting-state data is often not available in a clinical context).

Novelty:

Our gradient decomposition approach is relatively novel (2016, PNAS) and has not yet been used to explore functional recovery following stroke. White-box machine learning approaches (2014, IEEE Transactions on Evolutionary Computation) will also be used to identify where connectivity changes have the greatest predictive value. 

Timeliness:

Age is a major risk factor for stroke and the burden of this disease is increasing as people live to older age. This research may allow better prediction of outcomes and more effective deployment of therapy following stroke. The project is also expected to have high scientific impact, given the growing profile of gradient decomposition and machine learning.

Experimental Approach:

The project will use data from PLORAS, a large-scale study of 150 stroke survivors and 74 age-matched controls, who have structural and resting-state MRI. Stroke survivors completed assessments of cognition, language (semantics, phonology, orthography, speech fluency) and changes in lifestyle and wellbeing. We will test specific hypotheses about how these psychological changes relate to changes in gradient space: for example, we predict comprehension will be associated with reduced differentiation of connectivity patterns near the apex of the ‘principal gradient’, which captures the distinction between heteromodal memory and sensory-motor processes.

The York Biomedical Research Institute at the University of York is committed to recruiting extraordinary future scientists regardless of age, ethnicity, gender, gender identity, disability, sexual orientation, religion/belief, marital status, pregnancy and maternity, or career pathway to date. We understand that commitment and excellence can be shown in many ways and have built our recruitment process to reflect this. We welcome applicants from all backgrounds, particularly those underrepresented in science, who have curiosity, creativity and a drive to learn new skills.