In the most recent Global Burden of Diseases, Injuries, and Risk Factors Study (GBD), stroke was the 2nd largest cause of death globally and contributed to the 2nd highest disability-adjusted life-years (DALYs) worldwide. Minimising the time to treatment for patients is extremely important with quality of life is significantly improved by reducing time to treatment by 15 minutes [1]. Consequently, there is a need to optimize the treatment of acute stroke in order to reduce disability and cost. A major contributing factor to this delay is the failure of patients, first-responders, and paramedics to recognize signs of stroke. Clinical expertise is scarce and often not immediately accessible at the time that a stroke occurs. As a result, many patients either fail to receive timely treatment or are ineligible for acute stroke therapy. A technology that can accurately detect key signs of impaired upper limb coordination or unilateral weakness in the arms could go towards the creation of an automatic identification of neurological impairment in the upper limbs. Pose estimation has been identified as a promising methodology to track key physiological landmarks in video data of humans [2]. However, these models can often be computational expensive and face challenges differentiating 3D movements in the video’s 2D representation of the real-world. Moreover, existing models can often struggle to identify landmarks when they are obscured or when the video is captured in poor lighting or from an undesirable angle. This project aims to improve pose estimation modelling for the detection of stroke symptoms.

This project aims to develop novel, light-weight pose estimation models to address some of the deficiencies in the existing literature. This will involve undertaking research in the area of artificial intelligence and video analytics.
Objectives:
The focus of the project will be to develop novel artificial intelligence methods with the following specific objectives:
- Develop a novel pose-estimation model to approximate the location of obscured landmarks.
- Refine the developed model to find a minimal solution that maintains sufficient model performance whilst minimising the computational cost.
- Use the model to identify symptoms of stroke during standard neurological examinations; namely impaired upper limb coordination and unilateral weakness in the upper limbs.

[1] D. S. Liebeskind, R. Jahan, R. G. Nogueira, T. G. Jovin, H. L. Lutsep, and J. L. Saver, “Early arrival at the emergency department is associated with better collaterals, smaller established infarcts and better clinical outcomes with endovascular stroke therapy: Swift study,” Journal of neurointerventional surgery,vol. 8, no. 6, pp. 553–558, 2016.

[2] Desmarais, Yann, et al. "A review of 3D human pose estimation algorithms for markerless motion capture." Computer Vision and Image Understanding 212 (2021): 103275.

Funding Information

*Please note that the deadline for applications from international candidates closed on 28 February*

Please note that funding may be available for this project (still to be confirmed). To be eligible for consideration for a DfE Studentship (covering tuition fees and maintenance stipend of approx. £17,668 per annum), a candidate must satisfy all the eligibility criteria based on nationality, residency and academic qualifications. The Studentship is open to UK and ROI nationals, and to EU nationals with settled status in the UK, subject to meeting the specific DfE nationality and residency criteria. Full eligibility information can be viewed via: https://www.economy-ni.gov.uk/publications/student-finance-postgraduate-studentships-terms-and-conditions

Academic Requirements:
The minimum academic requirement for admission is normally an Upper Second Class Honours degree from a UK or ROI Higher Education provider in a relevant discipline, or an equivalent qualification acceptable to the University.