UCL EPSRC Centre for Doctoral Training in Intelligent Integrated Imaging in Healthcare


Segmentation of stroke lesions in hospital scans for predicting language impairment outcomes

Now Closed


8 June 2020

Primary supervisor: Prof John Ashburner
Secondary supervisor: Prof Cathy Price FRS

Project summary
A 4-year PhD studentship is available in the Wellcome Centre for Human Neuroimaging (WCHN) in the UCL Queen Square Institute of Neurology. The funding covers an annual tax-free stipend and tuition fees. Funding is jointly through Prof Price’s Wellcome Trust Principal Research Fellowship and the EPSRC, so standard EPSRC eligibility criteria apply. Please see the EPSRC website for further details. The successful candidate will join the UCL CDT in Intelligent, Integrated Imaging In Healthcare (i4health) and benefit from the activities and events organised by this centre. The anticipated start date is 21st September 2020.

There are over 1.2 million stroke survivors in the UK, many of who will have lost language functions. The locations of strokes in the brain can predict how well patients recover certain functions, so knowledge of the pattern of damage may allow programmes of speech therapy to be tailored to optimise patient outcomes. Stroke patients, and their families, also wish to know their prognosis. For more information about our work, see https://www.ucl.ac.uk/ploras/.

Your contribution will be to develop a general-purpose software framework for accurate semantic segmentation of regions of stroke damage in hospital brain scans.

In this PhD project, you will develop usable software for fully automated and accurate semantic segmentation of stroke damage from clinical brain scans (both MRI and CT). The software should be applicable to both CT scans and MRI scans acquired using a variety of different scanner settings. The urgency of acute stroke treatment means the scans you will be working with were probably acquired rapidly, so will be of poor quality and spatial resolution.

During this PhD, you will acquire transferable skills and experiences that should equip you to work in a variety of technical environments. Our aim is for you to gain insights and understanding about how machine learning algorithms work, rather than treat them as some sort of magical black box optimised via “grad-student descent”. Your work will be carried out as part of a collaborative, multidisciplinary team of computer scientists, physicists, biomedical engineers, psychologists, neuroscientists and clinicians, leading to it being a highly stimulating multidisciplinary environment for learning and for scientific research.

Project Approach
Your initial work will compare various approaches to stroke segmentation, identifying their strengths and weaknesses, and the extent to which they generalise across different types of brain scans. The remainder of the project will depend on results from these experiments and will be tailored to match your abilities.

The supervisory team expect that out-of-the-box convolutional neural network approaches may not handle the diversity of brain scans well enough to provide a sufficiently general-purpose solution. We anticipate that you will need to develop a semi-supervised generative modelling strategy, finding synergies between an existing widely used modelling approach and some more recent advances in deep learning. Your work will involve a lot of probability theory, along with an image registration component for overlaying deformable priors that encode what should be observed in unaffected brain regions. This strategy works well for healthy tissue and is routinely used in the neuroimaging field, but will need to be extended to handle pathology, such as stroke. This will require you to develop and integrate a more sophisticated generative model of binary images to encode prior knowledge about the variability of pathology.

You must have, or expect to obtain, a UK first class or 2:1 honours degree (or equivalent international qualifications or experience) in an appropriate engineering, statistics, mathematics or physics based subject. You must show a clear interest in medical image analysis, particularly as applied to neuroscience and healthcare. Experience in numerical computing and programming, in languages such as MATLAB, Julia or C/C++, will be required. Other desirable skills include creative problem solving and critical thinking, good writing and communication skills, the ability to work effectively both in a team and independently, and to be highly self-motivated.

To Apply
Please send a CV and covering letter expressing your interest, and names and contact details of two referees, to Prof John Ashburner (j.ashburner@ucl.ac.uk). The formal deadline is 21st June 2020, but applications will be continuously assessed so we recommend submitting early.