Reinforcement Learning for Brain Connectivity Mapping with Tractography
Primary Supervisor:
Prof Daniel C. Alexander, UCL Centre for Medical Image Computing
Introduction:
The project is a 4-year PhD studentship to be based in the UCL Centre for Medical Image Computing (CMIC). The studentships comes with a generous stipend approximately at the Wellcome Trust rates (the maximum possible for PhD studentships in the UK). The successful candidate will align with the UCL Centre for Doctoral Training (CDT) in Intelligent, Integrated Imaging in Healthcare (i4health) cohort and benefit from the activities and events organised by the centre.
Project Background:
The project will develop novel ideas using reinforcement learning to guide brain connectivity mapping via tractography. Tractography is a method for mapping white matter pathways in the brain, using information from diffusion-weighted magnetic resonance imaging. Tractography has revolutionised our understanding of the connectivity architecture of the brain over the last few decades, as well as providing new insights into how brain diseases develop and spread. However, the technique has substantial limitations in terms of false positives and false negative connections.
Reinforcement learning provides a promising avenue to address some of these limitations: the problem is similar to complex games at which reinforcement learning excels, in that an agent must make local moves to optimise a global objective. We will build upon work by Théberge et al (2021, 2023), who were the first to use reinforcement learning for tractography, and have demonstrated competitive results compared to other methods.
The student will be working within the CU-MONDAI project and alongside the POND group, who use computational modelling to improve our understanding of neurodegenerative diseases such as Alzheimer’s disease. They will have the opportunity to apply their advances to improve the accuracy of these models and shed new light on the biological mechanisms of neurodegenerative disease.
Research aims:
- Finding the optimal environment and state variables to facilitate accurate connectivity mapping
- Designing an appropriate reward function for this novel application of reinforcement learning
- Applying the new approach to model pathology spread in Alzheimer’s disease
Person specification & requirements:
- The ideal applicant will have a strong mathematical and/or computational background
- Candidates should have interest in neuroimaging and/or neurodegenerative disease
- Experience with any of image analysis and deep learning, particularly reinforcement learning, would be advantageous but not essential.
- Candidates must have a UK (or international equivalent) first class or 2:1 honours degree and an MSc in physics, computer science, mathematics, engineering, or a comparable subject.
Application Deadline: 3rd July 2023
How to Apply
Please complete the following steps to apply.
- Send an expression of interest and current CV to: [[email protected]and [email protected]] Please quote Project Code: 23028 in the email subject line.
- Make a formal application to via the UCL application portal. Please select the programme code Medical Imaging TMRMEISING01 and enter Project Code 23028 under ‘Name of Award 1’.