1st Supervisor: Wenjia Bai, Imperial College London
2nd Supervisor: Dr Emma Robinson, King’s College London
Clinical Supervisor: Prof Paul Matthews, Imperial College London
Aims of the Project:
- Developing machine learning methods for deep brain structure segmentation from partial annotations.
- Developing geometric deep learning methods for 3D shape analysis of deep brain structures and understanding of their longitudinal evolutions.
- Developing unsupervised learning methods for rare and novel class discovery in large-scale imaging datasets.
Understanding the variations of brain structures across populations is a classical problem that has attracted research attention for decades. Although numerous efforts have been made to segment and analyse brain structures, certain deep brain structures, including the substantia nigra (SN), subthalamic nucleus (STN) and red nucleus (RN), remain less investigated. Morphological and functional changes of these structures have been demonstrated to be associated with Parkinson’s disease (PD). For example, the loss of the dopaminergic projection neurons of SN plays a role in PD, which can be reflected as volumetric changes or diffusivity differences on magnetic resonance imaging.
There are certain technical challenges that inhibit the understanding of deep brain structures, including limited annotated samples for training an automated segmentation model, relatively low spatial resolution of these small structures and challenges in analysing subtle yet important differences of their morphologies. In this project, we aim to leverage large-scale datasets such as the UK Biobank and develop novel image and shape analysis methods to address these challenges, with the ultimate aim to further our understanding about deep brain structures.
The expected candidate will have a master’s degree from computing, engineering, statistics or other related background. Proficient programming skills, knowledge in machine learning and computer vision are required.