Developing Foundation Models for Multimodal Neuroimaging Data

Details

Background

Neuroimaging is a crucial technique to understand human brain functions and disorders. Artificial intelligence, particularly machine learning techniques, have become increasingly popular for neuroimaging data analysis, offering advanced approaches in various applications. These include disease diagnosis and prognosis, where algorithms can detect early signs of neurological conditions like Alzheimer’s or Parkinson’s disease. In brain-state decoding, machine learning helps interpret complex patterns of brain activity, enhancing our understanding of cognitive processes and emotional states. Furthermore, in brain connectivity analysis, these techniques map the intricate network of neural connections, providing insights into the functional architecture of the brain.

The field of artificial intelligence has been experiencing a significant paradigm shift with the emergence of pre-trained foundation models. These models have seen immense success in natural language processing and computer vision tasks, enabling them to learn from language and/or visual data on an unprecedented scale and to combine visual and textual data for richer AI interactions. They have greatly facilitated efficient cross-domain applications.

Aims and objectives

This project aims to leverage the success of pre-trained foundation models for neuroimaging analysis. The objective is to develop interpretable and training-efficient foundation models for multimodal neuroimaging data, thereby enhancing downstream neuroscience tasks.

The primary focus will be on utilizing functional Magnetic Resonance Imaging data from public repositories such as OpenNeuro, HCP, and UK Biobank. This will be followed by exploring additional data modalities, including the extraction and understanding of textual data (annotations, behavioural data, publications) using advanced natural language processing techniques. Additionally, efforts will be directed toward enhancing model interpretability to improve transparency in neuroscience applications.

Research environment

The successful candidate will benefit from a multi-disciplinary supervisory team and a vibrant research environment, including the Department of Computer Science, Centre for Machine Intelligence, and School of Medicine and Population Health. The team offers extensive experience in cross-disciplinary projects, open-source software development, and access to multicentre and multimodal neuroimaging datasets. In addition to Sheffield’s excellent doctoral training programme, the candidate will have the opportunity to engage in a wide range of research activities of the group, contributing to publications, gaining experience of writing funding applications and developing their teaching experience.

Entry requirements

Candidates must have achieved a minimum 2:1 undergraduate and/or postgraduate master’s qualification (MSc) in Computer Science, Neuroscience, or other relevant subjects, by the start of the PhD. The English language requirements must also be met by the start of the PhD. Candidates with a strong mathematical or statistical background and an interest in large datasets and open-source software development are desirable.

Interested candidates are strongly encouraged to contact the primary supervisor (Dr Shuo Zhou; [email protected]) to discuss their interest in and suitability for the project prior to submitting their application.

Interviews are likely to be held between 4 – 15 March. Students must be able to start by October 2024.

Applications are open to students from both the UK and overseas. We anticipate competition for these studentships to be very intense. We would expect applicants to have an excellent undergraduate degree in a relevant discipline. We would also expect applicants to have completed or be undertaking a relevant master’s degree to a similar very high standard (or have equivalent research experience).

Please ensure you pick the Department/Division of Neuroscience when filling in your application form, regardless of where your first supervisor sits.