Controlled Synthesis of Virtual Patient Populations with Multimodal Representation Learning

Project description

This project invites PhD candidates who are interested in developing generative models that can deal with complex multimodal patient data, contributing to safer, better and faster innovations of medical products via in-silico trials.

In-silico clinical testing/trials (ISTs) are an emerging approach to produce evidence for medical product innovation and regulation. Key to delivering ISTs are generative models, which can produce controlled virtual populations from vast real-world data sources. The generated virtual populations can expand/enrich the diversity of anatomical and physiological scenarios under which novel treatments can be tested within the in-silico trial framework. This thus ensures a broader and more equitable coverage of safety and efficacy for a given medical device.

You will work with multimodal data coming from radiological imaging, electronic health records, and wearable sensor data, etc. The synthesised virtual patient populations will comprise plausible instances of anatomy (from imaging) and physiology (from wearable sensors), and will not be traceable to any specific real patient data. The synthesis will be controllable, e.g., to impose specific conditions on the target virtual population to reproduce concrete inclusion and exclusion criteria in subsequent in-silico trials. You are expected to address modelling challenges in multimodal synthesis by considering at least one of the following aspects:
??? Limited availability of data (e.g., missing modality and insufficient patient instances) with potential inclusion biases.
??? Effective incorporation of domain medical knowledge in the synthesis (e.g., constraints arising from anatomical or physiological considerations).
??? Novel synthesis performance metrics that qualify and quantify the plausibility and representativity of synthetic populations.
??? Patient privacy protection and proven model trustworthiness.

The solutional development will build on state-of-the-art techniques in machine learning and computer vision areas, including but not limited to generative modelling, multimodal learning, zero-shot/few-shot learning, geometric deep learning, manifold learning, and differential privacy. You will have access to data resources available through INSILEX and INSILICO Programmes. The supervisory team has a strong record of research success in machine learning, medical imaging and computer vision in general, and supported by a unique cadre of clinical experts that will provide contextual guidance to the student.

We are eagerly inviting strong and passionate applicants who have:
??? High interest in this project, preferably being experienced in some of the above-mentioned machine learning or computer vision areas.
??? An excellent undergraduate degree in Computer Science or Mathematics (or related discipline), and preferably, a relevant M.Sc. degree.
??? Very good experience with computer programming of mathematical, imaging and/or machine learning models and algorithms.
??? Excellent report writing and presentation skills.
??? Excellent ability to communicate with fellow students and colleagues, and importantly medical experts.

Qualified applicants are strongly encouraged to informally contact the supervising academics Dr. Tingting Mu (tingting.mu@manchester.ac.uk) and Prof. Alejandro Frangi (alejandro.frangi@manchester.ac.uk) to discuss your application and research proposal prior to applying.