Nuclear medicine is an important branch of medical physics and is used to diagnose and treat cancers, among other conditions. There is a constant drive to improve the quality of imaging technology, as better images have a direct benefit to patient outcome and may even enable new diagnostic tests to be developed.
Artificial intelligence (AI) and machine learning are of great interest to the medical community and research is ongoing in a wide range of areas, including distinguishing between benign and cancerous tumours and improving detector performance. However, the nature of AI techniques often leads to ‘black box’ analysis software, which is a particular danger in medical imaging where analysis errors could be catastrophic. During this project you will develop and analyse machine-learning derived algorithms and compare these to known physical processes with the goal of opening these black boxes – deriving methods that may have applications in a wide range of areas.
The gamma energies used in nuclear medicine create challenges in detector design. When designing detectors for nuclear medicine there is often a trade-off to be made between energy resolution and sensitivity – both important parameters for medical imaging. This project will involve the application of AI and machine-learning techniques to data from detectors for nuclear medicine, to improve energy resolution while maintaining sensitivity.
One area of focus will be compound semiconductor detectors, which are used in sectors as diverse as medical imaging and astronomy due to their high detection efficiency, energy resolution, and ability to operate at high temperatures and in extreme environments. Recent developments in compound semiconductor detectors are of great interest for nuclear medicine, where their improved energy resolution over traditional detectors could allow the use of novel dual-isotope imaging techniques.
When detecting high energy photons (>100keV), as is the case for nuclear medicine applications, depth of interaction effects and charge sharing can lead to a degradation in energy resolution in compound semiconductor detectors. For detectors with small pixel sizes, a single detected photon may produce a signal across multiple pixels. The pattern of this signal is correlated to the amount of energy lost when this event is reconstructed. However, the large range of possible patterns means this reconstruction is difficult to solve using traditional techniques.
Artificial intelligence can provide a solution to this problem. The tools of AI can be roughly divided into knowledge-based and data-based techniques. The latter include machine learning algorithms based on large neural networks, which can recognise signature patterns after training on historical data. The knowledge-based techniques provide an explicit representation of specialist knowledge, in this case the physics of photon interactions within the detector. The knowledge-based techniques add context and sense-checking to the classification from the machine-learning algorithms. In this way, the project will link the analysis of machine-learning derived algorithms to the fundamental physical processes in the detector.
This PhD position is being offered as part of a collaboration between Loughborough University (contact [email protected]
), the University of Portsmouth (contact [email protected]
), and the Open University (contact [email protected]
Applicants may choose to base this PhD at any of these sites and should apply to the contact listed for their chosen institution.For relevant projects available at Portsmouth or the OU please contact the appropriate person above. If the appropriate funding is available, a successful applicant would be able to undertake placements at project partners during their PhD. Additional bench fees may be required, please contact your proposed supervisor to discuss.
Applicants are expected to have a background in either Physics or Computer Science – a 2:1 honours degree or equivalent or work experience in a related area - with an interest in applying their skills to interdisciplinary problems. This project will include experimental work in a laboratory and there is scope to collaborate with partners across institutions and sectors (academic, industrial, clinical).
How to apply
All applications should be made online: https://www.lboro.ac.uk/study/postgraduate/apply/research-applications/
Under school/department name, select 'Physics'. Please quote reference PH/SB-Un1/2020.
The deadline for applications is 31 March 2020.
Start date: July 2020, October 2020, January 2021
Full-time/part-time availability: Full-time (3 years), Part-time (6 years)
Fee band: Band RB (UK/EU: TBC; international: £22,350)
Links to supervisors' online staff profile pages: https://www.lboro.ac.uk/departments/physics/staff/academic/sarah-bugby/ https://www.lboro.ac.uk/departments/compsci/staff/academic-teaching/georgina-cosma/