This project is available for students starting in September 2020 and is suitable for a 1-year MSc in Physics. The project does not currently have funding attached, students must be able to fund the fees and their living costs either through their own funds or a scholarship. Current information on fees is available here.
The essence of the proposed project is to explore and develop the use of high-performance computing, in particular Graphics Processing Units (GPUs), applied to signal processing and visualization in Optical Coherence Tomography (OCT).
OCT is a technique that is able to produce a 3D visualizations of typically biological samples, based on reflection and absorption of laser light (typically infra-red) at different depths within the sample. As such, it can be used to generate images of sub-surface features, such as inner layers of skin, or into and behind the retina of a human eye. Crucially, OCT is non-invasive, using relatively low-power lasers, important in these applications to avoid any damage to the sample. Ophthalmology is one area of medicine where OCT is used extensively in diagnosis and treatment of a range of diseases.
The Applied Optics group at Kent, led by Prof. Podoleanu, have developed state of the art techniques that enables the capture of data representing in excess of 1 million “voxels” (pixels in a 3-dimensional grid) in under 0.5s, fast enough for real-time imaging of the human retina . The structure of the data generated by the OCT imaging hardware (that connects to a standard PC via a plug-in board) requires significant amounts of processing in order to produce a meaningful visualization, mostly Fourier transforms or other techniques for cross-correlation. Whilst OCT techniques and optics have advanced in recent years, the speed and capabilities of a typical PC processor have not, and this has presented a challenge for researchers in this area.
Many groups are now investigating the use of commodity Graphics Processing Units (GPUs) as the work-horse for OCT data processing, in addition to hardware based (FPGA or ASIC) solutions. This includes Kent, where GPUs have been used to process and visualize the data received in real-time, as it is generated by the OCT hardware . GPUs, whose development over the past decade has been driven by (primarily) the computer gaming market, are essentially regular grids of high-performance calculation engines, that operate in concert, performing the same calculations but on different pieces of data. For computer gaming, these calculations transform virtual models into real pixels on the player’s screen, with increasingly high speeds, resolutions and complexity. For OCT, the calculations transform the data captured by imaging hardware similarly, though these calculations are significantly more complex and time-consuming than the average modern computer game, and implementing them efficiently is a challenge – particularly considering the sheer quantity of data involved (hundreds of megabytes per “frame”, with gigabytes of reference data to reconstruct a 3D model).
As GPUs become more powerful and more readily available, the application scope of real-time data processing for OCT widens. “Functional imaging”, for example, involves the analysis of data over a number of captured data frames in order to reveal features such as blood flowing through vessels in the eye. OCT angiography (OCTA) is considered a revolution in ophthalmology as it allows visualization of vessels with no dye, ie no need of an injection, no fluorescence and totally non-invasive. This is an area of active research and has the potential for significant real-world impact. Building on existing work done at Kent involving GPUs in OCT, the proposed project will investigate and explore:
- Efficient methods for handling ever larger amounts of data from OCT imaging hardware, being able to deliver this to single or multiple GPU devices for processing.
- Algorithms for functional imaging in real-time, using data captured over time, and how to implement these algorithms effectively given the available hardware (combinations of GPUs and CPUs).
- Signal processing techniques for eliminating noise (that arises from various sources) to produce sharper and more detailed visualizations.
- Using increasingly powerful GPUs to provide enhanced 3D visualization techniques, such as being able to identify/highlight, peel-away or manipulate the geometry of layers or features.
The main discipline of the work is digital signal processing, parallel processing and algorithms, extending out into Physics (optics) and Mathematics/Electronics (signal processing).
A likely research student would be from a CS background, with excellent programming skills, strong mathematics and the desire and motivation to work in this interdisciplinary setting.
In terms of immediate medical applications, there is interest in real time visualization of surgical processes. A novel method patented by Podoleanu’s group is highly parallel and can only be progressed by securing parallel signal processing. There is interest from colleagues at the UCL-Institute of ophthalmology in London and from eye surgeons at the Est Kent Hospitals University NHS Foundation Trust. Both institutions fund and co-supervise research in Podoleanu’s group. Equipment Several OCT systems in Podoleanu’s group are equipped with state of the art graphic cards..
There is no deadline – applicants will be assessed on a rolling basis. The candidates must be in place in September 2019. For further information or informal enquiries, please contact Prof Adrian Podoleanu (A.G.H.Podoleanu@kent.ac.uk).