CUDA Spotlight: GPU-Accelerated Biophotonics & Biomedical Optics
This week's spotlight is on Alexander Doronin, a PhD candidate in the Biophotonics & Biomedical Imaging Research Group at the University of Otago in New Zealand.
His research interests include biophotonics, light-tissue interaction, Monte Carlo (MC) computational modeling and parallel programming on GPUs using CUDA.
NVIDIA: Alex, what is biophotonics?
NVIDIA: What's an example of an application that could benefit from biophotonics?
While the presence of cancer may be clear, it can be difficult to identify the type of cancer, or whether the lesion is the primary tumor or not. Despite the best laboratory practices, the rate of conclusive diagnoses by histological analysis for a range of cancers is only 65-75 percent.
In our research, we are investigating the use of circular polarized light -- and the manipulation of the coherent properties of light -- to improve cancer diagnostics. The technique has the potential to revolutionize the ability to detect cancer at the very early stage.
NVIDIA: What's another example of an area you are working on?
The biophotonics community is actively working on an imaging technique that is sensitive to the fast changes of oxy- and deoxy- haemoglobin and blood flow malformations on a time scale of milliseconds. The instrumentation is relatively low cost, portable, non-invasive, and compatible with fMRI. Used in conjunction with fMRI, the new tool has the potential to improve the accuracy of brain imaging and significantly reduce time and costs.
Both of these examples – the cancer diagnostics and the brain diagnostics - require intensive image/signal processing as well as computational modeling of light propagation in biological tissues. Using massively parallel CUDA GPUs will significantly speed up the time required for simulations of photon migration, image analysis and visualization.
NVIDIA: What role does GPU computing play in your work?
In MC modeling, computation time has always been a significant concern. The imitation of light propagation within biological tissues for a particular diagnostic system normally takes hours or even days for one simulation, depending on the technical parameters of the system and complexity of the structure of biological tissue.
Due to the SPMD (single program, multiple data) nature of MC, it is a highly-parallelizable problem. The emergence of NVIDIA GPUs and the CUDA programming model, which are specifically dedicated to parallel processing, allowed us to rethink and redesign our MC algorithms, achieving a dramatic speedup of our simulations (340-1000X).
With CUDA, our computational time was reduced to from hours to minutes, i.e. a near real-time solution. The major performance bottleneck of MC was solved employing NVIDIA GPU technology and led to the development of our new online MC tool.
NVIDIA: Tell us about the online Monte Carlo tool developed by your group.
Leveraging modern, web-based technology, we have created a free online MC computational tool for researchers in the area of biophotonics and biomedical optics. On the server side, the tool is accelerated by CUDA GPUs. On the client side, a lightweight, user-friendly web interface allows multiple clients to set up optical system parameters, perform modeling, and download results in a typical journal paper format. We've combined powerful GPU technology with a modern web application development approach, allowing researchers to use, check and validate our MC model using our group's GPU computing facilities. We are currently extending our GPU cluster with additional M2090s and are expecting even more performance.
The online MC tool is available to the worldwide biophotonics community through the Biophotonics & Biomedical Imaging Research Group, which is headed up by my supervisor, Dr. Igor Meglinski.
NVIDIA: Where do you see biophotonics going in the future?
Online Monte Carlo Computational Tool