The Usage Of GPU In Data Processing And Real-time Imaging Of FD-OCT

Optical Coherence Tomography(OCT) is an established medical optical imaging technique which is demonstrated in 1990 s. This technique can be used for scanning scattering medium such as tissue. Optical Coherence Tomography has achieved sub-micrometer axial resolution, millimeter penetration depth, and has many advantages such as high speed data acquisition, noninvasive, non-contact. Because of those merits, much research into the application of OCT to biological medical areas has been carried out, especially into the application in the ophthalmology to diagnosis and cure for various diseases.Since OCT was first demonstrated in 1991, the A-scan rate has gone up from 400 Hz to 20 MHz in experimental systems. Most of the commercially available SDOCT systems operate using line scan rates within 30 – 85 KHz to produce high quality 2D B-scan images. The fastest spectrometer-based SDOCT systems that have been developed operate with a dual camera configuration and a 500 KHz A-scan rate[1].GPU’s demonstration breaks the bottleneck of OCT data processing using traditional CPU system due to CPU’s lower computation power and clock rate. GPU is the microprocessor embedded in graphics card as a more computing device. NVIDIA, which is a graphics card company,first created a parallel computing platform and programming model and now implements to the main-stream graphics processing units they produces. This platform is called CUDA, which stands for Compute Unified Device Architecture. With CUDA, thousands of microprocessors can be used for computational-intensive work within threading processor mechanism.Facing the low-speed constraint that traditional CPU platform OCT data processing system brings, the focus of this paper is on using GPU of GTX Titan as a computational platform, parallelizing the data processing procedure with CUDA, then optimizing the algorithm and procedure of data processing. Finally, we implement a real-time display and storage software in Windows system. The result shows that GPU-based method improves the processing speed compare to CPU-based method. GPU-based method uses a commercial graphics card, is more economical, energy-saving, can be applied in many area in the future.