Super-high speed Fourier-domain optical coherence tomography (FD-OCT) sure would be useful for such clinical applications as surgery: Real-time 3D video would give surgeons a better understanding of tissue and provide instrument guidance during operations. But 3D imaging generates massive amounts of data, and while FD-OCT acquisition line (A-scan) speed has imcreased dramatically in recent years, no such progress has been made in terms of processing and visualization. So real-time video-rate displays have been limited to two dimensions.
Current methods for generating 3D OCT video involve capturing and storing image data for later processing. Data post-processing usually includes two computationally intensive steps: signal processing and volumetric visualization. While several solutions have been proposed and demonstrated, they require complex hardware modifications to specialized systems and, thus, are not commonly applicable. And they have other important limitations.
With a goal to "develop a general, low-cost method to solve the real-time data processing and visualization bottlenecks," researchers at Johns Hopkins University aimed for an approach that could "be easily applied on a regular nonlinear-k FD-OCT system without extensive hardware modification on the system architecture." Noting that OCT data is highly suitable for independent parallel processing, they looked to a graphics processing unit (GPU), which "devotes more transistors to processing than a standard GPU." They now report real-time 4D (3D plus time) operation using a standard FD-OCT system with a nonlinear-k space spectrometer.1
An ultra-high speed linear spline interpolation (LSI) method for λ-to-k spectral re-sampling at the GPU enables average interpolation speeds of >3,000,000 line/s for 1024-pixel OCT (1024-OCT) and >1,400,000 line/s for 2048-pixel OCT (2048-OCT). The complete GPU-based FD-OCT signal processing also includes fast Fourier transform (FFT) and post-FFT processing.
The researchers report maximum complete A-scan processing speeds of 680,000 line/s for 1024-OCT and 320,000 line/s for 2048-OCT, which correspond to 1 GB processing bandwidth. The team's experiment used a 2048-pixel CMOS camera running up to 70 kHz is for acquisition—which limited actual imaging speed to 128,000 line/s for 1024-OCT and 70,000 line/s for 2048-OCT. Their setup continuously acquires 3D data sets in real time at 1024-OCT mode, and immediately processes and visualizes them at rates as high as 10 volumes/second (12,500 A-scans/volume) by either en face slice extraction or ray-casting based volume rendering from 3D texture mapped in graphics memory.
The researchers say that no optical modification is needed, and that the technique can be easily integrated into most ultra-high speed FD-OCT systems. While their setup uses an NVIDIA Quadro FX 5800 GPU, they say that lower cost devices, such as NVIDIA's GTX series units, should offer similar performance. Overall, the hardware upgrade to realize real-time 3D FD-OCT can be "as low as $500."
- K. Zhang and J. Kang, Opt. Express 18, 11772-11784 (2010)