Graphics Processing Unit-Based Ultrahigh Speed Real-Time Fourier Domain Optical Coherence Tomography

Abstract

This paper describes a series of novel graphics processing unit (GPU)-based image reconstruction and visualization methods that we developed for realizing ultrahigh speed, real-time Fourier domain optical coherence tomography (FD-OCT). Several GPU-based algorithms including high-speed linear/cubic interpolation, gridding-based nonuniform fast Fourier transform (NUFFT), numerical dispersion compensation, and multi-GPU implementation were developed to improve the point-spread function, SNR roll-off, and stability of the system. Full-range complex-conjugate-free FD-OCT was also implemented on the GPU architecture to double the imaging range and to improve the SNR. The maximum processing speed of <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$&gt;$</tex></formula> 3 Gigavoxels/s ( <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">$&gt;$</tex></formula> 6 Mega-A-scans/s of 1024-pixel FD-OCT) was achieved using the latest GPU module from NVIDIA. The GPU-based volume rendering enabled real-time four-dimensional (4-D), i.e., 3-D + time, FD-OCT imaging, and a 5 volume/s 4-D FD-OCT system was demonstrated via in vivo tissue imaging. These GPU technologies were highly effective in circumventing the well-known imaging reconstruction and visualization bottlenecks existing among current ultrahigh speed FD-OCT systems and could open the door to realize interventional OCT imaging for applications in guiding microsurgery.