Real-time resampling in Fourier domain optical coherence tomography using a graphics processing unit

Abstract

Abstract. Fourier domain optical coherence tomography ͑FD-OCT͒ requires either a linear-in-wavenumber spectrometer or a computationally heavy software algorithm to recalibrate the acquired optical signal from wavelength to wavenumber. The first method is sensitive to the position of the prism in the spectrometer, while the second method drastically slows down the system speed when it is implemented on a serially oriented central processing unit. We implement the full resampling process on a commercial graphics processing unit ͑GPU͒, distributing the necessary calculations to many stream processors that operate in parallel. A comparison between several recalibration methods is made in terms of performance and image quality. The GPU is also used to accelerate the fast Fourier transform ͑FFT͒ and to remove the background noise, thereby achieving full GPU-based signal processing without the need for extra resampling hardware. A display rate of 25 frames/ sec is achieved for processed images ͑1024ϫ 1024 pixels͒ using a line-scan charge-coupled device ͑CCD͒ camera operating at 25.6 kHz. Optical coherence tomography ͑OCT͒ is a noncontact, noninvasive interferometric technique allowing micrometer resolution imaging of tissues. OCT can be classified into two categories: time domain ͑TD͒ OCT and spectral domain ͑SD͒ OCT. SD-OCT eliminates the need for mechanical scanning in TD-OCT to produce reflectivity profiles ͑A-scans͒ by recording the interference signal as a function of wavelength. This can be done either by using a spectrometer when performing Fourier domain ͑FD͒-OCT, to spatially encode the different optical frequencies, or by encoding them in time using a frequency scanning source, when performing swept source OCT. Both SD-OCT schemes have advantages of a better signal-to-noise ratio than that of the TD-OCT and a significantly improved imaging speed.1 The increased imaging speed requires very fast digital signal processing that is able to keep up with over tens of megahertz video rate data acquisitions. OCT digital signal processing mechanisms based on current central processing units ͑CPUs͒ are not capable of coping with such data processing speeds, and limit the sample rate dramatically.The rapid development of graphics processing units ͑GPUs͒ has introduced a revolution in numerical calculations. A GPU consists of many stream processors acting concurrently, thus favoring parallel programming. A recent report describes a GPU-based FD-OCT setup, where signal processing was performed partly by optical hardware and partly by the GPU 2 to enable real-time OCT imaging. We show here that similar data rates are possible by transferring all processing tasks to the GPU.FD-OCT employs a spectrometer to disperse the output of the interferometer onto a linear digital camera. A fast Fourier transform ͑FFT͒ of the linear camera signal provides the A-scan. However, the FFT operates in the optical frequency domain, while the equally spaced camera pixels produce an approximately linear representation of the spectrum in waveleng...

Medical image processing on the GPU – Past, present and future

“…The processing could be performed in real-time, such that volumes could be reconstructed and visualized with ray-casting at a frame rate of 10 Hz. Work reported by van der Jeught et al (2010) and Rasakanthan et al (2011) also include a GPU for resampling and inverse FFTs of OCT data in real-time. The resampling was eventually improved by using zero-padding and FFTs (Watanabe et al, 2010), instead of simple linear interpolation.…”

Simple Spectral Calibration Method and Its Application Using an Index Array for Swept Source Optical Coherence Tomography

“…After both array calculations are completed in the calibration step, the FPI and OSA are not used any more for regular imaging sessions. Another merit of our system is that it can decrease processing time by high-speed signal processing using GPU [16][17][18]. Most methods previously reported in the literature adapt an interpolation process in the k-domain linearization.…”

Imaging the eye fundus with real-time en-face spectral domain optical coherence tomography

“…Line rates larger than 300 kHz [17] have been reported using fast linear cameras in spectrometers in Sp-OCT and line rates as over several MHz using fast tunable lasers in SS-OCT [18]. To speed up the process, graphics processing units (GPU) [19] and field-programmable gate arrays [20] have been reported.…”