GPU-accelerated non-uniform fast Fourier transform-based compressive sensing spectral domain optical coherence...

Area of Science:

• Biomedical Optics
• Medical Imaging Technology
• Computational Imaging

Background:

• Spectral domain optical coherence tomography (SD OCT) is a crucial imaging modality.
• Compressive sensing (CS) techniques accelerate SD OCT acquisition but require efficient algorithms.
• Non-uniform sampling in k-space presents computational challenges for SD OCT.

Purpose of the Study:

• To implement and evaluate a graphics processing unit (GPU) accelerated compressive sensing (CS) non-uniform in k-space spectral domain optical coherence tomography (SD OCT) system.
• To compare the performance of a gridding-based non-uniform fast Fourier transform (NUFFT) algorithm against existing methods.
• To demonstrate the feasibility of real-time B-mode imaging using the developed system.

Main Methods:

• Implementation of GPU-accelerated CS SD OCT utilizing a gridding-based NUFFT algorithm.
• Independent use of Kaiser-Bessel (KB) and Gaussian functions as convolution kernels within the NUFFT.
• Comparison with GPU-accelerated modified non-uniform discrete Fourier transform (MNUDFT) and fast Fourier transform (FFT)-based CS SD OCT.

Main Results:

• The NUFFT-based implementation achieved comparable image quality to MNUDFT-based CS SD OCT.
• A speed enhancement of over 5 times was observed compared to MNUDFT-based CS SD OCT.
• The NUFFT approach demonstrated reduced background noise and side lobes compared to FFT-based CS SD OCT, eliminating k-space grid filling and k-linear calibration.
• Real-time B-mode imaging exceeding 30 fps was achieved on a standard desktop with three GPUs.

Conclusions:

• GPU-accelerated NUFFT-based CS SD OCT offers a significant speed improvement over existing methods.
• This approach enhances image quality by reducing noise and side lobes.
• The developed system enables real-time, high-frame-rate B-mode imaging for advanced OCT applications.