J Biomed Opt. 2012 Oct;17(10):100502. doi: 10.1117/1.JBO.17.10.100502.
The authors present the application of graphics processing unit (GPU) programming for real-time three-dimensional (3-D) Fourier domain optical coherence tomography (FdOCT) imaging with implementation of flow visualization algorithms. One of the limitations of FdOCT is data processing time, which is generally longer than data acquisition time. Utilizing additional algorithms, such as Doppler analysis, further increases computation time. The general purpose computing on GPU (GPGPU) has been used successfully for structural OCT imaging, but real-time 3-D imaging of flows has so far not been presented. We have developed software for structural and Doppler OCT processing capable of visualization of two-dimensional (2-D) data (2000 A-scans, 2048 pixels per spectrum) with an image refresh rate higher than 120 Hz. The 3-D imaging of 100×100 A-scans data is performed at a rate of about 9 volumes per second. We describe the software architecture, organization of threads, and optimization. Screen shots recorded during real-time imaging of a flow phantom and the human eye are presented.
作者提出了图形处理单元（GPU）编程在实时三维（3-D）傅里叶域光相干断层扫描（FdOCT）成像中的应用，实现了流动可视化算法。FdOCT 的一个限制是数据处理时间，通常比数据采集时间长。利用附加算法，如多普勒分析，进一步增加了计算时间。通用 GPU 计算（GPGPU）已成功用于结构 OCT 成像，但实时 3-D 流动成像尚未实现。我们开发了用于结构和多普勒 OCT 处理的软件，能够以高于 120 Hz 的图像刷新率可视化二维（2-D）数据（2000 个 A 扫描，每个光谱 2048 个像素）。以每秒约 9 个容积的速度对 100×100 A 扫描数据进行 3-D 成像。我们描述了软件架构、线程组织和优化。展示了在流动体模和人眼实时成像过程中记录的屏幕截图。
