光学相干层析(OCT)技术在活体成像应用中的无损、高速、超高分辨率特性使其在生物医学领域有着广阔的发展空间。通常情况下,OCT系统的数据采集量巨大,图像重建中包含的快速傅里叶变换(FFT)需要大量的计算时间,中央处理器(CPU)串行数据处理模式难以满足实时成像的需求。针对这一问题,将统一计算设备架构(CUDA)并行编程技术应用到皮肤组织成像的谱域相干光断层成像(SD-OCT)系统数据处理过程中,并在图形处理器(GPU)上予以实现。详述了系统算法并行化拆分以及对系统采集到的数据进行并行化处理等以提高成像速度的方法。利用搭建的SD-OCT系统对手指部位的皮肤组织进行成像并采集数据,用实验室现有数据处理平台MATLAB以及GPU分别对采集到的数据进行处理,对比了不同数据处理平台的成像速度和成像质量。结果表明,在保证成像质量不变的前提下,GPU+CPU混合编程技术比MATLAB数据处理平台的成像速率提高了10倍,满足了临床中对实时成像的实际要求。

The non-damage, high speed, and ultra-high resolution characteristics of optical coherence tomography (OCT) technology in in vivo imaging application make it a broad space for development in the biomedical imaging field. Generally, a large amount of OCT data are acquired , and fast Fourier transform (FFT) in the image reconstruction process requires much calculation time. Therefore, the traditional central processing unit (CPU) serial data processing mode is difficult to meet the requirements of real-time imaging. To this end, the compute unified device architecture (CUDA) parallel programming technique is applied to the data processing of skin tissue imaging in the spectral-domain optical coherence tomography (SD-OCT) system, and it is implemented in the graphics processing unit (GPU). We detail the parallelization of the system algorithm and the parallel processing of the data collected by the system to improve the imaging speed. In the experiment, we use the SD-OCT system to image the skin of the finger and collect the data. The data collected by the laboratory's existing data processing platform MATLAB and GPU are processed respectively, and the imaging speed and image quality of different data processing platforms are compared. The experimental results show that the GPU and CPU hybrid programming model has a processing speed up to 10 times faster than the CPU-based MATLAB method while the image quality remains the same, which meets the clinical requirement of real-time imaging in the clinic.