Dardikman Gili, Habaza Mor, Waller Laura, Shaked Natan T
Opt Express. 2016 May 30;24(11):11839-54. doi: 10.1364/OE.24.011839.
We suggest a new implementation for rapid reconstruction of three-dimensional (3-D) refractive index (RI) maps of biological cells acquired by tomographic phase microscopy (TPM). The TPM computational reconstruction process is extremely time consuming, making the analysis of large data sets unreasonably slow and the real-time 3-D visualization of the results impossible. Our implementation uses new phase extraction, phase unwrapping and Fourier slice algorithms, suitable for efficient CPU or GPU implementations. The experimental setup includes an external off-axis interferometric module connected to an inverted microscope illuminated coherently. We used single cell rotation by micro-manipulation to obtain interferometric projections from 73 viewing angles over a 180° angular range. Our parallel algorithms were implemented using Nvidia's CUDA C platform, running on Nvidia's Tesla K20c GPU. This implementation yields, for the first time to our knowledge, a 3-D reconstruction rate higher than video rate of 25 frames per second for 256 × 256-pixel interferograms with 73 different projection angles (64 × 64 × 64 output). This allows us to calculate additional cellular parameters, while still processing faster than video rate. This technique is expected to find uses for real-time 3-D cell visualization and processing, while yielding fast feedback for medical diagnosis and cell sorting.
我们提出了一种新的实现方法，用于快速重建通过层析相显微镜（TPM）获取的生物细胞三维（3-D）折射率（RI）图。TPM的计算重建过程极其耗时，使得对大数据集的分析速度慢得不合理，并且无法对结果进行实时三维可视化。我们的实现方法使用了新的相位提取、相位展开和傅里叶切片算法，适用于高效的CPU或GPU实现。实验装置包括一个连接到倒置显微镜的外部离轴干涉模块，该显微镜由相干光照明。我们通过微操作使单细胞旋转，以在180°角范围内从73个视角获得干涉投影。我们的并行算法是使用英伟达的CUDA C平台实现的，在英伟达的Tesla K20c GPU上运行。据我们所知，这种实现方法首次实现了对于具有73个不同投影角度（64×64×64输出）的256×256像素干涉图，三维重建速率高于每秒25帧的视频速率。这使我们能够计算额外的细胞参数，同时处理速度仍快于视频速率。预计这项技术可用于实时三维细胞可视化和处理，同时为医学诊断和细胞分选提供快速反馈。
