We demonstrate a high-throughput computation-efficient snapshot coherence tomographic imaging method by combining interferometric coding and compressive sampling. We first encode the depth distribution of a three-dimensional (3D) object into the spectrum of a light field, using the principle of optical coherence tomography (OCT), i.e., through a Michaelson interferometer, which generates an intermediate (x,y,λ)(x, y, \lambda) data-cube that encodes the raw (x,y,z)(x, y, z) data of the object. We then sample the spectral data using a well-established compressive spectral imaging technique, called the coded aperture snapshot spectral imaging (CASSI), which yields a compressed 2D (x,y)(x, y) measurement that captures the whole 3D tomographic information of the object. Finally, a developed iterative algorithm and end-to-end deep learning network are used for tomographic reconstruction from the single 2D measurement. Such integration of OCT and CASSI leads to a physically simple and computationally efficient system, allowing us to implement a large data size of more than 2000×20002000\times 2000 pixels in the transverse dimensions and up to 200 pixels (depth slices) in the axial dimension. Owning to the interferometry-based depth sensing mechanism, we achieve a high axial resolution of up to 13μm13\,\mu m within an axial field of view of 1.6mm\text{1.6}\,mm. Video-rate visualization of dynamic 3D objects at micrometer scale are shown through several examples.

中文翻译：

---

快照相干断层成像


我们通过结合干涉编码和压缩采样，展示了一种高通量计算效率的快照相干断层成像方法。我们首先使用光学相干断层扫描（OCT）原理，即通过迈克尔逊干涉仪，将三维（3D）物体的深度分布编码到光场的光谱中，该干涉仪生成中间（x，y， λ)(x, y, lambda) 数据立方体，对对象的原始 (x,y,z)(x, y, z) 数据进行编码。然后，我们使用成熟的压缩光谱成像技术（称为编码孔径快照光谱成像 (CASSI)）对光谱数据进行采样，该技术产生压缩的 2D (x,y)(x, y) 测量值，捕获整个 3D 断层摄影信息的对象。最后，使用开发的迭代算法和端到端深度学习网络从单个 2D 测量中进行断层扫描重建。 OCT 和 CASSI 的这种集成带来了一个物理简单且计算高效的系统，使我们能够实现横向尺寸超过 2000×20002000\times 2000 像素和轴向尺寸高达 200 像素（深度切片）的大数据量方面。得益于基于干涉测量的深度传感机制，我们在 1.6mm\text{1.6}\,mm 的轴向视场内实现了高达 13μm13\,\mu m 的高轴向分辨率。通过几个示例展示了微米级动态 3D 对象的视频速率可视化。