Extreme volumetric imaging using single-shot optical tomography with compressive sensing

Standard approaches to 3D imaging typically rely on taking a series of 2D images one after another and reconstructing the 3D volume. This puts an upper limit on the volumetric frame rate achievable, since it requires a certain amount of time to record the 10s-1000s of individual 2D images separately. This project aims to develop an imaging platform capable of full 3D volumetric imaging at acquisition rates exceeding 1000 volumes per second by acquiring multiple images in parallel using cost-effective high-speed cameras.

The approach is based on the use of 4 high-speed cameras capable of frame rates in excess of 1000 frames per second. Previously, cameras that could operate at such high frame rates were prohibitively expensive, but recent advances in the technology has seen their price reduce significantly and brought them in line with the cost of more standard scientific cameras. Each camera will be combined with a novel optical system that produces two side-by-side images on the camera sensor, each viewing the sample at a slightly different angle. The four cameras will be placed around the sample and acquire eight images from different directions simultaneously, running at 1000s of frames a second. Each set of 8 images can them be combined to reconstruct the 3D scene, producing a 3D data set running at 1000s of volumes per second.

The particular arrangement of the cameras and the approach to reconstruction will depend on the sample being imaged. In this project we will apply this extreme volumetric imaging system to two experimental investigations. To image semi-transparent biological samples a technique analogous to X-ray computed tomography, called optical projection tomography (OPT), will be applied. While typical OPT data sets require 100s of angularly resolved images, by applying advances computational analysis techniques we will be able to reconstruct the volume from the eight images acquired. This will permit OPT imaging of samples that have not been immobilized (e.g. no need for anesthesia) and can measure very fast biological processes in 3D, such as transient cell signaling events.

While the use of high-speed cameras has been limited in biological imaging, they have been routinely used in fluid dynamics experiments to capture very fast events (e.g. droplet-surface interactions). We will extend this to full 3D imaging of such interactions by imaging them from eight different directions simultaneously and performing 3D reconstructions for every set of images (i.e. running at 1000s of volumes per second). For single droplet interactions we will employ surface measurement and reconstruction techniques, while for sprays (i.e. large number of droplets) we will employ particle tracking and optical scattering to quantify the trajectories and size of all the droplets. These experiments will be used to confirm and develop computational fluid dynamic simulations.