In this talk, I will present a 3D snapshot compressive sensing microscope capable of imaging dynamic live cells in all 3 dimensions with around 100 nm resolution at 25 frame per second.
 
First, I will present Multifocal Plane Microscopy (MFM), an optical design that captures live cells’ 3D intensity information as a single “compressed” 2D image (Figure 1). For comparison, classic 3D imaging systems such as confocal microscopy require lengthy z-scan to measure each x-y plane intensity, which fails on moving objects. Our MFM system doesn’t need z-scanning, but rather encodes the xyz intensity information as a single 2D snapshot, enabling dynamic scene capture without temporal skew.
 
Next, I will describe our compressive sensing technique to recover the 3D volumetric object (Figure 2) from its 2D MFM image (Figure 1). The reconstruction of 3D from 2D is a large-scale ill-posed inverse problem, as it requires recovery of around 10 million unknowns from 1 million measurements. We apply a sparsity prior and non-negativity physical constraint on the volumetric intensity, then model and solve the reconstruction as a constrained optimization problem.
 
In the end, I will discuss 4D joint spatial-temporal reconstruction of MFM image sequences, and compare it with 3D tomography with uncalibrated rotation and translation. I will also talk about future work on using dictionary learning and deep learning to train the empirical basis from the data for better sparse representation.