Fig. 1: Principles of progressive optimization of an integrated microscope.
From: Large depth-of-field ultra-compact microscope by progressive optimization and deep learning
a Progressive optimization pipeline for a high-performance integrated microscope. In the first step, the integrated microscope consisting of plastic lenses is optimized with field of view (FOV) and chromatic merits through a canonical ray-tracing approach. In the second step, a diffractive optical element (DOE) containing cubic phase distribution is inserted in the front of the integrated microscope, and the system is further optimized for consistent modulation transfer function (MTF) across a 300-µm depth range. For every amplitude of the cubic phase (“Methods”), a corresponding lens system is optimized, and multiple candidate models are formed. In the third step, we separately train a deep neural network to retrieve clear images from captures of each of the models, and select the one yielding the best quality as the final optical design. As a comparison, directly optimizing all surfaces with deep optics algorithms needs 16 million calculation grids in an aperture of 4 mm and a feature size of 1 µm for each surface, consuming over 600 GB of memory. On the other hand, our progressive but systematic optimization can be finished in a desktop-level computer with 20 GB of memory. b Wireframe sketches of aspherical surfaces and DOE (here is a cubic phase plate) in the proposed integrated microscope. Irregular surfaces in our integrated microscope help achieving superior performance compared to spherical surfaces (Supplementary Fig. 1). c Modulation phase of the DOE in the integrated microscope (right) and corresponding surface fluctuations across the red dashed line (left). The valid area of DOE is lower than the surface of the surrounding glass to protect the component. d MTF characteristics of coded (with DOE) and uncoded (without DOE) PSFs in the focal depth (z = 0 µm) and defocused depth (z = 150 µm). e 3D PSF in the center of FOV across 300 µm depth. The maximum intensity projection (MIP) from x, y, and z axis are plotted below. f Spatial and frequency plots of the PSF without DOE and with DOE through different depths. White circles show a valid frequency range. g Strehl ratio across different FOV and depth of field (DOF) of uncoded microscope (without DOE, left) and coded microscope (with DOE, right). Red solid lines in each panel present the normalized FOV-averaged Strehl ratios. The colormap range has been adjusted for better visibility. h 3D rendering of the proposed integrated microscope. i The DOF changes with optical resolution of commercial 1×, 2×, 5×, 10×, and 20× objectives (blue). The proposed integrated microscope achieves high resolution that is comparable to the commercial 5× objectives but with 10 times increased DOF (green) and orders of magnitude smaller size.
