Abstract:Recovering 3D phase features of complex, multiple-scattering biological samples traditionally sacrifices computational efficiency and processing time for physical model accuracy and reconstruction quality. This trade-off hinders the rapid analysis of living, dynamic biological samples that are often of greatest interest to biological research. Here, we overcome this bottleneck by combining annular intensity diffraction tomography (aIDT) with an approximant-guided deep learning framework. Using a novel physics model simulator-based learning strategy trained entirely on natural image datasets, we show our network can robustly reconstruct complex 3D biological samples of arbitrary size and structure. This approach highlights that large-scale multiple-scattering models can be leveraged in place of acquiring experimental datasets for achieving highly generalizable deep learning models. We devise a new model-based data normalization pre-processing procedure for homogenizing the sample contrast and achieving uniform prediction quality regardless of scattering strength. To achieve highly efficient training and prediction, we implement a lightweight 2D network structure that utilizes a multi-channel input for encoding the axial information. We demonstrate this framework's capabilities on experimental measurements of epithelial buccal cells and Caenorhabditis elegans worms. We highlight the robustness of this approach by evaluating dynamic samples on a living worm video, and we emphasize our approach's generalizability by recovering algae samples evaluated with different experimental setups. To assess the prediction quality, we develop a novel quantitative evaluation metric and show that our predictions are consistent with our experimental measurements and multiple-scattering physics.

Robust single-shot 3D fluorescence imaging in scattering media with a simulator-trained neural network

Robust single-shot 3D fluorescence imaging in scattering media with a simulator-trained neural network

Untrained Neural Network Enabling Fast and Universal Structured-Illumination Microscopy

Universal and High-Fidelity Resolution Extending for Fluorescence Microscopy Using a Single-Training Physics-Informed Sparse Neural Network

Physical model simulator-trained neural network for computational 3D phase imaging of multiple-scattering samples

De-scattering Deep Neural Network Enables Fast Imaging of Spines Through Scattering Media by Temporal Focusing Microscopy

A Method Of 3d Light Field Imaging Through Single Layer Of Weak Scattering Media Based On Deep Learning

Wide-Field, High-Resolution Reconstruction in Computational Multi-Aperture Miniscope Using a Fourier Neural Network

Roles of scattered and ballistic photons in imaging through scattering media: a deep learning-based study

Deep learning-based scattering removal of light field imaging

Deep learning-assisted imaging through stationary scattering media

Single picture single photon single pixel 3D imaging through unknown thick scattering medium

3D Deep Learning Enables Fast Imaging of Spines through Scattering Media by Temporal Focusing Microscopy

Single-Shot 3D Widefield Fluorescence Imaging with a Computational Miniature Mesoscope

Advancing Computational Miniature Mesoscope with Simulator-based Deep Learning Reconstruction

Non-Invasive Imaging Through Scattering Media Based on Deep Learning

Imaging reconstruction through long-range scattering media by using deep learning

Deep Non-line-of-sight Imaging from Under-scanning Measurements.

Scalable Non-Invasive Imaging Through Dynamic Scattering Media at Low Photon Flux

Overcoming photon and spatiotemporal sparsity in fluorescence lifetime imaging with SparseFLIM

Computational Optical Sectioning with an Incoherent Multiscale Scattering Model for Light-Field Microscopy.