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.

Phase Imaging through Scattering Media Using Incoherent Light Source

Automatic Phase Recognition Method Based on Convolutional Neural Network

Optimal Phase Mask Design for Coherent Modulation Imaging by Deep Learning

Untrained Neural Network Enabling Fast and Universal Structured-Illumination Microscopy

An S-CNN-based phase conjugation method in imaging through random media

Intensity and phase imaging through scattering media via deep despeckle complex neural networks

Recognizing three-dimensional phase images with deep learning

Object Recognition Through Scattering Media Using Convolutional Neural Network

Real-time 3D Nanoscale Coherent Imaging via Physics-aware Deep Learning

Single‐Shot Recognition of 3D Phase Images with Deep Learning

Learning-based Phase Imaging Using a Low-Bit-depth Pattern

Imaging through a Scattering Medium under Different Intensities of Ambient Light Interference

AI-enabled high-resolution scanning coherent diffraction imaging

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

Three-dimensional coherent X-ray diffraction imaging via deep convolutional neural networks

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

Phase function estimation from a diffuse optical image via deep learning

Deep Phase Decoder: Self-calibrating phase microscopy with an untrained deep neural network

Computational Optical Sectioning by Phase-Space Imaging with an Incoherent Multiscale Scattering Model

Phase conjugation with spatially incoherent light in complex media