Dynamic Structured Illumination Microscopy with a Neural Space-time Model

Project Page | Paper | Experimental Data

Prerequisite

• CUDA 11.X (JAX distribution is CUDA version-specific)
• cuDNN
• Anaconda

Setup

Clone this project

git clone https://github.com/Waller-Lab/SpeckleFlowSIM.git

Set up & activate virtual env

conda create -n virtualenv_name python=3.9
conda activate virtualenv_name

Install dependencies

pip install https://storage.googleapis.com/jax-releases/cuda111/jaxlib-0.1.72+cuda111-cp39-none-manylinux2010_x86_64.whl
pip install -r requirements.txt  # install the rest of env via pip
conda install -c conda-forge jupyterlab nodejs ipympl  # for visualization

Install the in-house library

git clone --branch v0.0.1 https://github.com/rmcao/CalCIL.git
cd calcil
pip3 install -e .

Download experimental data

Download the data from Google Drive and place it under the project folder.

Open Jupyter

$ jupyter lab --no-browser --port=8899

Try Speckle Flow SIM on Jupyter notebooks

simulation.ipynb: simulation reconstruction on a dynamic Shepp-Logan phantom.

experiment.ipynb: experimental reconstruction on a absorptive USAF-1951 resolution target.

Folder structure

├── checkpoint          : folder to store model checkpoints
├── README.md           : README file
├── simulation.ipynb    : notebook for Speckle Flow SIM simulation
├── experiment.ipynb    : notebook for Speckle Flow SIM experiment
├── experiment.npz      : experimental data
├── requirement.txt     : dependencies to install
├── spacetime.py        : implementation of the neural space-time model
├── speckle_flow.py     : incorporating Speckle SIM forward model with neural space-time model for Speckle Flow SIM
└── utils.py            : utility functions for motion and dynamic scene generation.

Citation

@article{cao2022dynamic,
  title={Dynamic Structured Illumination Microscopy with a Neural Space-time Model},
  author={Cao, Ruiming and Liu, Fanglin Linda and Yeh, Li-Hao and Waller, Laura},
  journal={arXiv preprint arXiv:2206.01397},
  year={2022}
}