FRCnet

This repository contains the code and data for

Fourier Ring Correlation and anisotropic kernel density estimation improve deep learning based SMLM reconstruction of microtubules

by Andreas Berberich, Andreas Kurz, Sebastian Reinhard, Torsten Johann Paul, Paul Ray Burd, Markus Sauer, Philip Kollmannsberger

This work was part of the MSc thesis project of Andreas Berberich in the Computational Image Analysis group of Philip Kollmannsberger at the Center for Computational and Theoretical Biology of the University of Würzburg, carried out between October 2018 and March 2020.

This repository contains the following files:

• The notebooks Fig{1,2,3,4}.ipynb reproduce the figures in the manuscript.
• FRCNet.ipynb contains the code to train the neural network.
• frc_loss.py and aniso_kde.py contain the implementation of the tensorflow FRC loss and the anisotropic KDE filtering.
• The /data directory contains training/validation data and the data to generate the figures.

All code was written in Python3 using numpy, scipy, scikit-image, pandas, matplotlib and tensorflow 2.4.1. If you find this code useful and want to use it in your own project, please cite our paper.

To run the notebooks, create a new conda environment and install the dependencies:

conda install -c anaconda tensorflow-gpu=2.4.1 jupyter matplotlib pandas scikit-image

and then install tensorflow-addons (required for on-GPU image rotation) using pip:

pip install tensorflow-addons

The raw localization data and trained models can be downloaded from here and should be placed in the /data folder.