SPITFIRe

Contents

• Overview
• Repo Contents
• System Requirements
• Installation Guide
• Demo
• Runtime
• License
• Issues
• Citation

Overview

Modern fluorescent microscopy imaging is still limited by the optical aberrations and the photon budget available in the specimen. A direct consequence is the necessity to develop flexible and ”off-road” algorithms in order to recover structural details and improve spatial resolution, which is critical when restraining the illumination to low levels in order to limit photo-damages. We present a flexible method designed to restore 2D-3D images and videos, and subtract undesirable out-of-focus background. We assume that the images are sparse and piece-wise smooth, and are corrupted by mixed Poisson-Gaussian noise. To recover the unknown image, we consider a novel sparsepromoting regularizer defined as a mixed norm which gathers image intensity and spatial second-order derivatives. The resulting restoration algorithm named SPITFIR(e) utilizes the primal-dual optimization principle for fast energy minimization. Experimental results in various microscopy modalities from wide field up to lattice light sheet demonstrate the ability of the SPITFIR(e) algorithm to efficiently reduce noise, blur, and out-of-focus background, while avoiding the emergence of deconvolution artifacts.

Repo Contents

We provide 2 implementations of SPITFIRe. One is the original c++ implementation and works for 2D, 3D and 4D images. The second is a python pyTorch based implementation.

SPITFIRe C++

A c++ implementation of SPITFIRe is available in the SimgLib repository as a stand alone c++ command line interface. It is implemented for CPU and for GPU (cuda).

SPITFIRe python

A SPITFIRe python implementation using the pyTorch library is meant to provide an easy to use and modify full python implementation of the SPITFIRe regularizer. It is available in the sdeconv package.

System Requirements

Hardware Requirements

The SPITFIRe algorithm requires only a standard computer with enough RAM proportional to the input data size. Most of the experiment we made where done on a computer with about 8 GB of RAM. For optimal performance, we recommend a computer with the following specs:

RAM: 16+ GB
CPU: 4+ cores, 3.3+ GHz/core

The cuda SPITFIRe version needs a NVIDIA graphic card and the cuda drivers.

Software Requirements

OS Requirements

The SPITFIRe software is compatible with Windows 10, MacOS and Linux operating systems. The developmental version of the package has been tested on the following systems:

• Linux: Ubuntu 21.04
• Mac OSX: Mac OS Catalina 10.15.11
• Windows: 10

Installation Guide

Depending on the needed version they are several methods to install SPITFIRe. We recomand using the napari plugin or the sdeconv python library since it is available in PyPI

Using napari

The easiest method to install the napari plugin is using Conda.

conda create -y -n napari-env -c conda-forge python=3.9 pip
conda activate napari-env
python -m pip install "napari[all]"
python -m pip install napari-sdeconv

Napari can then be started with:

conda activate napari-env
napari

and SPITFIRe is in the plugin menu

Using the python version

The python primal-dual version can be installed using Conda

conda create -y -n spitfire -c conda-forge python=3.9 pip
conda activate spitfire
python -m pip install sdeconv

The SPITFIRe python class can then be used as follow:

from sdeconv.data import celegans
from sdeconv.deconv import Spitfire, SPSFGaussian

# load the blurry image
image = celegans()

# create a PSF
psf_generator = SPSFGaussian((1.5, 1.5), (13, 13))
psf = psf_generator()

filter_ = Spitfire(psf, weight=0.6, reg=0.995, gradient_step=0.01, precision=1e-7, pad=13)
deconv_image = filter_(image)

Using the c++ CLI

The C++ implementation of SPITFIRe is provided with a Command Line Interface.

A pre-compiled package is provided in the conda package simglib:

conda install -c sylvainprigent simglib

To compile SPITFIRe from C++ source code, cmake is needed. For people using windows, we recommand using Conda to manage cmake and the dependencies (libtiff, fftw3) and the visual compiler. The main step are (in unix command):

git clone https://github.com/sylvainprigent/simglib
cd simglib
mkdir build
cd build
cmake .. -DSL_USE_OPENMP=ON -Dsimglib_BUILD_CUDA=ON -Dsimglib_BUILD_TOOLS=ON
make
make install

Then the SPITFIRe CLI tool can be called as follows for the CPU version:

simgspitfiredeconv2d -i blurry_image.tif -o deblurred_image.tif -sigme 1.5 -regularization 12 -weighting 0.6 -niter 200

and as follow for the GPU version:

simgcuspitfire2ddeconv -i blurry_image.tif -o deblurred_image.tif -sigma 1.5 -regularization 12 -weighting 0.6 -niter 200

Demo

The demo using the graphical interface can be found in the napari plugin page

Run time