ConfocalGN

Overview

ConfocalGN is a MATLAB program that can simulate confocal imaging. Synthetic confocal image stacks can serve as test images for an image analysis program, allowing one to compare the output of the analysis with the 'ground truth' provided to built the synthetic images. Other types of microscopes can be simulated by using adequate PSF

ConfocalGN starts from a ground truth specified as a

the positions of fluorophores OR
a 3D image matrix of high resolution. This image is pixelized, and convolved with a Gaussian approximation of the PSF of the confocal microscope. Noise is also added, but ConfocalGN can automatically extract the characteristics of the noise from a sample confocal image, to ensure that the simulated images possess the same noise characteristics as the real images. In addition, ConfocalGN may optionally include fluorophore stochasticity and background fluorescence.

Requirements

To use the confocal generator, the user must provide :

1 - A ground truth truth, either

1A - truth.source : a set of fluorophore positions (as a matlab matrix or saved in a text file). This is then converted into a high resolution image (a 3D matrix). This conversion may include fluorophore stochasticity and background fluorescence.

1B - truth.img an image (a 3D matrix or a TIFF file) The ground truth image can be a high resolution image. It must be accompanied by truth.pix, the physical size of ground truth pixels. By default the ground truth image should be isotropic, but non-isotropic images are possible, provided conf.pix and conf.psf are scaled accordingly.

2 - A parameter structure conf, containing : conf.pix, the confocal voxel size in physical units (a 3x1 vector) and conf.psf, the microscope 2-way point spread function. conf.psf can be

an image (matlab matrix or tiff file), with the same pixel size as the ground truth image
a 3x1 vector; containing the standard deviation of the PSF in each direction (in physical units) The PSF is then assumed then to be a 3D Gaussian built from these parameters. The PSF can be obtained from PSF simulating software(e.g. Huygens/Icy) or from analysis of experimental data (e.g. with Huygens/Mosaic)

3 - The noise to imitate, in sample

3A - A noise distribution sample.noise and a signal value sample.sig -- or

3B - A sample image (matrix or image name) sample.img for the program to derive sample.noise and sample.sig ConfocalGN segments the image (using a default or user-provided segmentation function). The pixels above the threshold are considered as signal, the others as background. sample.noise is a 3x1 vector of the 3 first moments of the background pixel values. sample.sig is the mean of the signal pixel values.

Options

Optionally, the user can also specify:

Sampling and segmentation options by editing the file cgn_options.m
Options to generate the ground truth image by editing gtm_options.m
Custom segmentation program by indicating it in "provide_image_mask.m" By default, provide_image_mask uses the function segment_image, that can be replaced with another equivalent segmentation method The segmentation program is used to recognize background from signal The replacing function must be of the format [img,mask] = segment_image(image,options) see segment_image for the definitions of img,mask,image,options
Custom convolution program by indicating it in "convolve_with_psf.m" By default, convolve_with_psf uses either convolve_with_gaussian_psf or fftconvn

Operating ConfocalGN

To operate ConfocalGN, open MATLAB and switch to the ConfocalGN directory.

run cgn_startup.m (this will update the MATLAB path) An illustrative example is provided in cgn_example.m
Fill up the required input conf must contain conf.pix and conf.psf. truth must contain a ground truth image (truth.img) or fluorophore coordinates (truth.source). sample must contain either noise and signal (sample.noise and sample.sig), or a sample image (sample.img).
Use confocal_generator

[res,truth,sample]=confocal_generator(truth,conf,sample);

With the output :

res.stacks : simulated image stack
res.offset : translation needed to align with IMG
res.sig : mean signal value in the simulated image
res.noise : noise distribution in the simulated image
res.img : post-segmentation image of stack

Example :

Open MATLAB Switch current directory to the ConfocalGN main directory In the MATLAB command window, run:

cgn_startup.m cgn_example.m

Credits & Licence

Code is distributed under GPL3.0 Licence (see LICENCE.md) This code includes the librairies :

gausss3filter Copyright (C) Max W.K. Law

Code modified from GNU Octave :

randg
gamrnd
postpad
fftconvn see tiffread.m ; gamma_random.m ; readtext.m ; saveastiff.m ; postpad.m ; fftconvn.m for licencing information.

This code was developped by Serge Dmitrieff in 2016 in EMBL, with the support of François Nédélec. http://biophysics.fr

Changelog

-June 2016 : initial minimal version -July 2016 : correction of stacking -July 2016 : major overhaul for easier use, and usage of sample image -May 2017 : implemented the use of fluorophore coordinates -May 2017 : implemented general PSF support -June 2017 : major code cleanup and improved comments -June 2017 : v1.1 release

Source for gauss3filter

Max W. K. Law and Albert C. S. Chung, "Efficient Implementation for Spherical Flux Computation and Its Application to Vascular Segmentation", IEEE Transactions on Image Processing, 2009, Volume 18(3), 596�V612

SergeDmi/ConfocalGN