= Definition =
> In mathematics, deconvolution is an algorithm-based process used to reverse the effects of convolution on recorded data. The concept of deconvolution is widely used in the techniques of signal processing and image processing [[ https://en.wikipedia.org/wiki/Deconvolution | [wikipedia]]].
{F2292624, name="Raw/Deconvolved" , layout=center, width=512, alt="Raw / Deconvolved"}
= Point spread function=
> The point spread function (PSF) describes the response of an imaging system to a point source or point object [[ https://en.wikipedia.org/wiki/Point_spread_function | [wikipedia]]].
{F2292671, layout=center}
= Benefits =
- Denoising (useful for live imaging)
- Increase in resolution
- Tends to simplify image analysis pipelines
= Uses =
- Live samples: Allows to acquire noisier images (less laser power, more speed) and still be able to perform image analysis.
- Fixed samples: Allows for more precise localization of small structures, beneficial in [[../colocalization]] experiments for example.
- Noise removal: Makes most image analysis pipelines more robust, as the presence of noise scatters your results
- Widefield samples: Allows for 3D reconstructions from widefield samples.
= Requirements =
{F7264397,size=full}
Please consult the detailed [[ https://c4science.ch/w/bioimaging_and_optics_platform_biop/teaching/good_practices/ | Good Practices ]] page which details the notes below.
==Intensity range ==
(IMPORTANT) **DO NOT SATURATE your images !**
== Pixel (Voxel) Size ==
(NOTE) An Optimal deconvolution requires good sampling. Consult the [[ https://svi.nl/NyquistCalculator | SVI Calculator]] to match your acquisition settings to the deconvolution requirements.
(WARNING) Note that if the values are really too small for your sample (ie you observe bleaching during the acquisition), you can increase the voxel size given by SVI in XY by roughly a factor of 2 (closer to Nyquist sampling) without loosing too much deconvolution benefit. Try not to compromise the Z sampling if possible.
== 2D or 3D ==
(NOTE) Deconvolution is extremely powerful in 3D. In 2D you will mostly take advantage of the denoising effect, but not necessarily of the increase in resolution that deconvolution can offer.
(IMPORTANT) add explanatory image ?
=== Range ===
Deconvolution tends to generate edge artifacts if there are still intensities there (tissue samples, incomplete sample Z acquisition, ... ), so you should acquire:
- from Black to Black
(IMPORTANT) ADD images (XZ reslice to show incomplete vs complete Z and show images of edge artifacts
or
- plan to CROP a couple of slices at the top and bottom of your stack and also in XY
== Noise ==
(NOTE) Do not worry too much about noise. Deconvolution is a noise removal technique !
If you observe bleaching, you can decrease the number of averaging steps or the laser power and accept more noise, then let deconvolution work it out.
== Bleach ==
(IMPORTANT) TBD
== Confocal case optical slice matching==
(IMPORTANT) TBD
== Sample Refractive Index Matching ==
Deconvolution can help correct for spherical aberrations (PSF changing over the depth of the sample) in deep samples, but needs optimal sampling.
To avoid the effects of refractive index mismatch:
- Work close to the coverslip
- Choose mounting media and immersion that matches your sample (oil objectives for fixed samples, water objectives for Having a good refractive index match is
= Tools=
== Huygens Remote Manager ==
The BIOP chosen solution is [[ https://hrm-biop.epfl.ch/ | Huygens Remote Manager]] and you can follow this [[ https://c4science.ch/w/bioimaging_and_optics_platform_biop/image-processing/deconvolution/hrm/ | tutorial]].
== Deconvolution Lab ==
Deconvolution Lab [[ http://bigwww.epfl.ch/deconvolution/ | download]] and more (dataset, litterature) about deconvolution.
== CUDA deconvolution ==
[[ https://c4science.ch/w/bioimaging_and_optics_platform_biop/image-processing/deconvolution/cuda-deconvolution/ | tutorial ]]
TO BE CONTINUED...