This guide walks through 3D deconvolution of a microscopy z-stack using the C. elegans embryo dataset from EPFL. It covers generating a 3D PSF from physical parameters as well as loading a measured PSF from file.
Download one channel of the C. elegans embryo dataset from https://bigwww.epfl.ch/deconvolution/bio/:
Extract both the data and PSF zip files. This tutorial uses the DAPI channel but any of the three works.
Drag and drop CElegans-DAPI folder onto the input viewer, or use File > Open Image Folder.
The image appears in the input viewer. Use the frame slider on the top left to scroll through z-slices.
Adjust the viewer:
- In the viewer toolbar (right of the output viewer), uncheck Show Patch Grid (we will later set the patch grid to 1 x 1; the grid overlay is not useful with a single patch)
- Ensure Sync Views is checked so input and output viewers share the same orientation and zoom
- Enable View > Cross-Section Viewer to see XZ cross-sections of input and output side by side, this is very useful for inspecting the axial resolution improvement after deconvolution
Set the patch grid to 1 x 1:
In the sidebar (below the input/output viewers), select the Patch Grid tab. Set both Columns and Rows to 1. This way one single PSF is used for the entire volume. If you have image data with spatially varying blur, you would use a grid of patches and use a different PSF for each. For this tutorial, we keep it simple with a single patch.
Disable auto-deconvolution: Switch to the Deconvolution tab. Uncheck Auto-deconvolve on changes. 3D deconvolution is computationally expensive and should be triggered manually.
Select the 3D algorithm: In the Algorithm dropdown, select Richardson-Lucy 3D.
Set iterations:
Set Iterations to something between 10 and 32.
In the sidebar on the left, find the Generator dropdown and select 3D PSF Microscopy.
A settings panel appears below the generator dropdown. Enter the acquisition parameters:
| Setting | Value |
|---|---|
| Wavelength (nm) | 477 |
| Numerical Aperture | 1.4 |
| Immersion Index | 1.518 |
| Z Step (nm) | 160 |
| Num Z Planes | 104 |
| XY Pixel Size (nm) | 64.5 |
Set Num Z Planes to match the number of z-slices in the data (104) in case it is not already auto-detected.
The PSF grid size can be adjusted in Extras > Settings under Grid Size. For the loaded dataset a grid size of 768 works well.
The 3D PSF preview at the bottom of the sidebar shows XY Slice and XZ Section views.
You can play around with the Zernike coefficients to add aberrations to the PSF. For this it is best to use the fast mode for XY Scaling.
Click the Deconvolve button. A progress dialog appears showing the current iteration. Now wait...
When complete, the output viewer shows the deconvolved volume.
Instead of generating a PSF from physical parameters, you can load a measured or precomputed PSF.
Extract the PSF zip file (e.g., PSF-CElegans-DAPI.zip). This produces a folder of TIFF files, one per z-slice.
In the sidebar, change the Generator dropdown to From File. Click Open Folder... and select the folder containing the extracted PSF TIFF files. The info panel below shows the detected PSF dimensions and file count.
With the PSF loaded, click Deconvolve to run 3D Richardson-Lucy deconvolution using the measured PSF.
- To better compare input and output, set the same display range for both viewers. For this, right click on the display range slider and select Fit to Output Frame. Then you still can manually adjust the display range by moving the handles on the slider.
- To better inspect the PSF you can right click on it and select Log scale
That's it! Have fun experimenting with 3D PSF generation and deconvolution!