Reproduce NanoJ eSRRF results with NanoPyx

[arthursw]

Hello,

I try to reproduce a Fiji eSRRF computation with NanoPyx but I get different results.

Here is the image that I use:

You can download and unzip it:
C1-Bio-01.tifMod.tif.zip

Here is the NanoJ code I use in Fiji:
run("eSRRF - Analysis", "magnification=5 radius=3 sensitivity=3 #=200 vibration avg");

(I also tried run("eSRRF - Analysis", "magnification=5 radius=3 sensitivity=3 #=200 vibration avg wide-field #_0=25 "); but it does not change much).

And here is the resulting image:

And with this python code:

from nanopyx.methods import eSRRF
from nanopyx.core.transform.sr_temporal_correlations import calculate_eSRRF_temporal_correlations
import tifffile
from skimage import io

with tifffile.TiffFile('path/to/C1-Bio-01.tifMod.tif') as tif:
    dataset_original = tif.asarray()

frames_per_timepoint = 200

for i in range(dataset_original.shape[0] // frames_per_timepoint):

    block = dataset_original[i * frames_per_timepoint : (i + 1) * frames_per_timepoint]
    result = eSRRF(block, magnification=5, radius=3, sensitivity=3) # _force_run_type='_run_unthreaded'
    io.imsave(f"path/to/results/esrrf{i}.tif", result)
    esrrf_temporal_correlation = calculate_eSRRF_temporal_correlations(result, "AVG")
    io.imsave(f"path/to/results/esrrf{i}_temporal_correlation.tif", esrrf_temporal_correlation)

This is the result I get:

I tried on a mac M1 and linux with a GPU.

I also tried _force_run_type='_run_unthreaded' to avoid GPU problems (the pixel grid pattern makes me think of an misalignment issue) but I get the same result.

Do you have any idea why I get different results? Did I miss anything?

[brunomsaraiva]

Hi @arthursw,

Thanks for the report, I've tried running your image and was able to reproduce the issue with odd numbered magnification (those vertical lines), with an even numbered the vertical line artifact disappears (see image below with magnification set to 6), so we likely introduced a bug there.

The NanoPyx implementation also does not automatically correct for drift misalignment as the original NanoJ version. If you want to perform the alignment you can still do it by calling NanoPyx drift alignment.

from nanopyx.methods.drift_alignment import estimate_drift_alignment

aligned_img = estimate_drift_alignment(img_to_align)

One other thing, we've made several improvements to eSRRF calculation in the NanoPyx implementation, meaning the "best" parameters for eSRRF are not gonna be a 1 to 1 with the NanoJ ones. In any case there seems to be an excessive amount of grid patterning in the generated image which should be fixed.

I'll work on a fix and will let you know once we've made a release with the bug fixes.

Best,
Bruno

[arthursw]

All right, thank you very much for this very quick and detailed answer :) 👍

I'll try a magnification of 6 and play with parameters.

Best

[brunomsaraiva]

Hi @arthursw,

We've just released a new version of NanoPyx (v1.2.3) that fixes the issue with the odd numbered magnification.
To install the new version you can simply call:
pip install --upgrade nanopyx
and that should bring you to the newest release.

We've also changed how eSRRF works from nanopyx.methods, it now contains arguments to perform time projection on the method call. This means your previous script can be simplified to this:

from nanopyx.methods import eSRRF
import tifffile
from skimage import io

with tifffile.TiffFile('path/to/C1-Bio-01.tifMod.tif') as tif:
    dataset_original = tif.asarray()

out = eSRRF(magnification=5, radius=3, sensitivity=3, frames_per_timepoint=200, temporal_correlation="AVG")
io.imsave("path/to/results/esrrf_temporal_correlation.tif", out)

This should fix the problem. I'll close the issue but feel free to re-open if you encounter any issue :)

Best,
Bruno

[arthursw]

This is great! You not only solved the problem, but also greatly simplified my code! Thanks :) 👍 !

[arthursw]

I tried your new version with my image (uploaded above) and the code you provided.

It works, but I still get a lot of grid patterning, as you can see:

Do you have the same behavior on your machine?

[arthursw]

The grid pattern might come from the fact that the input image was thresholded (all values below 600 were set to 0).

[brunomsaraiva]

Hi @arthursw,
I can reproduce the issue on my end, I'll dive a bit more into the macro pixel correction to see if I spot any bugs and/or potential changes to solve the issue.
I'll let you know once I have an update

[arthursw]

Thanks!

[arthursw]

Hi, is there any update on this issue? The biologists I work with want me to step back and reuse NanoJ-eSRRF instead of NanoPyx because of this problem.

[brunomsaraiva]

Hi @arthursw, we are now testing a rollback to some of the changes we did to eSRRF that should sort out the issue until we have a better way of correcting the macro pixel patterning with the latest changes.
We have it currently as a PR and should have a release today or tomorrow, but I'll ping you once we have it

[brunomsaraiva]

While we finish the multi-platform testing, here is a preview of the eSRRF output with the upcoming version:

And a side-by-side comparison of NanoJ (left) vs NanoPyx (right):