File_format

[helvince]

Apologies in advance for the elementary question.

I am working with 3D data. The only way to process the data is through the featurize_directory, requiring a file_format input. Can you elaborate what this should be?

Copying “.*” (default for 2D) and naming mask and intensity files equivalantly “1.tif” in their respective directories did not work. After searching the documentation for explanation/examples and trying a few variations on what I found I still have not gotten it to work.

I look forward to hearing from you.

Kind regards,
Vincent

[friskluft]

Hi Vincent,

In order to process a dataset as 3-dimensional, you need to
(1) use the command-line version of Nyxus
(2) specify command line parameter --dim=3
(3) specify command line parameter --filePattern="file pattern matching your dataset".

Currently, Nyxus can read 3D datasets organized as a collection of image files representing z-slices. A z-slice index needs to be encoded in the file name in some uniform way, for example, s_k1.tiff, s_k2.tiff, ... s_k500.tiff . Note the pattern: "whatever k integer-number .tiff" where part _k followed by a number.

In order to tell Nyxus how to interpret file names of a 3D image collection, we use command line parameter --filePattern. Continuing our example, we tell Nyxus how to assemble the 3D shape from slices as --filePattern="{c+}_k{set d+}.tiff" . Here, {c+} is a placeholder of non-numeric characters, "_k" - a constant character sequence, {set d+} - a sequence of digits that should be interpreted by Nyxus as a unique z-slice identifier, and ".tiff" - constant character sequence.

Practical example: a collection of multi-channel slices whose z-indices are encoded as "_zDDD"

BRATS_001_z000_t000.ome.tif
BRATS_001_z000_t001.ome.tif
BRATS_001_z000_t002.ome.tif
BRATS_001_z000_t003.ome.tif
BRATS_001_z001_t000.ome.tif
BRATS_001_z001_t001.ome.tif
BRATS_001_z001_t002.ome.tif
BRATS_001_z001_t003.ome.tif
BRATS_001_z002_t000.ome.tif
BRATS_001_z002_t001.ome.tif
BRATS_001_z002_t002.ome.tif
BRATS_001_z002_t003.ome.tif
BRATS_001_z003_t000.ome.tif
BRATS_001_z003_t001.ome.tif

can be read with --filePattern="{c+}_{d+}_z{set d+}_t{d+}.ome.tif"

Will this z-slice based layout work for you?

My regards
=Andriy

[helvince]

Thank you very much Andriy for your for the quick response to such an elementary question.

I adapted to the multi-file ome.tiff which now also makes a lot more sense and ran "3D_ALL" features successfully. The 3D_ALL_INTENSITY argument was not valid, but the overhead is no real problem. I see the 3D measurements are a quite recent addition and will keep an eye out on the developments.

I still have some questions if you would be so kind to answer:

• Is it possible to specify anisotropic pixel size?
• Are there any shape/geometry features planned?
  Common volume/surface area features are absent. Or is this excluded by design and would you have any recommendations?
• Would you advice normalization as a pre-processing step? And if so what would be the optimal format (float32, uint16 or...)?

Best wishes,
Vincent

[friskluft]

Hello Vincent,

1. I'll add pixel anisotropy on the coming week and would appreciate getting your feedback ;
2. I plan to add volume and surface features then ;
3. Do you mean intensity normalization before extracting features? If the intensity varies significantly between z-slices, I would suggest normalizing it. Preferred format is uint16 or uint32.

Regards,
=Andriy

[helvince]

Great to hear! I will keep an eye out and report back my experience/struggles.

[helvince]

Ha Andriy,

Am I correct that the new version has not been pushed? Getting anxious to try your new additions :)

Kind regards,
Vincent

[friskluft]

Hi Vincent,

We are in progress testing the update before releasing it.
I will post a message in this thread when the update is available.
Thank you for your interest for Nyxus! : )

Cheers,
=Andriy

[sameeul]

@helvince : we have released a new version a few weeks back. Can you please give it a try and let us know?

[friskluft]

In that release, we added support of NIFTI format, so there is no need to convert a volume file to a "layout-A" dataset discussed above. NIFTI files are processed in a straightforward way, for example:

import nyxus
nyx = nyxus.Nyxus3D (["*3D_ALL*"])

# dataset
idir = "/dataset1/input"
mdir = "/dataset1/masks"

# selecting only uncompressed NIFTI files
features1 = nyx.featurize_directory (idir, mdir, file_pattern=".*\.nii")

# selecting only compressed NIFTI files
features2 = nyx.featurize_directory (idir, mdir, file_pattern=".*\.nii\.gz")