radxtools / collageradiomics Goto Github PK

I have a suggestion for the Linux/Mac installation. The documentation states:

Linux
Click here and follow the instructions to install Docker.

When you click on the link, it redirected me to Docker’s homepage. Is it supposed to redirect the user to Docker’s homepage or are there supposed to be step-by-step instructions?

Thanks!

When running the Docker image, my localhost does not display the Jupyter examples. Click here to see a screenshot of my localhost page.

Version Information
macOS 10.15.5
Docker 2.3.0.3
localhost port: 8888
Git 2.24.3 (Apple Git-128)

Steps to re-create problem
`Toms-Work-MacBook-Pro:Collageradiomics Tom$ git clone https://github.com/ccipd/collageradiomics.git

Cloning into 'collageradiomics'...
remote: Enumerating objects: 165, done.
remote: Counting objects: 100% (165/165), done.
remote: Compressing objects: 100% (65/65), done.
remote: Total 496 (delta 88), reused 155 (delta 85), pack-reused 331
Receiving objects: 100% (496/496), 1.88 MiB | 1.39 MiB/s, done.
Resolving deltas: 100% (246/246), done.

Toms-Work-MacBook-Pro:Collageradiomics Tom$ cd collageradiomics/
Toms-Work-MacBook-Pro:collageradiomics Tom$ ls -l

total 96
-rw------- 1 Tom staff 1463 Jun 29 16:05 LICENSE
-rw------- 1 Tom staff 26476 Jun 29 16:05 README.md
drwx------ 3 Tom staff 4096 Jun 29 16:05 docker
drwx------ 3 Tom staff 4096 Jun 29 16:05 jupyter
drwx------ 2 Tom staff 4096 Jun 29 16:05 module
drwx------ 2 Tom staff 4096 Jun 29 16:05 sample_data

Toms-Work-MacBook-Pro:collageradiomics Tom$ docker pull ccipd/collageradiomics-examples:latest

latest: Pulling from ccipd/collageradiomics-examples
Digest: sha256:896e3daabb23f5813ad271374a8d302283141fbb18afd0a29903c83e1903aae3
Status: Image is up to date for ccipd/collageradiomics-examples:latest
docker.io/ccipd/collageradiomics-examples:latest

Toms-Work-MacBook-Pro:collageradiomics Tom$ docker run -it -p 8888:8888 -v /root ccipd/collageradiomics-examples

[I 20:06:06.315 NotebookApp] Writing notebook server cookie secret to /tmp/notebook_cookie_secret
[W 20:06:06.473 NotebookApp] All authentication is disabled. Anyone who can connect to this server will be able to run code.
[I 20:06:06.477 NotebookApp] Serving notebooks from local directory: /root
[I 20:06:06.477 NotebookApp] The Jupyter Notebook is running at:
[I 20:06:06.477 NotebookApp] http://fc0e20af3a6c:8888/
[I 20:06:06.477 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).`

This is in regards to enhancements for https://collageradiomics.readthedocs.io

1. Should link to main github wiki readme page, to BrIC lab, and to their publication in overview.
2. "Calculates the collageradiomics algorithm." in a lot of docs, I see things like "This is the main entry-point to the Collage calculation system. You will likely want check out the execute() function [internal hyperlink]."
3. For the default values of 1/4*pi etc., in the docs we should say that the default values correspond to N/7*[Pi_symbol] for N=0...7 so people aren't confused about the decimals
4. For execute() I think we should state a bit more about the output, and what it expects to have set before calling the function
5. For from_multiple_images, I think we need to state explicitly the format of that numpy array
6. "Within mask coordinates" - I think we should dumb it down even more - say that it creates a rectangle at the x,y,w,h and only calculates collage inside that mask.
7. We should change the variable names to be consistent using "mask" instead of "patch"
8. We should remove the "mt" (if that was leftover from calling it "mahotas")
9. For collageradiomics.CollageCollection we should add more info on this at a high level
10. For DifferenceVarianceInterpretation, the formatting is off of the webpage
11. We should link to the haralick paper for DifferenceVarianceInterpretation and give more details as to the meaning. https://doi.org/10.1109/TSMC.1973.4309314
12. For these internal methods like bbox1, do they belong on the public readme? should we somehow exclude them from an outward-facing set of docs?

Instead of looping through the third dimension as slices, create the co-occurence matrix from the surrounding 3D voxels.

This will requires greycomatrix to be configured to process 3D images prior to passing the resulting matrix to the mahotas library for haralick calculations.

If the grey levels are 64, then the co-occurence matrix will still be 64x64, but will have pulled its data from surrounding voxels, not just surrounding pixels from the current slice.

Create unit tests or whatever python standard practice is for testing.

We should test it with invalid parameters being input, invalid datatypes, perhaps NaN's, etc.

/home/tothtech/code/collageradiomics-tothtech/module/collageradiomics.py in execute(self)
    829 
    830         haralick_feature_list = self.haralick_feature_list
--> 831         print(haralick_feature_list)
    832         if haralick_feature_list[0] == HaralickFeature.All:
    833             number_of_features = 13

IndexError: list index out of range

Also, doesn't appear to work for subsequent images after working past that error:

[<HaralickFeature.SumOfSquareVariance: 3>]
Calculating feature 4:
Calculated feature 4.
IMAGE:
width=440 height=596
Cropped Array Shape: (34, 33)
x = 208:251 (43)
y = 253:297 (44)
PaddedCropped Array Shape: (44, 43)
dx shape = (44, 43)
dominant angles shape = (34, 33)
svd radius = 5
svd diameter = 11
dx windows shape = (34, 33, 11, 11)
Center x: range(0, 33), Center y: range(0, 34)
x=8, y=0
angle=0.96076213752155
x=28, y=2
angle=1.0888919954206522
x=29, y=24
angle=0.702823824690135
Done calculating dominant angles.
0
[]

Core module should have minimum dependencies, such as mahotas and numpy, and only meant to input a numpy array and output numpy array.

The more full-featured module can have inputs for various image types, etc.

0 - Rectangular
1 - Non-rectangular mask
2 - Non-rectangular mask, multiple slices

Related to Issue Toth-Technology/collageradiomics#12, we can have nested docker images.

The core one would be to play around with our core pip module and the outer docker(s) could have playing around with image input, and potentially even a separate one for jupyter examples/visualizations (e.g. using matplotlib).

Before we break up into separate docker images (Issue #12), we should include the SimpleITK pip module on our test docker.

JUPYTER_RUNTIME_DIR=/tmp

Must loop through all windowed regions inside a masked patch and calculate the dominant angle from SVD and create a new dominant angle image.

I got an error as: "Non-contiguous masks are not supported" when the mask values were binary-0 and 1.
When I changed the values to 0 and 255, the code worked!

System Info

• Windows 10 64-bit
• Python 3.8.2
• Mahotas 1.4.10

Error

>>> import collageradiomics
Could not import submodules (exact error was: DLL load failed while importing _bbox: The specified module could not be found.).

There are many reasons for this error the most common one is that you have
either not built the packages or have built (using `python setup.py build`) or
installed them (using `python setup.py install`) and then proceeded to test
mahotas **without changing the current directory**.

Try installing and then changing to another directory before importing mahotas.
>>> import collageradiomics
>>>

Diagnosis

The problem seems to be with the Mahotas import, since calling import mahotas gives the same error (and has the same resolution).

The output of the dir command shows a limited install of Mahotas:

>>> dir(mahotas)
['', 'all', 'builtins', 'cached', 'doc', 'file', 'loader', 'name', 'package', 'path', 'spec_', 'citation', 'citation_text', 'e', 'short_citation_text', 'stderr', 'sys']

The expected output is:

>>> dir(mahotas)
['__all__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '__version__', '_bbox', '_center_of_mass', '_convex', '_convolve', '_distance', '_filters', '_histogram', '_labeled', '_morph', 'as_rgb', 'bbox', 'border', 'borders', 'bwperim', 'cdilate', 'center_of_mass', 'cerode', 'citation', 'citation_text', 'close', 'close_holes', 'colors', 'convolve', 'convolve1d', 'croptobbox', 'cwatershed', 'daubechies', 'demos', 'dilate', 'disk', 'distance', 'dog', 'edge', 'erode', 'euler', 'features', 'find', 'fullhistogram', 'gaussian_filter', 'gaussian_filter1d', 'get_structuring_elem', 'haar', 'histogram', 'hitmiss', 'idaubechies', 'ihaar', 'imread', 'imresize', 'imsave', 'internal', 'io', 'label', 'labeled', 'labeled_sum', 'laplacian_2D', 'locmax', 'locmin', 'mahotas_version', 'majority_filter', 'mean_filter', 'median_filter', 'moments', 'morph', 'open', 'otsu', 'overlay', 'polygon', 'rank_filter', 'rc', 'regmax', 'regmin', 'resize', 'resize_to', 'segmentation', 'short_citation_text', 'sobel', 'stretch', 'stretch_rgb', 'template_match', 'test', 'tests', 'thin', 'thresholding', 'wavelet_center', 'wavelet_decenter']

The output of its version command also is failing:

>>> print(mahotas.mahotas_version._version_)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'mahotas' has no attribute 'mahotas_version'

The expected output is:

>>> print(mahotas.mahotas_version.__version__)
1.4.10

Overall, it seems that only a limited version of Mahotas, without its modules, was installed when pip install collageradiomics was called.

Potential Resolution

calling import again seems to resolve the issue.

>>> import collageradiomics; import collageradiomics

To run through the full baseline pipeline in Octave to get final images for comparison.

During the testing of the software I found that the documentation for the syntax specifically the inputs, outputs, and function calls could be clarified in the Documentation section instead of being only in the examples. While I was able to manage using the included examples I believe that a separate section in usage could aid potential users in understanding what and how to use the function more clearly.