Improvements on stats and process_data

sct-pipeline / csa-atrophy

Evaluate the sensitivity of atrophy detection with SCT

https://csa-atrophy.readthedocs.io/

MIT License

1 stars 0 forks source link

Improvements on stats and process_data #16

Closed PaulBautin closed 4 years ago

PaulBautin commented 4 years ago

STATS

So far most stats plots were computed with arrays. The use of pandas simplifies and clarifies the plots.

DONE: -refactoring "csa_rescale_stat" plots with pandas

PROCESS_DATA

When using "sct_label_vertebrae" 20% of subject labelling were false. The use of "-scale-dist" improves vertebrae detection to only 5% false labelling. To prevent subject overlapping, subjects already in results were removed before re-run. To enable manual labelling commented lines should be activated.

DONE: -delete already existing subjects in results -add options for labelling (comments) -add scaling in intervertebral distance for labelling

TODO:

docstring "csa_rescale_stat" file and choose relevant plots

csa_persub_plot

PaulBautin commented 4 years ago

Last commits implement review comments. Could you do a second review @jcohenadad?

PaulBautin commented 4 years ago

User is now able to indicate in command line his vertebrae levels of interest. Code is ready for review @jcohenadad . I have tested code for intra-subject variabilty (transformed images are considered like different subjects) and packaging everything seems to be working well.

PaulBautin commented 4 years ago

Ok, I have updated branch for everything to be compatible with master. The branch has been tested and everything seems to work fine. However i have not created a 'config' file for user to choose vertebrae of interest, by default every vertebrae levels present in the csv files are used, i think this keeps things simple. What do you think @jcohenadad ?

PaulBautin commented 4 years ago

The branch is ready for review. I have added:

manual labeling with new script './manual_labeling_correction.sh' + update QC in Readme
fix comments, docstrings, file naming

PaulBautin commented 4 years ago

The new approach for the manual corrections script will be integrated in a new PR (issue has already been open #23 ). PR is ready for new review @jcohenadad .

PaulBautin commented 4 years ago

List of corrections completed in last commit:

Correct typos, improve comments and update files with pycharm
Added flag -o to csa_rescale_stat to regroup plots under same directory
By default second segmentation process (with labelling corrections) is done with transfo_values of first process (without labelling correction) The PR is ready for review @jcohenadad

PaulBautin commented 4 years ago

Ready for review again, @jcohenadad.

PaulBautin commented 4 years ago

To summarize before PR close.

This PR has addressed numerous subjects such as:

implementation of pandas dataframe for stats
new script './manual_labeling_correction.sh'
generalize the use of different vertebrae levels with flag -l
add scaling for intervertebral distance during labelling
add flag -o to output flags in same file
change transformation values distribution from gaussian to uniform
improve docstrings and comments