Canproco dataset analysis

[plbenveniste]

On branch plb/dataframe_analysis Created a folder (dataset_analysis) for the analysis of the CanProCo dataset. The objective is to analyze the CanProCo dataset and perform correlation over different metrics:

• age
• phenotype
• sex
• pathology
• edss
• nb of lesions
• volume of lesions
• locations of lesions (nb of lesions per chunk of the spinal cord)
• ...

So far, I have created the file generate_dataframe.py which constructs the dataframe and gathers information. Right now, I am testing in on a fake canproco dataset which contains lesion segmentations, spinal cord segmentation and vertebral levels (will be added soon).

For now, the dataframe looks like this:

More columns will be added as I will be able to use vertebral levels soon ...

This work relies on the segmentation of the spinal cord and vertebral levels issue 46 (on which @valosekj and @plbenveniste are working)

[valosekj]

Thanks for the script!

I have created the file generate_dataset.py which constructs the dataset and gathers information.

The phrase "constructs the dataset" in this sentence and also the name of the generate_dataset.py script is a little bit confusing. canproco itself is already a dataset (we usually use the term dataset when talking about a bunch of MRI data from a single study). In other words, we are not constructing or generating any dataset here; the dataset already exists. We can use phrases such as "parse dataset", "analyze dataset," or "analyze lesions" (since the purpose of the generate_dataset.py script is to obtain the lesion characteristics).

[plbenveniste]

• [x] Modification done regarding the name of the file : changed from "dataset" to "dataframe"

Analysis of lesions per spinal cord level

The code was modified to add the analysis of lesions per spinal cord levels. It takes the disc segmentation and the labelled lesions (done by sct_analyse_lesions). Then it looks at lesions that are between two levels. (assumption: a lesion is between two levels if its centre point, a.k.a. the average coordinate of voxel in lesion, is between the two disc levels). The computed data is added to the dataframe.

The dataframe contains :

• the number of lesions between the two consecutive discs
• the combined volume of lesions between the two consecutive discs
• the avg vertical length of lesions between the two consecutive discs

Tests were performed on a fake canproco dataset (which contains the required segmentations)

What next

• Look at where we are at regarding gathering more patient data (more clinical scores)
• Start data analysis pipeline (in a jupyter notebook)

[jcohenadad]

@plbenveniste could you please describe code evolution within a PR instead of within an issue? That will enable you to associate statements like "The code was modified" with an actual commit -- otherwise we don't know what code was modified, how it was modified, when it was modified and on what branch. Thanks