Analyse CSA

[sandrinebedard]

Description

The aim of the project is to get a smaller variaion of CSA across contrast then what we have on the GT.

The idea is to create a python script to analyse the output csv files with CSA.

CSA analysis of each training variation

• STD of CSA across 6 contrast persubject
• Mean (STD)
• Violin plot: of hard and soft GT

Impact of other factors:

• Contrast-agnostic
• Impact of soft averaged labels
• Non-binarization

[ebadrian]

Associated PR.

I created a preliminary notebook to visualize charts of interest with mock data. Currently we assume that we have a single pandas.Dataframe object containing all contrasts' CSA for each method. There are currently 2 valid type of charts:

• Violin plots: Individual contrast performance against a reference contrast for each method. These charts aim to showcase any bias of a specific contrast CSA value against the reference contrast CSA value. We use the pair-wise difference (in % w.r.t reference contrast) for the comparison.

  

• Violin plot: Methods' performance across contrasts using a measure of dispersion. Currently, the dispersion measure used is the standard deviation of all contrasts' CSA for each patient, for a given method. This chart aims to showcase the performance in CSA bias mitigation across contrasts of each method/baseline.

  

---

TODOs:

• [ ] A function that merges the intermediate sub-csv files. We assume that we will have one csv file per contrast. Each of these csv files will contain all CSA values of said contrast predicted/GT segmentation masks.
• [ ] A python script to run that will create all the charts once given the data folder containing all of these sub-csv files.

[naga-karthik]

@sandrinebedard this issue seems to be very old. Could you close the issue after you have confirmed? Thanks!