Cycle-GAN CBCT-CT Evaluation Pipeline

Evaluation pipleline for the CBCT to CT domain transfer using Cycle-GAN project. Uses Plastimatch to do all the necessary functions and finally calculates the Dice Similaruty Score, Hausdorff Distance and the Euclidean distance of the fiducial markers. This project utilizes Plastimatch, TotalSegmentor, 3D Slicer.

Folder Structure

Copy the {patient-number}-LinearTransforms.txt file from the LinearTransforms folder to the respective Patient folders and make sure the project structure look likes this before proceeding.

project/
├── datasets/
│   └── pelvic_reference/
│       ├── Pelvic-Ref-001/              # Patient with GT contours
│       │   ├── CBCT/                    # Contains the CBCT DICOM files
│       │   ├── CT/                      # Contains the CT DICOM files
│       │   ├── FDMs/                    # Contains the Fiducial Distance Markup files (if available)
│       │   ├── GT_contours/             # Contains ground truth contours
│       │   │   ├── CT/                  # CT contours in .mha format
│       │   │   └── CBCT/                # CBCT contours in .mha format
│       │   └── 001-LinearTransform.txt  # Affine transform values from 3D Slicer
│       ├── Pelvic-Ref-002/              # Patient without GT contours
│       │   ├── CBCT/
│       │   ├── CT/
│       │   └── 002-LinearTransform.txt
│       └── ... (additional patient data folders)
├── evaluation/
│   ├── __init__.py
│   ├── config.py
│   ├── fcsv.py
│   ├── params.py
│   ├── pipeline.py
│   ├── plastimatch.py
│   └── util.py
├── main.py
├── requirements.txt
└── README.md

Steps to Run the Code

1. Create a new environment

   conda create --name eval-pipeline python=3.10
   conda activate eval-pipeline

2. Install the dependencies

   a. Install PyTorch - PyTorch Installation.

   b. Install the Remaining Dependencies

   pip install -r requirements.txt

3. Run the pipeline in terminal (-d argument should be in glob format)

   python main.py -d ./datasets/pelvic_reference/Pel*

4. Run the pipeline from code

   from evaluation.pipeline import EvaluationPipeline
   from glob import glob
   import os
   
   data = glob(os.path.join(os.path.curdir, "datasets", "pelvic_reference", "Pel*"))
   pipeline = EvaluationPipeline()
   pipeline.evaluate(data)

Evaluation Pipeline Arguments

1. For Terminal

usage: main.py [-h] [-d] [-n] [-f] [-a] [-s] [-pw] [-dm] [-c] [-fc] [-p] [-r] [-w] [-m] [-fs]

options:
-h, --help            show this help message and exit
-d, --data            all patient dirs in glob format
-n, --nums            patient numbers to process (in csv format)
-f, --force           force run, deletes previous results and creates a new one. Otherwise skips the step if results already present
-a, --all             run all the steps
-s, --seg             run segmentation only
-pw, --pw-linear      run pw-linear transformation only
-dm, --dmap           run dmap calculation only
-c, --cxt             run cxt conversion only
-fc, --fcsv           run fcsv creation only
-p, --params          run register params.txt creation only
-r, --register        run plastimatch register only
-w, --warp            run warp only
-m, --metric          run calculate scores only
-fs, --fiducial-sep   run calculate fiducial distance only

Example:

  1. python main.py -d ./datasets/pelvic_reference/Pel* -f -n 3,4 -a        # Force runs all the steps for patients 3 and 4
  2. python main.py -d ./datasets/pelvic_reference/Pel* -f -n 3,4 -s -pw    # Force runs the segmentation and pw-linear transform steps for patients 3 and 4
  3. python main.py -d ./datasets/pelvic_reference/Pel* -f -s -pw           # Force runs the segmentation and pw-linear transform steps for all the patients
  3. python main.py -d ./datasets/pelvic_reference/Pel* -s -pw              # Initates the segmentation and pw-linear transform steps for all the patients, but skips if the result is already present

2. For Python

   nums = [] or [3, 4]   # Runs for all the patients if [] is passed
   force = False
   all = True            # Runs all the steps if True
   seg = False
   pw_linear = False
   dmap = False
   cxt = False
   fcsv = False
   params = False
   register = False
   warp = False
   metric = False
   fiducial_sep = False
   data = glob(os.path.join(os.path.curdir, "datasets", "pelvic_reference", "Pel*"))
   
   pipeline = EvaluationPipeline()
   pipeline.evaluate(data, force, nums, all, seg, pw_linear,
                        dmap, cxt, fcsv, params, register,
                        warp, metric, fiducial_sep)

References

1. Plastimatch
2. TotalSegmentator

---