Home
Application Setup
Process the Data
Run the Application
Extras
ITCR Connectivity

Processing the Data for FeTS

Note the ${fets_root_dir} from Setup.

Application Path
Data Arrangement
Running Pre-processing
Starting from pre-processed data

Application Path

cd ${download_location}
${fets_root_dir}/bin/FeTS # launches application

Please add the following path to your LD_LIBRARY_PATH when using FeTS: ${fets_root_dir}/lib:

export LD_LIBRARY_PATH=${fets_root_dir}/lib:$LD_LIBRARY_PATH

Data Arrangement

Note the ${fets_root_dir} from Setup.

For the first application of FeTS in volumetric brain tumor MRI scans, you should follow the pre-processing pipeline defined in the International Brain Tumor Segmentation (BraTS) Challenge:

Download the DICOM files of the structural multi-parametric MRI scans:
- T1-weighted pre-contrast (T1)
- T1-weighted post-contrast (T1Gd)
- T2-weighted (T2)
- T2 Fluid Attenuated Inversion Recovery (T2-FLAIR)

Each subject’s DICOM (or NIfTI) image is to be put in a separate folder:

Input_Data
│
└───Patient_001
│   │
│   └───T1
│   │   │ image_001.dcm
│   │   │ image_002.dcm
│   │   │ ...
│   └───T1GD
│   │   │ image_001.dcm
│   │   │ image_002.dcm
│   │   │ ...
│   └───T2
│   │   │ image_001.dcm
│   │   │ image_002.dcm
│   │   │ ...
│   └───T2FLAIR
│   │   │ image_001.dcm
│   │   │ image_002.dcm
│   │   │ .....
│   
└───Pat_JohnDoe
│   │ ...
│   
│ ...   
│   
└───SmithJoe
│   │ ...

Ensure that unexpected “special” string characters that could hamper the file system’s processing are not present (read this article for more details). Some examples of this include (and are not limited to): !, @, #, $, %, ^, &, *, ', ;, :, ,, ~

Construct a CSV (let’s call this raw_data.csv) containing the first DICOM images from each modality:

PatientID,T1,T1GD,T2,T2FLAIR
Patient_001,/path/to/Patient_001/T1/image_001.dcm,/path/to/Patient_001/T1GD/image_001.dcm,/path/to/Patient_001/T2/image_001.dcm,/path/to/Patient_001/T2FLAIR/image_001.dcm
Pat_JohnDoe,/path/to/Pat_JohnDoe/T1/image_001.dcm,/path/to/Pat_JohnDoe/T1GD/image_001.dcm,/path/to/Pat_JohnDoe/T2/image_001.dcm,/path/to/Pat_JohnDoe/T2FLAIR/image_001.dcm
...
SmithJoe,/path/to/SmithJoe/T1/image_001.dcm,/path/to/SmithJoe/T1GD/image_001.dcm,/path/to/SmithJoe/T2/image_001.dcm,/path/to/SmithJoe/T2FLAIR/image_001.dcm

One way to do this would be via the following command:

echo "PatientID,T1,T1GD,T2,T2FLAIR" > raw_data.csv
for d in $Input_data/*; do sub=`basename $d`; t1=`ls -1 $d/T1/* | head -n1`; tlce=`ls -1 $d/T1GD/* | head -n1`; t2=`ls -1 $d/T2/* | head -n1`; flair=`ls -1 $d/T2FLAIR/* | head -n1`; echo $sub,$t1,$tlce,$t2,$flair >> raw_data.csv; done

Please note that subject IDs should be unique strings:
- Problematic: “Patient_1”, “Patient_2”, …, “Patient_10”, …,”Patient_20”, …,”Patient_100”
- Acceptable: “Patient_001”, “Patient_002”, …, “Patient_010”, …, “Patient_020”, …, “Patient_100”, …

Pre-processing

Pass raw_data.csv as an input, along with an output directory, to the PrepareDataset executable (which internally calls the BraTSPipeline executable):
```
${fets_root_dir}/bin/PrepareDataset -i /path/to/raw_data.csv -o /path/to/output
```

Two output directories will be created under the specified output directory with the following structure::

/path/to/output
│   │
│   └───DataForFeTS # this is to be passed for inference/training
│   │   │
│   │   └───Patient_001 # this is constructed from the ${PatientID} header of CSV
│   │   │   │ Patient_001_brain_t1.nii.gz
│   │   │   │ Patient_001_brain_t1ce.nii.gz
│   │   │   │ Patient_001_brain_t2.nii.gz
│   │   │   │ Patient_001_brain_flair.nii.gz
│   │   │   
│   │   └───Pat_JohnDoe # this is constructed from the ${PatientID} header of CSV
│   │   │   │ ...
│   │
│   │
│   └───DataForQC # this is to be used for quality-control
│   │   │
│   │   └───Patient_001 # this is constructed from the ${PatientID} header of CSV
│   │   │   │ raw_${modality}.nii.gz
│   │   │   │ raw_rai_${modality}.nii.gz
│   │   │   │ raw_rai_n4_${modality}.nii.gz
│   │   │   │ ${modality}_to_SRI.nii.gz
│   │   │   │ brainMask_SRI.nii.gz # generated using BrainMaGe [https://github.com/CBICA/BrainMaGe/] or DeepMedic [https://cbica.github.io/CaPTk/seg_DL.html]
│   │   │   │ log.txt
│   │   │   
│   │   └───Pat_JohnDoe # this is constructed from the ${PatientID} header of CSV
│   │   │   │ ...

NOTE: For some OS variants, we have seen PrepareDataset executable to cause issues, for which we have an alternative with ${fets_root_dir}/bin/PrepareDataset.py, which has the exact same API and can be invoked in the following way:

cd ${fets_root_dir}/bin
./OpenFederatedLearning/venv/bin/python \ # virtual environment that was set up in previous section
  ./PrepareDataset.py -i /path/to/raw_data.csv -o /path/to/output

Starting from Pre-processed data

NOTE: Skip this step if you have used PrepareDataset as described in https://fets-ai.github.io/Front-End/process_data#pre-processing

If you have processed data from a prior study that you would like to include in the FeTS federation, please ensure that all the data is co-registered within each patient and the annotations are in the same space. Once that is assured, follow these steps:

Arrange your data by using the processed image files in respective columns
Run PrepareDataset as shown above

In DataForQC, under each patient, the transformation matrices will be generated per modality. Use the T1CE_to_SRI.mat file (the assumption here is that the data is co-registered within each patient) to transform the annotation (which is in the patient space) in the following manner:

${fets_root_dir}/bin/Preprocessing \
  -i /path/to/patient_X/annotation.nii.gz \
  -rFI ${fets_root_dir}/data/sri24/atlasImage.nii.gz \
  -o /path/to/output/DataForFeTS/patient_X/annotation_final_seg.nii.gz \
  -rIA /path/to/output/DataForQC/patient_X/T1CE_to_SRI.mat \
  -reg Rigid -rIS 1 -rSg 1

Load the transformed images and corresponding annotation to ensure they are aligned correctly:
- /path/to/output/DataForFeTS/patient_X/brain_*
- /path/to/output/DataForFeTS/patient_X/annotation_final_seg.nii.gz

Back To Top ↑

Next: Run Application