convert.py

import os
import pydicom
import pydicom_seg
import pandas as pd
import SimpleITK as sitk

from glob import glob
from tqdm import tqdm


data = pd.read_csv('metadata.csv')  # This file should be created by NBIA Data Retriever
patient_ids = data['PatientID'].unique()

img_reader = sitk.ImageSeriesReader()
seg_reader = pydicom_seg.SegmentReader()

for pid in tqdm(patient_ids):

    row = data[data['PatientID'] == pid]
    out_fn = f'NSCLC-Radiomics-NIFTI/{pid}'
    os.makedirs(out_fn)

    dicom_names = img_reader.GetGDCMSeriesFileNames(row[row['Modality'] == 'CT'].SeriesInstanceUID.values[0])
    img_reader.SetFileNames(dicom_names)
    image = img_reader.Execute()

    sitk.WriteImage(image, os.path.join(out_fn, f'image.nii.gz'), True)

    if pid == 'LUNG1-128': continue  # LUNG1-128 missing segmentation

    dicom_names = img_reader.GetGDCMSeriesFileNames(row[row['Modality'] == 'SEG'].SeriesInstanceUID.values[0])
    dcm = pydicom.dcmread(dicom_names[0])
    result = seg_reader.read(dcm)

    for segment_number in result.available_segments:
        image = result.segment_image(segment_number)  # lazy construction
        sitk.WriteImage(image, os.path.join(out_fn, f'seg-{result.segment_infos[segment_number].SegmentDescription}.nii.gz'), True)