INIT

README.md

@@ -0,0 +1,29 @@
+
+## [Pre- to Post-Contrast Breast MRI Synthesis for Enhanced Tumour Segmentation]()
+
+In SPIE Medical Imaging 2024.
+
+![examples](docs/examples.png)
+
+## Getting Started
+The [Duke Dataset](https://sites.duke.edu/mazurowski/resources/breast-cancer-mri-dataset/) used in this study is available on [The Cancer Imaging Archive (TCIA)](https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=70226903).
+
+You may find some examples of synthetic nifti files in [synthesis/examples](synthesis/examples).
+
+### Synthesis Code
+- [Config](synthesis/pix2pixHD/scripts/train_512p_duke_2D_w_GPU_1to195.sh) to run a training of the image synthesis model.
+- [Config](synthesis/pix2pixHD/scripts/test_512p_duke_2D_w_GPU_1to195.sh) to run a test of the image synthesis model.
+- [Code](synthesis/utils/convert_to_nifti_whole_dataset.py) to transform Duke DICOM files to NiFti files.
+- [Code](synthesis/utils/nifti_png_conversion.py) to extract 2D pngs from 3D NiFti files.
+- [Code](synthesis/utils/png_nifti_conversion.py) to create 3D NiFti files from axial 2D pngs.
+- [Code](synthesis/utils/get_training_patient_ids.py) to separate synthesis training and test cases.
+- [Code](synthesis/utils/metrics.py) to compute the image quality metrics such as SSIM, MSE, LPIPS, and more. 
+- [Code](synthesis/utils/fid.py) to compute the Frèchet Inception Distance (FID) based on ImageNet and [RadImageNet](https://github.com/BMEII-AI/RadImageNet).  
+
+### Segmentation Code
+- [Code](nnUNet/custom_scripts/convert_data_to_nnunet_204.py) to prepare 3D single breast cases for nnunet segmentation.
+- [Train-test-splits](nnUNet/nnunetv2/nnUNet_preprocessed/Dataset208_DukePreSynthetic/splits_final_pre_post_syn.json) of the segmentation dataset.
+- [Script](nnUNet/custom_scripts/full_pipeline.sh) to run the full nnunet pipeline on the Duke dataset.
+
+## Acknowledgements
+This repository borrows code from the [pix2pixHD](https://github.com/NVIDIA/pix2pixHD) and the [nnUNet](https://github.com/MIC-DKFZ/nnUNet) repositories. The 254 tumour segmentation masks used in this study were provided by [Caballo et al](https://doi.org/10.1002/jmri.28273).

nnUNet/custom_scripts/convert_data_to_nnunet_204.py

@@ -0,0 +1,66 @@
+import os
+import SimpleITK as sitk
+import numpy as np
+from scipy import ndimage
+import csv
+import random
+
+import shutil
+
+def save_image(save_array, reference_img, save_path):
+    image = sitk.GetImageFromArray(save_array)
+    image.SetDirection(reference_img.GetDirection())
+    image.SetSpacing(reference_img.GetSpacing())
+    image.SetOrigin(reference_img.GetOrigin())
+
+    sitk.WriteImage(image, os.path.join(save_path))
+
+
+def save_the_half(paths): #patient, pre_path, post_path
+
+    patient = paths[0]
+    post_path = paths[1]
+
+    if os.path.exists(post_path):
+        post = sitk.ReadImage(post_path)
+        save_name = post_path.split('/')[-1]
+        print(save_name)
+
+        if not os.path.exists(os.path.join('/workspace/AutomaticSegmentation/nnUNet/nnunetv2/nnUNet_raw/Dataset204_DukePhaseOneHalf/imagesTr', save_name)):
+
+            # create the half image based on where the mask is:
+            post_array = sitk.GetArrayFromImage(post)
+
+            # find which side the segmentation is - right or left
+            if os.path.exists(os.path.join('/data/Duke-Breast-Cancer-MRI-Nifti-Whole-Preprocessed/masks_preprocessed', patient + '.nii.gz')):
+                segmentation = sitk.GetArrayFromImage(sitk.ReadImage(os.path.join('/data/Duke-Breast-Cancer-MRI-Nifti-Whole-Preprocessed/masks_preprocessed', patient + '.nii.gz')))
+                segmentation_coordinates = ndimage.center_of_mass(segmentation) 
+
+                middle_coordinate = segmentation.shape[-1]/2
+                if segmentation_coordinates[-1] < middle_coordinate:
+                    new_post = post_array[:, :, :np.int64(middle_coordinate)]
+                    new_segmentation = segmentation[:, :, :np.int64(middle_coordinate)]
+                else:
+                    new_post = post_array[:, :, np.int64(middle_coordinate + 1):]
+                    new_segmentation= segmentation[:, :, np.int64(middle_coordinate + 1):]
+                save_image(new_post, post, os.path.join('/workspace/AutomaticSegmentation/nnUNet/nnunetv2/nnUNet_raw/Dataset204_DukePhaseOneHalf/imagesTr', patient+'_0000.nii.gz'))
+                save_image(new_segmentation, post, os.path.join('/workspace/AutomaticSegmentation/nnUNet/nnunetv2/nnUNet_raw/Dataset204_DukePhaseOneHalf/labelsTr',  patient+'.nii.gz'))
+
+def main():
+    processed_nifti_path = '/data/Duke-Breast-Cancer-MRI-Nifti-Whole-Preprocessed'
+    pre_post_list = []
+
+    # step 1
+    for patient in os.listdir(processed_nifti_path):
+
+        post = os.path.join(processed_nifti_path, patient, patient + '_0001.nii.gz')
+        save_the_half([patient, post])
+    #     # pre_post_list.append([patient, pre, post])
+
+    # # step 2
+    # # remove multifocal images
+
+    # step 3
+    # select test data and remove it from the training set
+
+main()

nnUNet/custom_scripts/convert_data_to_nnunet_205.py

@@ -0,0 +1,67 @@
+import os
+import SimpleITK as sitk
+import numpy as np
+from scipy import ndimage
+import csv
+import random
+
+import shutil
+
+def save_image(save_array, reference_img, save_path):
+    image = sitk.GetImageFromArray(save_array)
+    image.SetDirection(reference_img.GetDirection())
+    image.SetSpacing(reference_img.GetSpacing())
+    image.SetOrigin(reference_img.GetOrigin())
+
+    sitk.WriteImage(image, os.path.join(save_path))
+
+
+def save_the_half(paths): #patient, pre_path, post_path
+
+    patient = paths[0]
+    post_path = paths[1]
+
+    if os.path.exists(post_path):
+        post = sitk.ReadImage(post_path)
+        save_name = post_path.split('/')[-1]
+        print(save_name)
+
+        if not os.path.exists(os.path.join('/workspace/AutomaticSegmentation/nnUNet/nnunetv2/nnUNet_raw/Dataset205_DukePreHalf/imagesTr', save_name)):
+
+            # create the half image based on where the mask is:
+            post_array = sitk.GetArrayFromImage(post)
+
+            # find which side the segmentation is - right or left
+            if os.path.exists(os.path.join('/data/Duke-Breast-Cancer-MRI-Nifti-Whole-Preprocessed/masks_preprocessed', patient + '.nii.gz')):
+                segmentation = sitk.GetArrayFromImage(sitk.ReadImage(os.path.join('/data/Duke-Breast-Cancer-MRI-Nifti-Whole-Preprocessed/masks_preprocessed', patient + '.nii.gz')))
+                segmentation_coordinates = ndimage.center_of_mass(segmentation) 
+
+                middle_coordinate = segmentation.shape[-1]/2
+                if segmentation_coordinates[-1] < middle_coordinate:
+                    new_post = post_array[:, :, :np.int64(middle_coordinate)]
+                    new_segmentation = segmentation[:, :, :np.int64(middle_coordinate)]
+                else:
+                    new_post = post_array[:, :, np.int64(middle_coordinate + 1):]
+                    new_segmentation= segmentation[:, :, np.int64(middle_coordinate + 1):]
+                save_image(new_post, post, os.path.join('/workspace/AutomaticSegmentation/nnUNet/nnunetv2/nnUNet_raw/Dataset205_DukePreHalf/imagesTr', patient+'_0000.nii.gz'))
+                save_image(new_segmentation, post, os.path.join('/workspace/AutomaticSegmentation/nnUNet/nnunetv2/nnUNet_raw/Dataset205_DukePreHalf/labelsTr',  patient+'.nii.gz'))
+
+
+def main():
+    processed_nifti_path = '/data/Duke-Breast-Cancer-MRI-Nifti-Whole-Preprocessed'
+    pre_post_list = []
+
+
+    # step 1
+    for patient in os.listdir(processed_nifti_path):
+        post = os.path.join(processed_nifti_path, patient, patient + '_0000.nii.gz')
+        save_the_half([patient, post])
+
+    # step 2
+    # remove multifocal images
+
+    # step 3
+    # select test data and remove it from the training set
+
+
+main()