If you find this code useful, please provide a reference to my github page for others www.github.com/brianmanderson , thank you!
This builds off of the function found here https://github.com/brianmanderson/Dicom_Data_to_Numpy_Arrays and into https://github.com/brianmanderson/Data_Generators
You can now resample the images during this process!
This takes the 3D images created from Dicom_to_Numpy_Arrays and creates a folder called Single_Images3D which identifies which slices have annotations and creates 2D images of the entire process
The reason for this is that you can manipulate how many images you want to load with the data generators, saving time by making it that you don't need to load the entire patient
from Make_Single_Images import run_main
path = path_to_split_Images
pickle_path = path_to_patient_info.pkl
extension = 50
run_main(path=path,
pickle_path=os.path.join(path,'patient_info_Ablation_Zones.pkl'),
resample=True, desired_output_spacing=(None,None,2.5),extension=extension)
This is saying we want to grab 50 images above and below the slices which have annotations
Also now supporting TFDataset by making records
image_processors do not need to be given, defaults are provided for 2D and 3D creation
from Make_Single_Images.Make_TFRecord_Class import write_tf_record, os
from Make_Single_Images.Image_Processors.Image_Processors_TFRecord import *
path = r'D:\Liver_Work'
image_processors = [Normalize_to_annotation(1), Clip_Images_By_Extension(16), Distribute_into_3D()]
write_tf_record(os.path.join(path, 'Train'), is_3D=False, rewrite=True, record_name='bbox_test', image_processors=image_processors)