process.py

import os
# os.environ["nnUNet_raw"] = "/home/head_neck/algorithm-template_2/nnunet_raw"
# os.environ["nnUNet_preprocessed"] = "/home/head_neck/algorithm-template_2/nnunet_preprocessed"
# os.environ["nnUNet_results"] = "/home/head_neck/algorithm-template/nnunet_results_5"
os.environ["nnUNet_raw"] = "./nnunet_raw"
os.environ["nnUNet_preprocessed"] = "./nnunet_preprocessed"
# os.environ["nnUNet_results"] = "/home/head_neck/algorithm-template/nnunet_results_task_2"
# os.environ["nnUNet_results"] = "./nnunet_results_task_2"
from typing import Dict
import tempfile
import subprocess
import SimpleITK as sitk
from nnunetv2.inference.predict_from_raw_data import nnUNetPredictor
from batchgenerators.utilities.file_and_folder_operations import load_json, join, isfile, maybe_mkdir_p, isdir, subdirs, \
    save_json
# import torch
import numpy as np

from base_algorithm import BaseSynthradAlgorithm
from revert_normalisation import get_ct_normalisation_values, revert_normalisation_single_modified

import torch
import shutil

import os

os.environ["OPENBLAS_NUM_THREADS"] = "1"

force_cpu = os.getenv("FORCE_CPU", "0") == "1"
device = torch.device("cuda:0" if torch.cuda.is_available() and not force_cpu else "cpu")


class SynthradAlgorithm2(BaseSynthradAlgorithm):
    """
    This class implements a simple synthetic CT generation algorithm that segments all values greater than 2 in the input image.

    Author: Suraj Pai (b.pai@maastrichtuniversity.nl)
    """

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

    def predict(self, input_dict: Dict[str, sitk.Image]) -> sitk.Image:
        assert list(input_dict.keys()) == ["image", "mask", "region"]

        region = input_dict["region"]
        mr_sitk = input_dict["image"]
        mask_sitk = input_dict["mask"]

        mr_np = sitk.GetArrayFromImage(mr_sitk).astype("float32")
        mask_np = sitk.GetArrayFromImage(mask_sitk).astype("float32")

        mr_np[mask_np == 0] = 0

        preprocessed_mr_sitk = sitk.GetImageFromArray(mr_np)
        preprocessed_mr_sitk.CopyInformation(mr_sitk)

        if region == "Head and Neck":
            dataset_name = "Dataset542"
            result_folder = "nnUNetTrainerMRCT_loss_masked_perception_masked__nnUNetResEncUNetLPlans__3d_fullres"
            plans_path = "./542_gt_nnUNetResEncUNetLPlans.json"
        if region == "Abdomen":
            dataset_name = "Dataset540"
            result_folder = "nnUNetTrainerMRCT_loss_masked_perception_masked__nnUNetResEncUNetLPlans__3d_fullres"
            plans_path = "./540_gt_nnUNetResEncUNetLPlans.json"
        if region == "Thorax":
            dataset_name = "Dataset544"
            result_folder = "nnUNetTrainerMRCT_loss_masked_perception_masked__nnUNetResEncUNetLPlans__3d_fullres"
            plans_path = "./544_gt_nnUNetResEncUNetLPlans.json"

        # predictor = nnUNetPredictor(
        #     tile_step_size=0.5,
        #     use_gaussian=True,
        #     use_mirroring=True,
        #     perform_everything_on_device=True,
        #     device=torch.device('cuda', 0),
        #     verbose=True,
        #     verbose_preprocessing=True,
        #     allow_tqdm=True
        # )
        predictor = nnUNetPredictor(
            tile_step_size=0.50,         
            use_gaussian=True,
            use_mirroring=False,      
            perform_everything_on_device=(device.type == "cuda"),
            device=device,
            verbose=True,
        )
        predictor.initialize_from_trained_model_folder(
            join(os.environ["nnUNet_results"], f'{dataset_name}/{result_folder}'),
            # use_folds=(0, 1, 2, 3, 4),
            use_folds=(0,),
            checkpoint_name='checkpoint_final.pth',
        )

        sitk_spacing = mr_sitk.GetSpacing()      
        sitk_origin  = mr_sitk.GetOrigin()       
        sitk_dir     = mr_sitk.GetDirection()    

        props = {
            'sitk_stuff': {
                'spacing': tuple(sitk_spacing),   
                'origin':  tuple(sitk_origin),    
                'direction': tuple(sitk_dir),     
            },
            'spacing': [sitk_spacing[2], sitk_spacing[1], sitk_spacing[0]]  
        }

        img = sitk.GetArrayFromImage(mr_sitk).astype(np.float32)
        img = np.expand_dims(img, 0) 
        ret = predictor.predict_single_npy_array(img, props, None, 'TRUNCATED', False)
        
        pred_path = "./TRUNCATED.nii.gz"
        pred_sitk = sitk.ReadImage(pred_path)

        ct_mean, ct_std = get_ct_normalisation_values(plans_path)
        mask_sitk = sitk.Cast(mask_sitk, sitk.sitkUInt8)

        pred_sitk = revert_normalisation_single_modified(pred_sitk, ct_mean, ct_std, mask_sitk=mask_sitk)
        os.remove(pred_path)
        shutil.rmtree("./imagesTs", ignore_errors=True)  
        shutil.rmtree("./predictions", ignore_errors=True)  
        return pred_sitk

# if __name__ == '__main__':
    
#     # Run the algorithm on the default input and output paths specified in BaseSynthradAlgorithm.
#     SynthradAlgorithm().process()

# if __name__ == '__main__':
# #     # test Brain
# #     #start_time = time.time()
# #     # img_s_path = '//datasets/work/hb-synthrad2023/source/raw_data/Task1/brain/1BA001/mr.mha'
# #     # img_m_path = '//datasets/work/hb-synthrad2023/source/raw_data/Task1/brain/1BA001/mask.mha'
# #     # img_fakeB_path = '//datasets/work/hb-synthrad2023/work/bw_workplace/output/task1_brain/p2p3D/ensemble_t1_brain_final_e2/test_predictions/1BA001/ct_fakeB.mha'
# #     # region = 'Head and Neck'

# #     # test Pelvis
#     img_s_path = '/home/head_neck/algorithm-template_2/task2/2ABA033_0000.mha'
#     img_m_path = '/home/head_neck/algorithm-template_2/task2/2ABA033_mask.mha'
# #     #img_fakeB_path = '//datasets/work/hb-synthrad2023/work/bw_workplace/output/task1_pelvis/p2p3D/exp6_data_7_size_256_256_56_batch_3_lr_0.0002_aug3d_fold0_resumed_multiscale_2dsample_5/epoch_best/epoch75/test_predictions/1PA005/ct_fakeB.mha'
#     region = 'Abdomen'


#     # start test
#     input_dict = {
#         "image": sitk.ReadImage(img_s_path),
#         "mask": sitk.ReadImage(img_m_path),
#         "region": region
#     }
#     algorithm = SynthradAlgorithm()
#     img_t = algorithm.predict(input_dict)