def compute_warped_image_label(img_label_txt_pth,phi_pth,phi_type, saving_pth):
img_label_pth_list = read_txt_into_list(img_label_txt_pth)
phi_pth_list = glob(os.path.join(phi_pth,phi_type))
f = lambda pth: sitk.GetArrayFromImage(sitk.ReadImage(pth))
img_list = [f(pth[0]) for pth in img_label_pth_list]
label_list = [f(pth[1]) for pth in img_label_pth_list]
num_img = len(img_list)
for i in range(num_img):
fname = get_file_name(img_label_pth_list[i][0])
img = torch.Tensor(img_list[i][None][None])
label = torch.Tensor(label_list[i][None][None])
f_phi = lambda x: get_file_name(x).find(fname)==0
phi_sub_list = list(filter(f_phi, phi_pth_list))
num_aug = len(phi_sub_list)
phi_list = [f(pth) for pth in phi_sub_list]
img = img.repeat(num_aug,1,1,1,1)
label = label.repeat(num_aug,1,1,1,1)
phi = np.stack(phi_list,0)
phi = np.transpose(phi,(0,4,3,2,1))
phi = torch.Tensor(phi)
sz = np.array(img.shape[2:])
spacing = 1./(sz-1)
phi, _ = resample_image(phi,spacing,[1,3]+list(img.shape[2:]))
warped_img = compute_warped_image_multiNC(img,phi,spacing,spline_order=1,zero_boundary=True)
warped_label = compute_warped_image_multiNC(label,phi,spacing,spline_order=0,zero_boundary=True)
save_image_with_given_reference(warped_img,[img_label_pth_list[i][0]]*num_aug,saving_pth,[get_file_name(pth).replace("_phi","")+'_warped' for pth in phi_sub_list])
save_image_with_given_reference(warped_label,[img_label_pth_list[i][0]]*num_aug,saving_pth,[get_file_name(pth).replace("_phi","")+'_label' for pth in phi_sub_list])
def get_test_file_for_brainstorm_color(test_path,transfer_path,output_txt):
#atlas_image_9023193_image_test_iter_0_warped.nii.gz
file_label_list = read_txt_into_list(test_path)
file_list, label_list = [file[0] for file in file_label_list],[file[1] for file in file_label_list]
f = lambda x: "atlas_image_"+x+"_test_iter_0_warped.nii.gz"
new_file_list = [os.path.join(transfer_path,f(get_file_name(file))) for file in file_list]
new_file_label_list = [[new_file_list[i],label_list[i]] for i in range(len(file_label_list))]
write_list_into_txt(output_txt,new_file_label_list)
def init_test_env(setting_path, output_path, file_list, fname_list):
"""
create test environment, the file list would be saved into output_path/reg/test/file_path_list.txt,
a corresponding auto-parsed filename list would also be saved in output/path/reg/test/file_name_list.txt
:param setting_path: the path to load 'cur_task_setting.json' and 'cur_data_setting.json' (optional if the related settings are in cur_task_setting)
:param output_path: the output path of the task
:param image_path_list: the image list, each item refers to the abstract path of the image
:param l_path_list:optional, the label of image list, each item refers to the abstract path of the image
:return: tuple of ParameterDict, datapro (optional) and tsk_set
"""
dm_json_path = os.path.join(setting_path, 'cur_data_setting.json')
tsm_json_path = os.path.join(setting_path, 'cur_task_setting.json')
assert os.path.isfile(tsm_json_path), "task setting not exists"
dm = DataTask('task_reg',
dm_json_path) if os.path.isfile(dm_json_path) else None
tsm = ModelTask('task_reg', tsm_json_path)
file_num = len(file_list)
os.makedirs(os.path.join(output_path, 'seg/test'), exist_ok=True)
os.makedirs(os.path.join(output_path, 'seg/res'), exist_ok=True)
file_txt_path = os.path.join(output_path, 'seg/test/file_path_list.txt')
fn_txt_path = os.path.join(output_path, 'seg/test/file_name_list.txt')
has_label = len(file_list[0]) == 2
if fname_list is None:
if has_label:
fname_list = [
get_file_name(file_list[i][0]) for i in range(file_num)
]
else:
fname_list = [get_file_name(file_list[i]) for i in range(file_num)]
write_list_into_txt(file_txt_path, file_list)
write_list_into_txt(fn_txt_path, fname_list)
data_task_name = 'seg'
cur_task_name = 'res'
if dm is not None:
dm.data_par['datapro']['dataset']['output_path'] = output_path
dm.data_par['datapro']['dataset']['task_name'] = data_task_name
tsm.task_par['tsk_set']['task_name'] = cur_task_name
tsm.task_par['tsk_set']['output_root_path'] = os.path.join(
output_path, data_task_name)
return dm, tsm
def get_pair_txt_for_oai_reg_net(train_txt_path,warped_folder,warped_type, num_train,output_txt):
train_pair_list = read_txt_into_list(train_txt_path)
warped_file_list = glob(os.path.join(warped_folder,warped_type))
name_set = [get_file_name(pair[0]).split("_")[0] for pair in train_pair_list]
name_set = set(name_set)
name_file_dict = {name:[] for name in name_set}
extra_weight = 2
for pair in train_pair_list:
fname = get_file_name(pair[0]).split("_")[0]
for i in range(extra_weight):
name_file_dict[fname].append(pair[0])
name_file_dict[fname].append(pair[1])
for file in warped_file_list:
fname = get_file_name(file).split("_")[0]
name_file_dict[fname].append(file)
num_per_patient = int(num_train/len(name_set))
train_list = []
for name,values in name_file_dict.items():
num_sample = 0
while num_sample < num_per_patient:
pair = random.sample(name_file_dict[name],2)
if get_file_name(pair[0])==get_file_name(pair[1]) or get_file_name(pair[0]).split("_")[1]==get_file_name(pair[1]).split("_")[1]:
continue
else:
train_list.append(pair)
num_sample += 1
write_list_into_txt(output_txt, train_list)
deformation_np, mask_np, spacing = read_deformation_and_mask(deformation_path_list[i], mask_path_list[i],itk_format)
extra_res = compute_jacobi_map(deformation_np,spacing,[fname_list[i]],mask_np,crop_boundary=True,save_jacobi_map=True,saving_folder=saving_folder)
jacobi_val_res += extra_res[0] * batch_size
jacobi_num_res += extra_res[1] * batch_size
average_jacobi_masked_res += extra_res[2]*batch_size
records_jacobi_val_np[i] = extra_res[0]
records_jacobi_num_np[i] = extra_res[1]
average_jacobi_masked_np[i] = extra_res[2]
print("id {} and current pair name is : {}".format(i,os.path.split(f_path)[-1]))
print('the running averge jocobi val sum is {}'.format(jacobi_val_res/(i+1)/batch_size))
print('the running averge jocobi num sum is {}'.format(jacobi_num_res/(i+1)/batch_size))
print('the running averge jocobi masked average is {}'.format(average_jacobi_masked_res/(i+1)/batch_size))
jacobi_val_res = jacobi_val_res/num_samples
jacobi_num_res = jacobi_num_res/num_samples
average_jacobi_masked_res = average_jacobi_masked_res/num_samples
print("the average {}_ jacobi val sum: {} :".format('test', jacobi_val_res))
print("the average {}_ jacobi num sum: {} :".format('test', jacobi_num_res))
print("the average {}_ average_jacobi_masked average: {} :".format('test', average_jacobi_masked_res))
np.save(os.path.join(saving_folder,'records_jacobi'),records_jacobi_val_np)
np.save(os.path.join(saving_folder,'records_jacobi_num'),records_jacobi_num_np)
np.save(os.path.join(saving_folder,'records_average_jacobi_masked'),average_jacobi_masked_np)
deformation_mask_folder_path = "/playpen-raid1/zyshen/data/reg_new_lung/testing_lddmm/reg/res/records/original_sz"
deformation_path_list =glob.glob(os.path.join(deformation_mask_folder_path,'*img_inv_phi.nii.gz'), recursive=True)
f_mask_list =[f.replace('img_inv_phi','img_moving_l') for f in deformation_path_list]
fname_list = [get_file_name(f).replace("_img_inv_phi","") for f in deformation_path_list]
saving_path = "/playpen-raid1/zyshen/data/reg_new_lung/testing_lddmm/jacobi_analysis"
compute_jacobi(deformation_path_list,fname_list, saving_path, f_mask_list,itk_format=False)