def preprocessing_valid(self):
self.z_valid = z_calc(self.valid_pat, self.valid_sub_pat)
self.z_startvalid, self.z_endvalid = calculating_slicesofinterest_segtrain(self.valid_pat, self.valid_sub_pat, self.z_valid, self.roi_interest)
self.z_startvalid, self.z_endvalid = calculating_slicesofinterestfor3D(self.valid_pat,self.z_startvalid, self.z_endvalid)
self.x_centvalid, self.y_centvalid = calculate_centroid_segvalid(self.valid_pat,
self.valid_sub_pat, self.z_startvalid, self.z_endvalid, self.roi_interest)
np.save(os.path.join(self.main_pred_dir, 'valid.npy'), self.valid_pat)
np.save(os.path.join(self.main_pred_dir, 'valid_sub.npy'), self.valid_sub_pat)
np.save(os.path.join(self.main_pred_dir, 'z_valid.npy'), self.z_valid)
np.save(os.path.join(self.main_pred_dir, 'z_startvalid.npy'), self.z_startvalid)
np.save(os.path.join(self.main_pred_dir, 'z_endvalid.npy'), self.z_endvalid)
np.save(os.path.join(self.main_pred_dir, 'x_centvalid.npy'), self.x_centvalid)
np.save(os.path.join(self.main_pred_dir, 'y_centvalid.npy'), self.y_centvalid)
def preprocessing_train(self):
self.z_TRAIN = z_calc(self.train_pat, self.train_pat_sublist)
self.z_starttrain, self.z_endtrain = calculating_slicesofinterest_segtrain(self.train_pat, self.train_pat_sublist, self.z_TRAIN, self.roi_interest)
self.z_starttrain, self.z_endtrain = calculating_slicesofinterestfor3D(self.train_pat,self.z_starttrain, self.z_endtrain)
self.x_centtrain, self.y_centtrain = calculate_centroid_segtrain(self.train_pat,
self.train_pat_sublist, self.z_starttrain, self.z_endtrain, self.roi_interest)
np.save(os.path.join(self.main_pred_dir, 'train.npy'), self.train_pat)
np.save(os.path.join(self.main_pred_dir, 'train_sub.npy'), self.train_pat_sublist)
np.save(os.path.join(self.main_pred_dir, 'z_train.npy'), self.z_TRAIN)
np.save(os.path.join(self.main_pred_dir, 'z_starttrain.npy'), self.z_startvalid)
np.save(os.path.join(self.main_pred_dir, 'z_endtrain.npy'), self.z_endtrain)
np.save(os.path.join(self.main_pred_dir, 'x_centtrain.npy'), self.x_centtrain)
np.save(os.path.join(self.main_pred_dir, 'y_centtrain.npy'), self.y_centtrain)
def predict(self):
self.test_pat = np.load(os.path.join(self.data_dir, 'test.npy'))
self.test_sub = np.load(os.path.join(self.data_dir, 'test_sub.npy'))
self.z_test = z_calc(self.test_pat, self.test_sub)
#localization
self.test_imgs_in, self.test_imgs_roi = self.create_locnpys(
self.test_pat, self.test_sub, self.z_test)
cust = {
'channel_RSDice': channel_RSDice,
'channel_RSloss': channel_RSloss
}
self.loc_network = load_model('locnetwork_2.h5', cust)
imgs_roi_pred = self.loc_network.predict(self.test_imgs_in,
batch_size=1,
verbose=1)
save_coarse_predictions(self.loc_dir, imgs_roi_pred)
#OAR segmentation
oar_pred_dir = '/data/M1_FM2'
self.preprocessing_test(oar_pred_dir, 2)
cust = {
'weighted_dice_loss3D': weighted_dice_loss3D,
'dice_coef': dice_coef,
'GroupNormalization': GroupNormalization,
'DropBlock3D': DropBlock3D
}
self.loc_network = load_model('FM2_network.h5', cust)
f = K.function([self.loc_network.layers[0].input,
K.learning_phase()],
[self.loc_network.layers[-1].output])
self.predict_with_uncertainty(f,
oar_pred_dir,
5,
self.test_imgs_in,
1,
n_iter=50)
oar_pred_dir = '/data/M1_FM3'
self.preprocessing_test(oar_pred_dir, 3)
cust = {
'weighted_dice_loss3D': weighted_dice_loss3D,
'dice_coef': dice_coef,
'GroupNormalization': GroupNormalization,
'DropBlock3D': DropBlock3D
}
self.loc_network = load_model('FM3_network.h5', cust)
f = K.function([self.loc_network.layers[0].input,
K.learning_phase()],
[self.loc_network.layers[-1].output])
self.predict_with_uncertainty(f,
oar_pred_dir,
5,
self.test_imgs_in,
1,
n_iter=50)
oar_pred_dir = '/data/M1_PB'
self.preprocessing_test(oar_pred_dir, 8)
cust = {
'weighted_dice_loss3D': weighted_dice_loss3D,
'dice_coef': dice_coef,
'GroupNormalization': GroupNormalization,
'DropBlock3D': DropBlock3D
}
self.loc_network = load_model('PB_network.h5', cust)
f = K.function([self.loc_network.layers[0].input,
K.learning_phase()],
[self.loc_network.layers[-1].output])
self.predict_with_uncertainty(f,
oar_pred_dir,
5,
self.test_imgs_in,
1,
n_iter=50)
oar_pred_dir = '/data/M1_Bladder'
self.preprocessing_test(oar_pred_dir, 5)
cust = {
'weighted_dice_loss3D': weighted_dice_loss3D,
'dice_coef': dice_coef,
'GroupNormalization': GroupNormalization,
'DropBlock3D': DropBlock3D
}
self.loc_network = load_model('Bladder_network.h5', cust)
f = K.function([self.loc_network.layers[0].input,
K.learning_phase()],
[self.loc_network.layers[-1].output])
self.predict_with_uncertainty(f,
oar_pred_dir,
5,
self.test_imgs_in,
1,
n_iter=50)
oar_pred_dir = '/data/M1_Rectum'
self.preprocessing_test(oar_pred_dir, 4)
cust = {
'weighted_dice_loss3D': weighted_dice_loss3D,
'dice_coef': dice_coef,
'GroupNormalization': GroupNormalization,
'DropBlock3D': DropBlock3D
}
self.loc_network = load_model('Rectum_network.h5', cust)
f = K.function([self.loc_network.layers[0].input,
K.learning_phase()],
[self.loc_network.layers[-1].output])
self.predict_with_uncertainty(f,
oar_pred_dir,
4,
self.test_imgs_in,
1,
n_iter=50)
oar_pred_dir = '/data/M1_CTV'
self.preprocessing_test(oar_pred_dir, 7)
cust = {
'weighted_dice_loss3D': weighted_dice_loss3D,
'dice_coef': dice_coef,
'GroupNormalization': GroupNormalization,
'DropBlock3D': DropBlock3D
}
self.loc_network = load_model('Rectum_network.h5', cust)
f = K.function([self.loc_network.layers[0].input,
K.learning_phase()],
[self.loc_network.layers[-4].output])
self.predict_with_uncertainty(f,
oar_pred_dir,
7,
self.test_imgs_in,
1,
n_iter=50)
