def __getitem__(self, idx):
datamap = self.datamap_list[idx]
if datamap[0]:
label, data_idx, aug_idx = datamap
data_path = self.positive_list[data_idx]
volume_overbound = np.load(data_path)
volume_fitted = da.extract_augment_random(
aug_idx, volume_overbound, self.data_size,
self.translation_num, self.translation_range,
self.rotation_num, self.flip_num)
else:
label, data_idx = datamap
data_path = self.negative_list[data_idx]
volume_fitted = np.load(data_path)
if np.linalg.norm(np.int_(volume_fitted.shape) -
self.data_size) != 0:
volume_fitted = mt.local_crop(
volume_fitted,
np.rint(np.array(volume_fitted.shape) / 2).astype(int),
self.data_size
) #the sample should be cropped to fit the data size
#volume = volume_fitted + np.random.random_sample(volume_fitted.shape) * random.choice(self.noise_range)
volume = da.add_noise(volume_fitted, self.noise_range)
if self.mode not in self.norm_mode_dict.keys():
norm_mode = "CT_lung"
else:
norm_mode = self.norm_mode_dict[self.mode]
volume = self.data_preprocess(volume, norm_mode, num_channels=1)
'''
volume_normalized = mt.medical_normalization(volume_fitted, input_copy=False)
volume_noisy = volume_normalized + np.random.random_sample(volume_normalized.shape) * random.choice(self.noise_range)
volume_reshaped = volume_noisy.reshape(1, volume_noisy.shape[0], volume_noisy.shape[1], volume_noisy.shape[2])
volume = torch.from_numpy(volume_reshaped).float()
'''
output = {'data': volume, 'label': label}
if self.file_retrieval:
output['path'] = data_path
#return {'data': volume, 'label': label, 'path': data_path}
#else: return {'data': volume, 'label': label}
return output
elif pfile.split('/')[0].find("tianchi") >= 0:
patient_uid, nodule_diameter = mt.get_luna_annotation_informations(
pfile, annotation2)
else:
patient_uid = mt.get_volume_informations(pfile)[0]
nodule_diameter = 0
data_volume = da.extract_volumes(
data_volume,
np.int_([REGION_SIZE, REGION_SIZE, REGION_SIZE]),
nodule_diameter=nodule_diameter,
scale_augment=AUGMENTATION,
translation_augment=AUGMENTATION,
rotation_augment=AUGMENTATION)
else:
data_volume = mt.local_crop(
data_volume,
np.rint(np.array(data_volume.shape) /
2).astype(int), REGION_SIZE)
data_volume = data_volume.reshape(
(1, data_volume.shape[0], data_volume.shape[1],
data_volume.shape[2]))
data_label = np.zeros(shape=(data_volume.shape[0], 2),
dtype=float)
data_label[:, 1 - int(isnodule)] = 1
if "data_volumes" not in dir():
data_volumes = data_volume
data_labels = data_label
else:
data_volumes = np.concatenate(
(data_volumes, data_volume), axis=0)
data_labels = np.concatenate((data_labels, data_label),
axis=0)
