def load_mri_volumes(self, normalize) -> np.ndarray:
patient_path = os.path.join(self.data_path, self.patch_name)
flair = load_nifi_volume(os.path.join(patient_path, self.flair),
normalize)
t1 = load_nifi_volume(os.path.join(patient_path, self.t1), normalize)
t2 = load_nifi_volume(os.path.join(patient_path, self.t2), normalize)
t1_ce = load_nifi_volume(os.path.join(patient_path, self.t1ce),
normalize)
modalities = np.stack((flair, t1, t2, t1_ce))
return modalities
def load_patient() -> Tuple[np.ndarray, np.ndarray]:
flair = load_nifi_volume(
os.path.join(dataset_path, patient, f"{patient}_flair.nii.gz"))
t1 = load_nifi_volume(
os.path.join(dataset_path, patient, f"{patient}_t1.nii.gz"))
t1ce = load_nifi_volume(
os.path.join(dataset_path, patient, f"{patient}_t1ce.nii.gz"))
t2 = load_nifi_volume(
os.path.join(dataset_path, patient, f"{patient}_t2.nii.gz"))
masks = load_nifi_volume(
os.path.join(dataset_path, patient, f"{patient}_seg.nii.gz"))
modalities = np.asarray([t1, t1ce, t2, flair])
return modalities, masks
def get_brain_mask():
data = load_nifi_volume(os.path.join(dataset_path, patient,
f"{patient}_flair.nii.gz"),
normalize=False)
brain_mask = np.zeros(data.shape, np.float)
brain_mask[data > 0] = 1
return brain_mask
def get_brain_mask(self):
patient_path = os.path.join(self.data_path, self.patch_name)
data = load_nifi_volume(os.path.join(patient_path, self.flair), False)
brain_mask = np.zeros(data.shape, np.float)
brain_mask[data > 0] = 1
return brain_mask
def read_preds_from_models(model_list: list, patient_name: str) -> np.ndarray:
seg_maps = [
nifi_volume.load_nifi_volume(os.path.join(model_path, patient_name),
normalize=False)
for model_path in model_list
]
return np.stack(seg_maps)
def compute_normalization(input_dir, output_dir, ground_truth_path):
file_list = sorted(
[file for file in os.listdir(input_dir) if "unc" in file and "nii.gz"])
file_list_all = sorted(
[file for file in os.listdir(input_dir) if "nii.gz" in file])
max_uncertainty = 0
min_uncertainty = 10000
for uncertainty_map in tqdm(file_list,
total=len(file_list),
desc="Getting min and max"):
# Load Uncertainty maps
patient_name = uncertainty_map.split(".")[0].split("_unc")[0]
path_gt = os.path.join(ground_truth_path, patient_name,
f"{patient_name}_flair.nii.gz")
flair = nifi_volume.load_nifi_volume(path_gt, normalize=False)
brain_mask = np.zeros(flair.shape, np.float)
brain_mask[flair > 0] = 1
path = os.path.join(input_dir, uncertainty_map)
unc_map, _ = nifi_volume.load_nifi_volume_return_nib(path,
normalize=False)
tmp_max = np.max(unc_map[brain_mask == 1])
tmp_min = np.min(unc_map[brain_mask == 1])
if tmp_max > max_uncertainty:
max_uncertainty = tmp_max
if tmp_min < min_uncertainty:
min_uncertainty = tmp_min
for uncertainty_map_path in tqdm(file_list_all,
total=len(file_list_all),
desc="Normalizing.."):
path = os.path.join(input_dir, uncertainty_map_path)
output_path = os.path.join(output_dir, uncertainty_map_path)
unc_map, nib_data = nifi_volume.load_nifi_volume_return_nib(
path, normalize=False)
if "unc" in uncertainty_map_path:
uncertainty_map_normalized = brats_normalize(
unc_map, max_unc=max_uncertainty, min_unc=min_uncertainty)
print(f"Saving to: {output_path}")
nifi_volume.save_segmask_as_nifi_volume(uncertainty_map_normalized,
nib_data.affine,
output_path)
else:
nifi_volume.save_segmask_as_nifi_volume(unc_map, nib_data.affine,
output_path)
post_result = {"prediction": segmentation_post}
predict.save_predictions(data[idx], post_result, model_path,
f"{task}_post_processed")
results["prediction"] = prediction_map
predict.save_predictions(data[idx], results, model_path, task)
if compute_metrics:
patient_path = os.path.join(data[idx].data_path,
data[idx].patch_name, data[idx].seg)
data_path = os.path.join(data[idx].data_path, data[idx].patch_name,
data[idx].flair)
if os.path.exists(patient_path):
volume_gt = data[idx].load_gt_mask()
volume = nifi_volume.load_nifi_volume(data_path)
metrics = compute_wt_tc_et(prediction_map, volume_gt, volume)
logger.info(f"{data[idx].patient} | {metrics}")
print("Normalize uncertainty for brats!")
if uncertainty_flag:
input_dir = os.path.join(model_path, task)
output_dir = os.path.join(model_path, task, "normalized")
gt_path = data[0].data_path
compute_normalization(input_dir=input_dir,
output_dir=output_dir,
ground_truth_path=gt_path)
print("All done!!!! Be happy!")
def load_gt_mask(self) -> np.ndarray:
patient_path = os.path.join(self.data_path, self.patch_name)
volume = load_nifi_volume(os.path.join(patient_path, self.seg),
normalize=False)
return volume
def volume():
patient = "BraTS20_Training_001_p0_64x64x64"
gen_path = "/Users/lauramora/Documents/MASTER/TFM/Data/2020/train/random_tumor_distribution/"
volume_path = os.path.join(gen_path, patient, f"{patient}_seg.nii.gz")
return nifi_volume.load_nifi_volume(volume_path, normalize=False)
"subject_ID", "Grade", "Center", "Size", "Dice WT", "HD WT",
"Recall WT", "Precision WT", "F1 WT", "Dice TC", "HD TC",
"Recall TC", "Precision TC", "F1 TC", "Dice ET", "HD ET",
"Recall ET", "Precision ET", "F1 ET"
])
for patient in tqdm(data, total=len(data)):
patient_data = []
gt_path = os.path.join(patient.data_path, patient.patient,
f"{patient.seg}")
data_path = os.path.join(patient.data_path, patient.patient,
f"{patient.flair}")
prediction_path = os.path.join(
patient.data_path, patient.patient,
f"{patient.patient}_prediction.nii.gz")
if not os.path.exists(prediction_path):
print(f"{prediction_path} not found")
continue
patient_data.extend(
[patient.patient, patient.grade, patient.center, patient.size])
volume_gt_all, _ = nifi_utils.load_nifi_volume(gt_path)
volume_pred_all, _ = nifi_utils.load_nifi_volume(prediction_path)
volume, _ = nifi_utils.load_nifi_volume(data_path)
patient_data = compute_wt_tc_et(volume_pred_all, volume_gt_all,
volume)
writer.writerow(patient_data)
def volume():
patient = "BraTS20_Training_001"
gen_path = "/Users/lauramora/Documents/MASTER/TFM/Data/2020/train/no_patch/"
volume_path = os.path.join(gen_path, patient, f"{patient}_seg.nii.gz")
volume = nifi_utils.load_nifi_volume(volume_path)
return volume
output_dir = os.path.join(model_path, f"{task}/{setx}/normalized")
if not os.path.exists(output_dir):
os.makedirs(output_dir)
file_list = sorted([file for file in os.listdir(input_dir) if "unc" in file and "nii.gz"])
file_list_all = sorted([file for file in os.listdir(input_dir) if "nii.gz" in file])
max_uncertainty = 0
min_uncertainty = 10000
for uncertainty_map in tqdm(file_list, total=len(file_list), desc="Getting min and max"):
# Load Uncertainty maps
patient_name = uncertainty_map.split(".")[0].split("_unc")[0]
path_gt = os.path.join(ground_truth_path, patient_name, f"{patient_name}_flair.nii.gz")
flair = load_nifi_volume(path_gt, normalize=False)
brain_mask = np.zeros(flair.shape, np.float)
brain_mask[flair > 0] = 1
path = os.path.join(input_dir, uncertainty_map)
unc_map, _ = load_nifi_volume_return_nib(path, normalize=False)
tmp_max = np.max(unc_map[brain_mask == 1])
tmp_min = np.min(unc_map[brain_mask == 1])
if tmp_max > max_uncertainty:
max_uncertainty = tmp_max
if tmp_min < min_uncertainty:
min_uncertainty = tmp_min
