def make_overlap_us_seg(us_path, seg_paths, save_path, pass_without_seg=True):
us_path = norm_path(us_path)
seg_paths = [norm_path(p) for p in seg_paths]
save_path = norm_path(save_path)
if not os.path.exists(save_path):
os.makedirs(save_path)
us_dict = image_dict(us_path)
key = lambda p: os.path.splitext(os.path.split(p)[-1])[0].split('#')[0]
seg_dict_list = [image_dict(p, key=key) for p in seg_paths]
for name in us_dict:
seg_path_list = []
for seg_dict in seg_dict_list:
if name in seg_dict:
seg_path_list.append(seg_dict[name])
else:
seg_path_list.append(None)
if pass_without_seg and None in seg_path_list:
continue
img = overlap_us_seg(us_dict[name], seg_path_list)
save_file = os.path.join(save_path, name + '.png')
print(save_file)
cv2.imwrite(save_file, img)
def getNonKidneyFiles(self, path):
us_files = image_dict(path)
seg_files = image_dict(args.seg_data)
non_kidney_files = [
us_files[file] for file in us_files if file not in seg_files
]
return non_kidney_files
def getKidneyFiles(self, path):
us_files = image_dict(path)
seg_files = image_dict(args.seg_data)
kidney_files = [
us_files[file] for file in us_files if file in seg_files
]
return kidney_files
def load_kidney(self, dataset_dir, subset, seg_dir=None):
"""Load a subset of the nuclei dataset.
dataset_dir: Root directory of the dataset
subset: Subset to load. Either the name of the sub-directory,
such as stage1_train, stage1_test, ...etc. or, one of:
* train: stage1_train excluding validation images
* val: validation images from VAL_IMAGE_IDS
"""
if subset != "":
dataset_subset_dir = os.path.join(dataset_dir, subset)
else:
dataset_subset_dir = dataset_dir
self.class_list = [
'bg', # class_id 1
'kidney'
] # class_id 2
# add classes
for id, name in enumerate(self.class_list, start=1):
self.add_class("kidney", id, name)
print("add class: {} {}".format(id, name))
# seg_dir is None on detect mode
if seg_dir:
mask_dict = dict()
for id, name in enumerate(self.class_list, start=1):
mask_class_dict = image_dict(os.path.join(seg_dir, name))
for _name, _path in mask_class_dict.items():
mask_dict[_name] = [id, _path]
print("find mask id:{} name:{} count:{}".format(
id, name, len(mask_class_dict)))
# Add images
for file in image_list(dataset_subset_dir):
_, name, _ = split_path(file)
if name not in mask_dict.keys():
continue
class_id, mask_path = mask_dict[name]
self.add_image(source="kidney",
image_id=name,
path=file,
mask_path=mask_path,
class_id=class_id)
else:
image_files = image_list(dataset_subset_dir)
# Add images
for file in image_files:
_, name, _ = split_path(file)
self.add_image(source="kidney",
image_id=name,
path=file,
mask_path=None,
class_id=None)
print('count added image', subset, len(self.image_info))
def load_kidney(self, dataset_dir, subset, seg_dir=None):
"""Load a subset of the nuclei dataset.
dataset_dir: Root directory of the dataset
subset: Subset to load. Either the name of the sub-directory,
such as stage1_train, stage1_test, ...etc. or, one of:
* train: stage1_train excluding validation images
* val: validation images from VAL_IMAGE_IDS
"""
# Add classes. We have one class.
# Naming the dataset nucleus, and the class nucleus
self.add_class("kidney", 1, "kidney")
# Which subset?
# "val": use hard-coded list above
# "train": use data from stage1_train minus the hard-coded list above
# else: use the data from the specified sub-directory
# assert subset in ["train", "val", ""]
if subset != "":
dataset_subset_dir = os.path.join(dataset_dir, subset)
else:
dataset_subset_dir = dataset_dir
# seg_dir is None on detect mode
if seg_dir:
# read mask path
mask_dict = image_dict(seg_dir)
# read image path
image_files = image_list(dataset_subset_dir)
# Add images
for file in image_files:
_, name, _ = split_path(file)
# detect mode ?
if seg_dir:
# only use image with mask
if name not in mask_dict.keys():
continue
mask_path = mask_dict[name]
else:
mask_path = None
self.add_image(
source="kidney",
image_id=name,
path=file,
mask_path=mask_path)
print('cnt img', subset, len(self.image_info))
def main(us_path,
mask_path,
result_path,
scale=None,
view_scale=None,
use_top_n_greatest_width=True,
size=None,
horizontally=True,
crop=False,
denoise=True):
'''
center boolean: 신장의 중앙 정렬 여부
scale float: 영상 축적 고정
only_greatest_width boolean: 신장 영상중 가장 큰것만 사용할지 여부
resize (W, H): 최종 이미지 크기
'''
us_dict = image_dict(norm_path(us_path))
mask_dict = image_dict(norm_path(mask_path))
result_path = norm_path(result_path, True)
for patient, dicoms in Per_patient_fast():
patient_datas = list()
for dicom in dicoms:
Name = dicom['Name'][:-4] # remove file ext
Diagnosis = dicom['Diagnosis']
AccNo = dicom['AccNo']
PhysicalUnitsXDirection = dicom['PhysicalUnitsXDirection']
PhysicalUnitsYDirection = dicom['PhysicalUnitsYDirection']
PhysicalDeltaY = dicom['PhysicalDeltaY'] # cm per pixels
PhysicalDeltaX = dicom['PhysicalDeltaX'] # 1 픽셀이 몇 cm 인지 나타냄
# checking the kidney
if Name not in mask_dict:
# pass the non-kidney images
continue
# for debug
# if Name != '1.2.840.113663.1500.1.295077244.3.8.20101014.92756.625':
# continue
# checking physical unit(mm)
if PhysicalUnitsXDirection != '3' or PhysicalUnitsYDirection != '3' \
and not PhysicalDeltaY and not PhysicalDeltaX:
continue
PhysicalDeltaY = float(PhysicalDeltaY)
PhysicalDeltaX = float(PhysicalDeltaX)
# read image
us_img = cv2.imread(us_dict[Name], cv2.IMREAD_GRAYSCALE)
print(us_dict[Name])
mask_img = cv2.imread(mask_dict[Name], cv2.IMREAD_GRAYSCALE)
# assert scale != view_scale, "use only scale or view_scale"
if horizontally:
angle = calculate_angle(mask_img)
mask_img = rotate_bound(mask_img, angle)
us_img = rotate_bound(us_img, angle)
if scale:
us_img = resize_physical_unit(us_img,
(PhysicalDeltaX, PhysicalDeltaY),
scale)
mask_img = resize_physical_unit(
mask_img, (PhysicalDeltaX, PhysicalDeltaY), scale)
if view_scale:
bbX, bbY, bbW, bbH = find_bounding_square(mask_img)
fx = view_scale[0] / bbW
fy = view_scale[1] / bbH
us_img = cv2.resize(us_img, None, fx=fx, fy=fy)
mask_img = cv2.resize(mask_img, None, fx=fx, fy=fy)
if size:
#numpy 좌표값을 얻어낸다 -> argwhere
mask_pts = np.argwhere(mask_img == 255)
#평균을 구한다
mask_cx, mask_cy = int(np.mean(mask_pts[:, 1])), int(
np.mean(mask_pts[:, 0]))
#캔버스를 만들고 붙혀넣는다
us_canvas = np.zeros(size, dtype=np.uint8)
mask_canvas = np.zeros(size, dtype=np.uint8)
# # for debug
# us_canvas[:] = 100
# mask_canvas[:] = 100
canvas_h, canvas_w = size
canvas_cx, canvas_cy = canvas_w // 2, canvas_h // 2
dst_x, dst_y = canvas_cx - mask_cx, canvas_cy - mask_cy
us_img = paste(dst=us_canvas,
src=us_img,
dst_x=dst_x,
dst_y=dst_y)
mask_img = paste(dst=mask_canvas,
src=mask_img,
dst_x=dst_x,
dst_y=dst_y)
if crop:
mask = mask_img.astype(np.bool)
us_img = us_img * mask
if denoise:
us_img = cv2.fastNlMeansDenoising(us_img, None, 10, 7, 21)
data = dict()
data['us'] = us_img
data['mask'] = mask_img
data['info'] = dicom
patient_datas.append(data)
# sort by kidney value
patient_datas = sorted(
patient_datas,
key=lambda x: np.count_nonzero(x['mask'] == 255),
reverse=True)
# use one or all
if use_top_n_greatest_width:
cut_n = use_top_n_greatest_width
patient_datas = patient_datas[:cut_n]
accNo = patient['AccNo']
diagnosis = patient['Diagnosis']
train_val = 'val' if accNo in isangmi_accno_list else 'train'
save_path = os.path.join(result_path, train_val, diagnosis, accNo)
''' if args.preprocess_denoise:
np_out = np.asarray(out)
np_out = cv2.fastNlMeansDenoising(np_out, None, 10, 7, 21)
out = Image.fromarray(np.uint8(np_out))
return out'''
# save image
for order, data in enumerate(patient_datas):
us_img = data['us']
name = data['info']['Name'][:-4]
save_file = os.path.join(save_path, name + '.png')
if not os.path.exists(save_path):
os.makedirs(save_path)
print(save_file)
cv2.imwrite(save_file, us_img)
# save order of images
if patient_datas:
save_file = os.path.join(save_path, 'order.txt')
with open(save_file, 'wt') as f:
for order, data in enumerate(patient_datas):
name = data['info']['Name'][:-4]
f.write(name + '\n')