def __getitem__(self, idx):
imgdir = self.testImg[idx]
img_arr = dicom2array(imgdir) # 获取具体的图片数据，二维数据
# (hieght * width)
origi_shape = np.array(img_arr.shape)
if img_arr.shape != (256, 256):
img_aug = cv2.resize(img_arr, (256, 256))
else:
img_aug = img_arr
# 'studyUid','seriesUid','instanceUid','zindx'
tag_list = ['0020|000d', '0020|000e', '0008|0018', '0020|0013']
studyUid, seriesUid, instanceUid, zindx= dicom_metainfo(imgdir, tag_list)
if self.transform is not None:
img_aug = self.transform(img_aug) # 在这里做transform，转为tensor等等
# print(img_aug.shape)
# print(img_aug)
return origi_shape, img_aug, instanceUid, seriesUid, studyUid, zindx
def dicoｍ2TestIdx(img_file):
"""
input:some ill people's folder which contain the dicom img
return:{"studyUid":001,"SOPInstaceUID":4.286.2,"img_dir":"./data/test.dic"}
"""
img_fileName = os.listdir(img_file)
# make the img list sort by id,like [img1.dcm,img2.dcm,img3.dcm,……]
img_fileName.sort(key=lambda x: int(x[5:-4]))
# split the file by dfferent seriesDescription
# seriesNumber ,InstanceNumber, StudyInstanceUID, other info tag :'0008|103e' '0002|0003'
GroupElement=['0020|0011', '0020|0013', '0020|000d']
res = []
for i in range(1, len(img_fileName)+1):
dicImg_path = os.path.join(img_file, img_fileName[i-1])
# img_info have three item[]
img_info = dicom_metainfo(dicImg_path, GroupElement)
img_info.append(dicImg_path)
print(img_info)
res.append(img_info)
return res
def get_info(trainPath, jsonPath):
"""
:rturn: 图像的位置和标注信息
"""
# path = "./train150.npy" if "150" in trainPath else "./valid50.npy"
# path = "./train203.npy" if "203" in trainPath else "./valid150.npy"
# global config
# if config["valid"] == 1:
# if "round2train" in trainPath:
# path="./round2train_pc.npy"
# if "round2_valid53" in trainPath:
# path = "./round2_valid53_pc.npy"
# else:
path = "noSuchfile"
if not os.path.exists(path):
annotation_info = pd.DataFrame(columns=('studyUid', 'seriesUid', 'instanceUid', 'annotation'))
# print(jsonPath)
# print(os.getcwd())
json_df = pd.read_json(jsonPath)
for idx in tqdm(json_df.index):
studyUid = json_df.loc[idx, "studyUid"]
seriesUid = json_df.loc[idx, "data"][0]['seriesUid']
instanceUid = json_df.loc[idx, "data"][0]['instanceUid']
annotation = json_df.loc[idx, "data"][0]['annotation']
row = pd.Series(
{'studyUid': studyUid, 'seriesUid': seriesUid, 'instanceUid': instanceUid, 'annotation': annotation})
annotation_info = annotation_info.append(row, ignore_index=True)
dcm_paths = glob.glob(os.path.join(trainPath, "**", "**.dcm")) # 具体的图片路径
# print(dcm_paths)
# 'studyUid','seriesUid','instanceUid'
tag_list = ['0020|000d', '0020|000e', '0008|0018']
# ["检查实例号：唯一标记不同检查的号码.", "序列实例号：唯一标记不同序列的号码.", "SOP实例"]
dcm_info = pd.DataFrame(columns=('dcmPath', 'studyUid', 'seriesUid', 'instanceUid'))
for dcm_path in dcm_paths:
try:
studyUid, seriesUid, instanceUid = dicom_metainfo(dcm_path, tag_list)
row = pd.Series(
{'dcmPath': dcm_path, 'studyUid': studyUid, 'seriesUid': seriesUid, 'instanceUid': instanceUid})
# print("try: ", dcm_path)
dcm_info = dcm_info.append(row, ignore_index=True)
except:
# print("except: ", dcm_path)
continue
result = pd.merge(annotation_info, dcm_info, on=['studyUid', 'seriesUid', 'instanceUid'])
# result = result.set_index('dcmPath')['annotation'] # 然后把index设置为路径，值设置为annotation
result = result[["dcmPath", "annotation"]].values
idx_wrong = []
print("result len:",len(result))
for idxx in range(len(result)):
annotation = result[idxx][1] # 获取图片的标签
if len(annotation[0]['data']['point']) == 11:
continue
else:
idx_wrong.append(idxx)
result_new = np.delete(result, idx_wrong, 0)
print("result_new len:", len(result_new))
#delete some row which label wrong
np.save(path, result_new)
else:
result_new = np.load(path, allow_pickle=True)
return result_new