Duke liver datasets
如果大家是做医学图像处理的,刚好又是做肝脏那边的,对这个数据集应该不会陌生
我发现网上现有的资料很少有说怎么读取这个数据集的,由于我之前一直都是做的nii格式的数据集,现在碰到了dicom格式的数据集页研究了几天,终于把这个数据集如何创建dataset等给弄出来了
数据集部分文件夹
> `class DukeLiverDataset(Dataset):
def __init__(self, root_dir, csv_file, transform=None):
"""
初始化数据集
:param root_dir: 数据集的根目录,例如 'D:\\datasets\\Duke Liver Dataset (MRI) v2\\Segmentation'
:param csv_file: 包含序列和标签信息的CSV文件路径
:param transform: 对图像应用的转换(预处理或数据增强)
"""
self.root_dir = root_dir
self.transform = transform
self.data_frame = pd.read_csv(csv_file)
self.series_image_paths = []
self.series_mask_paths = []
self.labels = []
for _, row in self.data_frame.iterrows():
patient_id = format(row['DLDS'], '04d')
series_id = row['Series']
label = row['Label']
series_path = os.path.join(self.root_dir, patient_id, str(series_id))
if os.path.exists(series_path):
image_dir = os.path.join(series_path, 'images')
mask_dir = os.path.join(series_path, 'masks')
image_paths = [os.path.join(image_dir, f) for f in os.listdir(image_dir) if f.endswith('.dicom')]
mask_paths = [os.path.join(mask_dir, f) for f in os.listdir(mask_dir) if f.endswith('.dicom')]
self.series_image_paths.append(sorted(image_paths))
self.series_mask_paths.append(sorted(mask_paths))
self.labels.append(label)
def __len__(self):
return len(self.series_image_paths)
def __getitem__(self, idx):
label = self.labels[idx]
# 读取所有的DICOM图像和对应的DICOM掩码
image_files = self.series_image_paths[idx]
mask_files = self.series_mask_paths[idx]
images = [pydicom.dcmread(img_file).pixel_array for img_file in image_files]
masks = [pydicom.dcmread(mask_file).pixel_array for mask_file in mask_files]
# 转换为NumPy数组
images = np.array(images, dtype=np.float32)
masks = np.array(masks, dtype=np.float32)
location = _get_annotation_float(masks)
transformed_images = []
if self.transform is not None:
for img in images:
# 将 numpy 数组转换为 PIL 图像
img = Image.fromarray(np.uint8(img))
# 应用定义的转换
img = self.transform(img)
# 将转换后的图像添加到列表中
transformed_images.append(img)
transformed_images = torch.stack(transformed_images)
transformed_images = transformed_images.permute(1, 0, 2, 3)
return transformed_images, location
def get_series_path(self, idx):
return self.series_image_paths[idx][0] # 返回第一个图像的路径作为代表
def get_path_details(self, idx):
# 从保存的信息中返回特定索引的详细路径
image_dir = os.path.join(self.series_image_paths[idx][0]) # 取第一个图像的路径作为示例
series_directory = os.path.dirname(os.path.dirname(image_dir)) # 回退两级目录
return series_directory`