Revert "Initial commit"

This reverts commit c604086.

mar/reconstruct_nii.py

@@ -0,0 +1,201 @@
+import numpy as np
+import os                #遍历文件夹
+import nibabel as nib    #nii格式一般都会用到这个包
+from PIL import Image
+import cv2
+import h5py
+from typing import List
+import re
+'''
+**MAR usage**
+将二维图像序列重建为3D nii数组
+
+说明：
+    输入input_folder 为二维图像序列文件夹；
+    输出目录output_folder指定了输出文件夹，配合output_name指定输出的文件名，确定保存的文件路径；
+    refer_file为每个重建的数组对应的原CT文件，主要提取其中的位置信息来构建标注；
+    head 为头部插入空白层数，tail 为尾部插入空白层数，为0就表示不插入序列。最终要保证结果的层数与ct的层数一致才能打开。
+'''
+## ---- ****
+# 注：在此修改相关参数。
+input_folder = r"outputH5\output-data3"
+output_folder = r"outputH5"
+output_name = 'pred-data3.nii.gz'
+
+refer_file = r"data2.nii.gz"
+# head = 408
+# tail = 626-569
+
+## 获得读入的图片所在原CT的位置，其他位置插入0序列补全，这个通常用于补充分割mask的结果
+slice_start = 0
+slice_end = 608
+
+# 整个nii文件的slice数量
+total_length = 699
+# mode 处理模式：如果是png图片用"image", 如果是h5文件用"h5"
+mode = "image"
+## ----
+
+
+
+
+## -----------------
+def read_image_tonumpy(filename: str):
+    """Read an image in numpy format"""
+    img = Image.open(filename)
+    # 转换成np.ndarray格式
+    img_array = np.array(img)
+    img.close()
+    return img_array
+
+def read_niifile(nii_filepath: str):
+    # return 3D image
+    img_nii = nib.load(nii_filepath)  # 读取nii
+    return img_nii
+
+def get_all_files(folder: str):
+    """Get Image Files with number id in an folder.
+    Does not contain path, only name.
+    注：测试时每次只能测试一个实例，因为文件名特别针对数字进行了排序，不能使用多个实例。
+    """
+    namelist = [name for name in os.listdir(folder)]
+    if len(namelist) == 0:
+        return namelist
+    # to contruct re expression
+    rexp = re.compile(r"\d+")
+    # sort by number
+    namelist.sort(key=lambda s: int(re.findall(rexp, s)[0]) )
+    return namelist
+
+def get_3darray(filelist: List[str]):
+    # filelist = [name for name in os.listdir(folder)]
+    # filelist = [os.path.join(folder, name) for name in os.listdir(folder)]
+    datalist = []
+    for filename in filelist:
+        img_array = read_image_tonumpy(filename)
+        ## 镜像翻转，矩阵需要进行转置复原
+        img_array = img_array.T
+        datalist.append(img_array)
+
+    data = np.asarray(datalist)
+    return data
+
+def insert_blankarray(data_array: np.ndarray, head: int, tail: int):
+    """
+    在3d数组前后插入一定数量的序列。head为头部插入的层数（数量>=0），tail为尾部插入的层数（数量>=0）。
+    """
+    z,y,x = data_array.shape
+    if head > 0:
+        head_part = np.zeros((head, y, x), dtype=data_array.dtype)
+        data_array = np.concatenate([head_part, data_array], axis=0)
+    if tail > 0:
+        tail_part = np.zeros((tail, y, x), dtype=data_array.dtype)
+        data_array = np.concatenate([data_array, tail_part], axis=0)
+
+    return data_array
+
+def save_to_nii(data_array: np.ndarray, filename: str, affine_array: np.ndarray):
+    ## refer_data is nii_img
+    data_array = data_array.astype(np.float)
+    nii_img = nib.Nifti1Image(data_array, affine_array)
+    nib.save(nii_img, filename)
+
+
+
+
+
+
+############ For MAR usage
+
+def read_exported_datah5_slices(filename: str, normalize=False, window_size = (None, None), mask_threshold: float or None = None):
+    '''
+    Use for test process. `window_size` is the tuple with (minHU, maxHU).
+    Currently do not truncate values.
+    '''
+    with h5py.File(filename, mode='r') as f:
+        data_ma = np.array(f['image']) # read with shape (H, W)
+
+        if mask_threshold is None:
+            data_metalmask = np.array(f['metal_mask']) # read with shape (H, W)
+        else:
+            data_metalmask = data_ma > mask_threshold
+
+    if normalize and window_size[1] is not None:
+        minval, maxval = window_size
+        # minval = minval / 1000 * miuWater + miuWater
+        # maxval = maxval / 1000 * miuWater + miuWater
+
+        data_ma = (data_ma - minval) / (maxval - minval)
+
+
+    # return all data with single image shape (H, W)
+    data = {
+        "filename": filename,
+        # "gt_image": data_gt,
+        "ma_image": data_ma,
+        "metal_mask": data_metalmask,
+
+    }
+    return data
+
+def get_3darray_fromH5file(filelist: List[str]):
+    """
+    Read image array from h5 files.
+    """
+    # filelist = [name for name in os.listdir(folder)]
+    # filelist = [os.path.join(folder, name) for name in os.listdir(folder)]
+    datalist = []
+    for filename in filelist:
+        img_array = read_exported_datah5_slices(filename, normalize=False)['ma_image']
+        ## 镜像翻转，矩阵需要进行转置复原
+        img_array = img_array.T
+        # img_array = img_array.astype(float)
+        # img_array = cv2.resize(img_array, dsize=(obj_size, obj_size), interpolation=cv2.INTER_LINEAR)
+        datalist.append(img_array)
+
+    data = np.asarray(datalist)
+    # data = np.int32(data.round())
+    return data
+
+
+
+
+
+
+
+def main():
+    # filelist = ['GAO JIA1', 'GAO JIA2', 'GAO JIA10', 'GAO JIA12', 'GAO JIA15']
+    namelist = get_all_files(input_folder)
+    filelist = [os.path.join(input_folder, name) for name in namelist]
+    print("Processing folder: ", os.path.split(input_folder)[1])
+    # data_array = get_3darray(filelist)
+    if mode == "image":
+        data_array = get_3darray(filelist)
+    elif mode == "h5":
+        data_array = get_3darray_fromH5file(filelist)
+    else:
+        data_array = get_3darray(filelist)
+
+    ## insert blank part
+    head_num = slice_start - 0
+    tail_num = total_length - 1 - slice_end
+    if head_num > 0 or tail_num > 0:
+        data_array = insert_blankarray(data_array, head_num, tail_num)
+
+    print("处理完的层数一共有：",data_array.shape[0])
+    # transpose the shape
+    data_array = np.transpose(data_array, [1,2,0])
+    print("处理完的数组形状为：", data_array.shape)
+
+    ## save to nii_file
+    refer_data = nib.load(refer_file)
+    save_to_nii(data_array, filename=os.path.join(output_folder, output_name), affine_array=refer_data.affine )
+
+    print("End of the process. ")
+
+
+
+if __name__ == '__main__':
+    main()
+
+