什么ct要4000基于深度学习的CT图像肺结节自动检测技术二——训练数据处理

def normalize_hu(image):

    """
    将输入图像的像素值(-4000 ~ 4000)归一化到0~1之间
    :param image 输入的图像数组
    :return: 归一化处理后的图像数组
    """

    MIN_BOUND = -1000.0
    MAX_BOUND = 400.0
    image = (image - MIN_BOUND) / (MAX_BOUND - MIN_BOUND)
    image[image > 1] = 1.
    image[image < 0] = 0.
    return image


def load_patient_images(src_dir, wildcard="*.*", exclude_wildcards=[]):
    """
        读取一个病例的所有png图像，返回值为一个三维图像数组
    :param image 输入的一系列png图像
    :return: 三维图像数组
    """

    src_img_paths = glob.glob(src_dir + wildcard)
    for exclude_wildcard in exclude_wildcards:
        exclude_img_paths = glob.glob(src_dir + exclude_wildcard)
        src_img_paths = [im for im in src_img_paths if im not in exclude_img_paths]
    src_img_paths.sort()
    images = [cv2.imread(img_path, cv2.IMREAD_GRAYSCALE) for img_path in src_img_paths]
    images = [im.reshape((1,) + im.shape) for im in images]
    res = np.vstack(images)
    return res


def save_cube_img(target_path, cube_img, rows, cols):
    """
        将3维cube图像保存为2维图像,方便勘误检查
    :param 二维图像保存路径, 三维输入图像
    :return: 二维图像
    """

    assert rows * cols == cube_img.shape[0]
    img_height = cube_img.shape[1]
    img_width = cube_img.shape[1]
    res_img = np.zeros((rows * img_height, cols * img_width), dtype=np.uint8)

    for row in range(rows):
        for col in range(cols):
            target_y = row * img_height
            target_x = col * img_width
            res_img[target_y:target_y + img_height,
                    target_x:target_x + img_width] = cube_img[row * cols + col]

    cv2.imwrite(target_path, res_img)


if __name__ == '__main__':
    dicom_dir = './data/dicom_demo/'
   
    slices = load_patient(dicom_dir)                  # 读取dicom文件的元数据(dicom tags)
    
    pixel_spacing = slices[0].PixelSpacing           # 获取dicom的spacing值
    pixel_spacing.append(slices[0].SliceThickness)
    print('The dicom spacing : ', pixel_spacing)
    
    image = get_pixels_hu_by_simpleitk(dicom_dir)    # 提取dicom文件中的像素值
    # 标准化不同规格的图像尺寸, 统一将dicom图像缩放到1mm:1mm:1mm的尺度
    image = rescale_patient_images(image, pixel_spacing, 1.00)
    for i in tqdm(range(image.shape[0])):
        img_path = "./temp_dir/dcm_2_png/img_" + str(i).rjust(4, '0') + "_i.png"
        
        org_img = normalize_hu(image[i])             # 将像素值归一化到[0,1]区间
        
        cv2.imwrite(img_path, org_img * 255)         # 保存图像数组为灰度图(.png)

    # 加载上一步生成的png图像
    pngs = load_patient_images("./temp_dir/dcm_2_png/", "*_i.png")
    # 输入人工标记的结节位置: coord_x, coord_y, coord_z
    cube_img = get_cube_from_img(pngs, 272, 200, 134, 64)
    print(cube_img)
    save_cube_img('./temp_dir/chapter3_3dcnn_img_X.png', cube_img, 8, 8)