用于深度学习的医学图像预处理

      用于深度学习的医学图像数据，往往非常庞大，如果从网上下载公开数据集数据，往往有几十GB的图象数据，我们需要先进行预处理，将其转换成适合深度学习网络训练的形式：

       变为     

 或者我们还需要：Dicom变为nii，变为JPG；去除扫描床影响；为每个图象重命名等等。

          

 

一、文件批量重命名

 1、文件夹重命名：

import os,sys

def File_Rename(path):
    filelist = os.listdir(path)
    total_num = len(filelist)

    i = 0
    for file_name in filelist:
        os.rename((path + file_name), (path + str(i).zfill(2)))  # 子文件夹重命名
        print(file_name, "has been renamed successfully! New name is: ",str(i).zfill(2))
        i = i + 1

if __name__ == '__main__':
        path = r'E:/DeepLearningWorkSpace/PyCharmWorkSpace/Test/Bo_Test/thing/'  
        File_Rename(path)            #调用定义的函数

           

 2、文件重命名：dicom文件为例

import os

def rename(path):
    filelist = os.listdir(path)
    total_num = len(filelist)

    i = 0
    for item in filelist:
        if item.endswith('.dcm'):
            src = os.path.join(os.path.abspath(path), item)
            dst = os.path.join(os.path.abspath(path), str(i).zfill(3) + '.dcm')
            os.rename(src, dst)
            print(item, "has been renamed successfully! New name is: ", str(i).zfill(3) + '.dcm')
            i = i + 1
    print('total %d to rename & converted %d dcms' % (total_num, i))

if __name__ == '__main__':
    path = r'E:/DeepLearningWorkSpace/PyCharmWorkSpace/Test/Bo_Test/thing/00/'
    rename(path)

    

但是这里，出现了一个问题：没有按照顺序命名，它将倒数第二各10_384_time.....命名成000，但我想将1_384_time....命名为000，然后依次往下。

（这里明白了，计算机是按照第一位，第二位，这样一位一位识别的，10_,11_,1_,先识别10_，然后是11_,和1_。所以以后对文件命名时，尽量写成001，002.....010。而且再重命名前做好备份，以防万一。）

现在遇到上面这种情况，可以：

获取文件名   中：_384前面的内容，再命名。

file_name = '10_384_time_20210824-16-11-58-316.dcm'
print(file_name[0:file_name.rfind('_384')])

>>10

3、 批量文件夹内文件重命名：

import os

def rename(path):
    filelist = os.listdir(path)
    total_num = len(filelist)

    i = 0
    for item in filelist:
        if item.endswith('.dcm'):
            src = os.path.join(os.path.abspath(path), item)
            dst = os.path.join(os.path.abspath(path), str(i)zfill(3) + '.dcm')
            os.rename(src, dst)
            print('converting %s to %s ...' % (src, dst))
            i = i + 1
    print('total %d to rename & converted %d dcms' % (total_num, i))

Dir_pathes = 'E:/DeepLearningWorkSpace/PyCharmWorkSpace/date/NBIA_lung/CBCT'
Dir_list = os.listdir(Dir_pathes)
Dir_num = len(Dir_list)

for k in Dir_list:
    path = os.path.join(Dir_pathes,k)
    rename(path)

然后对文件重命名结束之后，可以对dicom文件进行处理。

二、移除扫描床处理

Fuction_quchuang.py 

import matplotlib.pyplot as plt
import numpy as np
import pydicom
import cv2
import SimpleITK as sitk
from PIL import Image

from skimage.morphology import disk, rectangle, binary_dilation, binary_erosion, binary_closing, binary_opening

def read_dicom_data(file_name):
    file = sitk.ReadImage(file_name)
    data = sitk.GetArrayFromImage(file)
    print(data.shape)
    data = np.squeeze(data, axis=0)
    print(data.shape)
    data = np.int32(data)

    dicom_dataset = pydicom.dcmread(file_name)
    slice_location = dicom_dataset.SliceLocation         #获取层间距
    return data, data.shape[0], data.shape[1], slice_location

def window(window, img_data):
    if window == 'Lung':
        img_data[img_data < -1150] = -1150
        img_data[img_data > 350] = 350
    elif window == 'Pelvic':
        img_data[img_data < -138] = -138
        img_data[img_data > 238] = 238
    elif window == 'Chest':
        img_data[img_data < -160] = -160
        img_data[img_data > 240] = 240
    elif window == 'Chest_scatter':
        img_data[img_data < -752] = -752
        img_data[img_data > 838] = 838
    elif window == 'Pelvic_scatter':
        img_data[img_data < -300] = -300
        img_data[img_data > 240] = 240
    else:
        img_data[img_data < 0] = 0
        img_data[img_data > 80] = 80
    return img_data

def find_max_region(mask_sel):
    contours, hierarchy = cv2.findContours(mask_sel, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
    # 找到最大区域并填充
    area = []
    for j in range(len(contours)):
        area.append(cv2.contourArea(contours[j]))
    max_idx = np.argmax(area)
    max_area = cv2.contourArea(contours[max_idx])
    for k in range(len(contours)):
        if k != max_idx:
            cv2.fillPoly(mask_sel, [contours[k]], 0)
    return mask_sel

def quchuang(dcm_path,window1,window2):
    # 读入dicom图像
    dcm_path = dcm_path
    pixel_array, rows, columns, slice_location = read_dicom_data(dcm_path)
    pixel_array1 = pixel_array.copy()
    pixel_array2 = pixel_array.copy()

    pixel_array1 = window(window1, pixel_array1)
    imageData1 = (pixel_array1 - pixel_array1.min()) * 255.0 / (pixel_array1.max() - pixel_array1.min())
    imageData1 = np.uint8(imageData1)

    pixel_array2 = window(window2, pixel_array2)
    imageData2 = (pixel_array2 - pixel_array2.min()) * 255.0 / (pixel_array2.max() - pixel_array2.min())
    imageData2 = np.uint8(imageData2)

    # 二值化
    ret, binary = cv2.threshold(imageData2, 3, 255, cv2.THRESH_BINARY)

    # 腐蚀
    selem = disk(3)
    fushi = binary_erosion(binary, selem)

    # 找最大连通区域
    binary = np.uint8(fushi)
    max_region = find_max_region(binary)

    # 膨胀
    selem = disk(3)
    pengzhang = binary_dilation(max_region, selem)
    
    #最左最右边两列像素值为0
    pengzhang[:, 0] = 0
    pengzhang[:, 511] = 0
    
    #填充
    A = np.uint8(pengzhang)
    h, w = A.shape[:2]
    #print(pengzhang.shape[:2])
    mask_tp = np.zeros((h + 2, w + 2), np.uint8)
    temp = A.copy()
    temp2 = A.copy()
    cv2.floodFill(temp, mask_tp, (1, 1), 255)
    cv2.floodFill(temp2, mask_tp, (w - 2, h - 2), 255)
    rt = cv2.bitwise_not(temp)

    Tianchong = rt
    Tianchong[rt > 0] = 255

    image_process = np.uint8((Tianchong / 255) * imageData1)
    return image_process

if __name__ == '__main__':
    dcm_path = "E:/DeepLearningWorkSpace/PyCharmWorkSpace/date/CBCT_Pancreatic/22/CT/22/24.dcm"
    window1 = 'Pelvic'
    window2 = 'Pelvic_scatter'
    image_array = quchuang(dcm_path, window1, window2)
    #image = Image.fromarray(image_array)
    save_path = 'E:/DeepLearningWorkSpace/PyCharmWorkSpace/date/CBCT_Pancreatic/JPG/buchong'
    cv2.imwrite(save_path + '/22_24.jpg', image_array)

 以上是处理一幅dicom并将其保存为JPG格式的代码。

from Function_quchuang import read_dicom_data,window,find_max_region,quchuang  #调用上一个代码函数
import os
import cv2
import re

Dir_pathes = 'E:/DeepLearningWorkSpace/PyCharmWorkSpace/date/Pancreatic/22/CT/'
Dir_list = os.listdir(Dir_pathes)
Dir_num = len(Dir_list)

for k in Dir_list:
    path = os.path.join(Dir_pathes, k)
    dir_list = os.listdir(path)
    dir_num = len(dir_list)
    for j in dir_list:
        window1 = 'Pelvic'
        window2 = 'Pelvic_scatter'
        image_array = quchuang(path + '/' + j, window1, window2)
        save_path = 'E:/DeepLearningWorkSpace/PyCharmWorkSpace/date/Pancreatic/JPG/train_B'
        cv2.imwrite(save_path + '/' + k + '_'+ re.sub("\D", "", j) +'.jpg', image_array)
        print('%s libingren, di %s fu image' % (k, j))

 以上是批量处理移除扫描床的代码.

一些去床的具体内容在： DICOM的理解与学习2_张小懒君的博客-CSDN博客