kaggle CBIS-DDSM 依照breast density分类图片

Question

kaggle上的CBIS-DDSM下面又有两个文件夹[csv， jpeg]。 csv文件夹中有六个文件分别为：[calc_case_description_test_set.csv calc_case_description_train_set.csv dicom_info.csv mass_case_description_test_set.csv mass_case_description_train_ set.csv meta.csv]。 另外， jpeg 文件夹中分别有好几个名子类似为“1.3.6.1.4.1.9590.100.1.2.100018879311824535125115145152454291132”的文件夹里面有“cropped images”，“full mammogram images”，“ROI mask images”。 请帮我依据“breast density”的等级分成四个文件夹1 2 3 4。 并且1 2 3 4文件夹中又个别有子文件夹“cropped images”，“full mammogram images”，“ROI mask images”。
声明： 请勿用chatGPT， 因为他给的答案是错的

Answer 1

import os
import shutil
import pandas as pd
import re
import cv2

#区分roi和cropped图片
def classify_img(filename):
    img=cv2.imread(filename)
    gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    r, c = gray_img.shape[:2]
    piexs_sum = r * c

    dark_points = (gray_img < 20)
    target_array = gray_img[dark_points]
    dark_sum = target_array.size
    dark_prop = dark_sum / (piexs_sum)
    if dark_prop >= 0.85:
        return True
    else:
        return False
    
#复制文件函数
def copy_files(dd):
    global num
    for n in range(len(dd)):
        #取出full images的路径
        d=re.findall(r'/(.*)/', dd.loc[n,'image file path'])[0]
        a=re.findall(r'/(.*)', d)[0]
        old_path=f"D:/workstation/CBIS_DDSM/jpeg/{a}"
        patient_id=dd.loc[n,'patient_id']
        abnormality_id=dd.loc[n,'abnormality id']
        left_or_right=dd.loc[n,'left or right breast']
        pathology=dd.loc[n,'pathology']
        _class={'BENIGN_WITHOUT_CALLBACK':0,'BENIGN':1,'MALIGNANT':2}
        view=dd.loc[n,'image view']
        density=dd.iat[n,1]
        if os.path.exists(old_path):
            full_filenames=os.listdir(old_path)
            #完整图片只有一个
            for i in full_filenames:
                shutil.copy(f"{old_path}/{i}", f'd:/workstation/images/{density}/full/{patient_id}_{left_or_right}_{view}_{abnormality_id}_{_class[pathology]}_{density}.jpg')
                num+=1
        #取出cropped 和roi image的路径，同一个路径下        
        d=re.findall(r'/(.*)/', dd.loc[n,'cropped image file path'])[0]
        a=re.findall(r'/(.*)', d)[0]
        old_path=f"D:/workstation/CBIS_DDSM/jpeg/{a}"
        #裁剪图片和标注图片各一个，同一个文件夹下，使用图像识别
        if os.path.exists(old_path):
            cropped_filenames=os.listdir(old_path)
            for i in cropped_filenames:
                if classify_img(f"{old_path}/{i}"):
                    shutil.copy(f"{old_path}/{i}", f'd:/workstation/images/{density}/roi/{patient_id}_{left_or_right}_{view}_{abnormality_id}_{_class[pathology]}_{density}.jpg')
                    num+=1
                else:
                    shutil.copy(f"{old_path}/{i}", f'd:/workstation/images/{density}/cropped/{patient_id}_{left_or_right}_{view}_{abnormality_id}_{_class[pathology]}_{density}.jpg')
                    num+=1
                    
name_list={'calc_test':'calc_case_description_test_set','calc_train':'calc_case_description_train_set',\
           'mass_test':'mass_case_description_test_set','mass_train':'mass_case_description_train_set'}
#创建4个文件夹及其子文件夹
for i in range(5):
    os.makedirs(f"d:/workstation/images/{i}/full",exist_ok=True)
    os.makedirs(f"d:/workstation/images/{i}/cropped",exist_ok=True)
    os.makedirs(f"d:/workstation/images/{i}/roi",exist_ok=True)
num=0#记录复制文件数
#每个csv的结构是一样的，所以调用同一个函数
for i in name_list.keys():
    dd=pd.read_csv(f'D:/workstation/CBIS_DDSM/csv/{name_list[i]}.csv')
    copy_files(dd)
    print(f'复制了{num}个文件')