【Radiomics】Use 3Dslicer or Python to extract radiomics features

---

1. Using 3Dslicer software to extract radiomics features

1. Install plugin : SlicerRadiomics
2. Import image files : breast1_label.nrrd (mask file) and breast1_image.nrrd (personal image file).
   
3. Toggle plugin : Welcome to Slicer → Informatics → Radiomics
   
4. Setting parameters (as shown in the figure) :
   ◆ Select Input Volume and Segmentation
   input image Volume : breast1_image (personal image file)
   input regionsROI: breast1_label (mask file)
   ◆ Extraction Customization : Manual Customization
   Featrue Classesselects which class features to extract, such as firstorder and gldm
   Resampling and Filteringresampling→ Resampled voxel size: 3 ,3,3 (The distance between voxels is 3mm, it is recommended that the voxel intervals in each direction be the same) → LoG kernel sizes: 3,4 (If you use the Gaussian Laplace filter, you need to set it, you can set it multiple sizes). Check Wavelet-based features(whether to extract the features of the wavelet filter)
   Other default
   ◆ output
   output table : Table1 (modify the output table name), after setting the table name Apply.
   
5. After Apply, a table is generated in the lower right corner of the window :
   Image type: The original image is still a picture generated by some kind of filter. The previous diagnostics is the basic information of the image. Starting from log-sigma-3-0-mm-3D, it is useful information for feature extraction. (log filter, sigma size 3.0).
   Feature class: feature classification
   
6. Export table : format selection csv.
   
7. When doing radiomics research, cases are usually used as rows and features as columns, so it is necessary to transpose the csv table for subsequent research: select all data → select transpose when pasting to a new table.
   
   

---

2. Using python to extract radiomics features

• Basic version: To extract the characteristics of a case,
  you need to install pyradiomics first.
  Reference: [Radiomics] Problems with installing the pyradiomics package in windows system

  # pip install -i https://pypi.tuna.tsinghua.edu.cn/simple pyradiomics
  from radiomics import featureextractor 
  
  
  imageFile = 'file5/breast1/breast1_image.nrrd'
  maskFile = 'file5/breast1/breast1_label.nrrd'
  
  extractor = featureextractor.RadiomicsFeatureExtractor() # 初始化，给这一长串起个简写的名字
  featureVector = extractor.execute(imageFile, maskFile) # 
  # print(featureVector.items()) 结果太多太杂乱
  # for循环遍历提取所需的信息 featureName
  for featureName in featureVector.keys():
      print("%s: %s" % (featureName, featureName[featureName]))
  

  

● Common parameter setting

• Select by features to be extracted

  extractor.disableAllFeatures()  # 禁用所有特征
  extractor.enableFeaturesByName(firstorder=['Mean', 'Skewness']) # 只选几个特征提取
  
  featureVector = extractor.execute(imageFile, maskFile)
  for featureName in featureVector.keys():
      print("%s: %s" % (featureName, featureVector[featureName]))
  

  

• Select according to the type of feature to be extracted

  extractor.disableAllFeatures()  # 禁用所有特征
  extractor.enableFeatureClassByName('glcm') # 输出想要类型的特征，如 glcm
  
  featureVector = extractor.execute(imageFile, maskFile)
  for featureName in featureVector.keys():
      print("%s: %s" % (featureName, featureVector[featureName]))
  

• Filter settings : Extract features on the image with the filter (modified image).

  extractor.enableImageTypes(Original={
        
        }, LoG={
        
        }, Wavelet={
        
        }) 
  # Original原始图像 LoG滤波器 Wavelet小波滤波器
  
  featureVector = extractor.execute(imageFile, maskFile)
  for featureName in featureVector.keys():
      print("%s: %s" % (featureName, featureVector[featureName]))
  

  

• Batch processing to extract radiomics features

  import os
  import pandas as pd
  
  basePath = 'data/featureExtraction'
  folders = os.listdir(basePath) # 读取featureExtraction文件夹下所有文件名
  print(folders)
  
  df = pd.DataFrame() 
  for folder in folders:  # 遍历文件夹
      files = os.listdir(os.path.join(basePath,folder)) # 拼接文件夹下的文件路径
  #     print(files) 区分影像文件和 mask 文件
      for file in files:
          if file.endswith('image.nrrd'):  
              imageFile = os.path.join(basePath,folder,file) # 影像文件路径
          if file.endswith('label.nrrd'):
              maskFile = os.path.join(basePath,folder,file) # mask文件路径
  #     print(imageFile, maskFile) 特征提取
      extractor = featureextractor.RadiomicsFeatureExtractor()
      featureVector = extractor.execute(imageFile, maskFile)  # 得到结果是字典 dict
      # 从featureVector字典中提取数据转换成数据框，并 .T 转置（特征作为列）
      df_new = pd.DataFrame.from_dict(featureVector.values()).T  
      df_new.columns = featureVector.keys() # 列名
      df = pd.concat([df,df_new])  # 合并之前的数据框
  df.to_excel(os.path.join(basePath,'results.xlsx'))