医学分割十项全能挑战，是从nnunet那里听说来的，下载心脏数据集之后发现有19个训练集，10个测试集，label标记为0背景1左心房。

数据集是TCIA比赛中选出来的，由于是MRI的nii.gz格式，并且图像的spacing都是一致的，打算先进行N4校正。

中途报labelid=8的时候想当然觉得是int8，后来师兄告诉我id=8是float32的意思。

import os
import SimpleITK as itk
import numpy as np
from sklearn import preprocessing



def pre(files,img_type=itk.sitkFloat32):
    for i in files:
        img=itk.ReadImage(path+i,img_type)  #read images float32
        filename, fileexp = os.path.splitext(i)  # get filename

        mask = sitk.OtsuThreshold(img, 0, 1, 200)
        vol=itk.N4BiasFieldCorrection(img,mask)  #N4
        #vol=img
        #vol_a = itk.GetArrayFromImage(vol)  # 转换成numpy形式,
        #vol_a=np.float32(vol_a)
        out=itk.Normalize(vol)    #z标准化
        #average=np.average(vol_a)
        #sigma=np.std(vol_a)
        #vol_z=(vol_a-average)/sigma

        #out=itk.GetImageFromArray(vol_z)  #save as .nii

        '''
        resample = itk.ResampleImageFilter()
        resample.SetOutputDirection(out.GetDirection())
        resample.SetOutputOrigin(out.GetOrigin())
        resample.SetSize(out.GetSize)
        resample.SetOutputSpacing(out.Getspacing)
        resample.SetOutputPixelType(itk.sitkFloat32)  #save as float32
        out = resample.Execute(out)  #execute
        '''

        itk.WriteImage(out,'D:/test/'+filename+'.nii')  #save as nii




path = 'D:/Bratrain/'  # volume文件所在路径
files = os.listdir(path)  # 获取文件
pre(files)

再做z-scores归一化，最后保存为float32文件。对于label不进行处理。

这里报错了

Traceback (most recent call last):
  File "C:/Users/HP/PycharmProjects/pythonProject7/main.py", line 33, in <module>
    pre(files)
  File "C:/Users/HP/PycharmProjects/pythonProject7/main.py", line 15, in pre
    vol_z=preprocessing.scale(vol_a)    #z标准化
  File "C:\Users\HP\PycharmProjects\pythonProject4\venv\lib\site-packages\sklearn\utils\validation.py", line 63, in inner_f
    return f(*args, **kwargs)
  File "C:\Users\HP\PycharmProjects\pythonProject4\venv\lib\site-packages\sklearn\preprocessing\_data.py", line 161, in scale
    X = check_array(X, accept_sparse='csc', copy=copy, ensure_2d=False,
  File "C:\Users\HP\PycharmProjects\pythonProject4\venv\lib\site-packages\sklearn\utils\validation.py", line 63, in inner_f
    return f(*args, **kwargs)
  File "C:\Users\HP\PycharmProjects\pythonProject4\venv\lib\site-packages\sklearn\utils\validation.py", line 716, in check_array
    raise ValueError("Found array with dim %d. %s expected <= 2."
ValueError: Found array with dim 3. the scale function expected <= 2.

明天再仔细看看报错原因

6/12 今天重新一块块跑了一遍，发现居然跑通了，生成一个nii大概需要跑十分钟左右，师兄告诉我是因为N4耗费太多时间了。

这里需要记录一下标准化用的代码，应该是Z-SCORES标准化，sitk库里就可以找得到，输入图像输出图像。这个函数我完全没有了解，网上也找不到，官方sitk手册也读不懂，唉，瞎弄了一下，幸亏是输入image，不过这下子就不用转成numpy形式了，真好。

import os
import SimpleITK as itk
import numpy as np
from sklearn import preprocessing


def pre(files,img_type=itk.sitkFloat32):
    for i in files:
        img=itk.ReadImage(path+i,img_type)  #read images float32
        filename, fileexp = os.path.splitext(i)  # get filename

        mask = itk.OtsuThreshold(img, 0, 1, 200)
        vol = itk.N4BiasFieldCorrection(img, mask)  # N4


        out = itk.Normalize(vol)  # z标准化

        resample = itk.ResampleImageFilter()
        resample.SetOutputDirection(out.GetDirection())
        resample.SetOutputOrigin(out.GetOrigin())
        resample.SetSize(out.GetSize())
        resample.SetOutputSpacing(out.GetSpacing())
        resample.SetOutputPixelType(itk.sitkFloat32)  # save as float32
        out = resample.Execute(out)  # execute


        itk.WriteImage(out,'D:/test2/'+filename+'.nii')  #save as nii



path = 'D:/Bratrain/'  # volume文件所在路径
files = os.listdir(path)  # 获取文件
pre(files)

6/12后续是对test数据同样处理，不过之后应该也要裁一下

6/17今天做完lits的crop之后做心脏数据集的crop，照例对z轴crop完之后发现它的轴是反的，应该是y轴对应z才对，于是修改了一下代码

import os
import SimpleITK as itk
import numpy as np


def crop(segfiles,volfiles):
    for i,j in zip(segfiles,volfiles):
        seg = itk.ReadImage(path2 + i)  # read seg
        vol=itk.ReadImage(path1+j)     #read vol
        filename1, fileexp1 = os.path.splitext(i)  # get filename:filename1:seg
        filename2,fileexp2=os.path.splitext(j)  #filename2:vol
        seg_size = seg.GetSize()  # get seg size
        vol_size=vol.GetSize()   #get vol size
        # print(seg_size)
        # print(vol_size)


        seg_a=itk.GetArrayFromImage(seg)
        mask_index = np.argwhere(seg_a!= 0)  # return mask!=0 whole coords
        (d1,h1,w1),(d2,h2,w2)=np.max(mask_index,0),np.min(mask_index,0)  #zyx
        h_max=h1 +3   #side
        h_min=h2 -3

        newvol=vol[:,h_min:h_max,:]  #new vol,对于3D图像可以这样直接CROP xyz
        newseg=seg[:,h_min:h_max,:]  #new seg

        itk.WriteImage(newvol,'H:/MedicinePublicData/Task02_Heart/imgtrain_cropy/'+filename2+'.nii')  #save vol
        itk.WriteImage(newseg, 'H:/MedicinePublicData/Task02_Heart/label_cropy/' + filename1 + '.nii')  #save seg


# def crop(volfiles,segfiles):
#     for i,j in (volfiles,segfiles):
#         vol=itk.ReadImage(path1+i) #read vol
#
#         filename1,fileexp1=os.path.split(i)  #get filename
#         filename2,fileexp2=os.path.split(j)
#         vol_size=vol.GetSize()  #get vol size
#         seg_size=seg.GetSize()  #get seg size
#
#         print(vol_size)
#         print(seg_size)
#
#         # mask_index=np.argwhere(seg!=0)  #return mask!=0 whole coords
#         # (d1,h1,w1),(d2,h2,w2)=np.max(mask_index,0),np.min(mask_index,0)

path1='H:/MedicinePublicData/Task02_Heart/imgtrain_pro/'   #get vol
path2='H:/MedicinePublicData/Task02_Heart/label_pro/'  #get segmentation
volfiles=os.listdir(path1)
segfiles=os.listdir(path2)
# crop(volfiles,segfiles)
crop(segfiles,volfiles)

到这里大概是预处理完毕了