Data Preprocessing for MRI Datasets

A lot of effort in solving any machine learning problem goes into preparing the data. In this tutorial, we will see how to load and preprocess/augment data from a non-trivial dataset. The various image format classes give full or selective access to the header (meta) information and access to the image data is made available via NumPy arrays.

To run this tutorial, please make sure the following packages are installed:

scikit-image: For image io and transforms.

nibabel: Read / write access to some common neuroimaging file formats

matplotlib: which allows mathematical representations of all kinds to be created and visualized.

Parameter:

Path: A path-like object representing a file system path.

Return Type: This method returns a string value that represents the directory name from the specified path.

Download the sample nibabel files from my github which is in nii.giz form

Import Libraries:

import os
import nibabel as nib
import numpy as np
import matplotlib.pyplot as plt
}

Directory for MRI data
Importing Datasets: dirName = r"file_path"

function to read data from dirName

def FileRead(file_path):
    nii = nib.load(file_path)
    data = nii.get_data()
    #if (np.shape(np.shape(data))[0] == 3): #If channel data not present
     #   data = np.expand_dims(data, 3)
    return data

function returns a 2 dimensional numpy array from a 3 dimensional data

def Nifti3Dto2D(Nifti3D):
    Nifti3DWOChannel = Nifti3D#[:,:,:,0] #Considering there is only one channel info
    Nifti2D = Nifti3DWOChannel.reshape(np.shape(Nifti3DWOChannel)[0], np.shape(Nifti3DWOChannel)[1] * np.shape(Nifti3DWOChannel)[2])
    return Nifti2D

function returns a 1 dimensional numpy array from a 2 dimensional array

def Nifti2Dto1D(Nifti2D):
    Nifti1D = Nifti2D.reshape(np.shape(Nifti2D)[0] * np.shape(Nifti2D)[1])
    return Nifti1D

reshapes a 1 dimensional array to 2 dimensional data

def Nifti1Dto2D(Nifti1D, height):
    Nifti2D = Nifti1D.reshape(height,int(np.shape(Nifti1D)[0]/height))
    return Nifti2D

reshapes 2 dimensional array to 3 dimensional data

def Nifti2Dto3D(Nifti2D):
    Nifti3DWOChannel = Nifti2D.reshape(np.shape(Nifti2D)[0],np.shape(Nifti2D)[0],np.shape(Nifti2D)[1]//np.shape(Nifti2D)[0])
    #Nifti3D = np.expand_dims(Nifti3DWOChannel, 3)
    return Nifti3DWOChannel

normalize data between o-1

def normalize(x):
    min_val = np.min(x)
    max_val = np.max(x)
    x = (x-min_val) / (max_val-min_val)
    return x

To copy data to local storage

def FileSave(data, file_path):
    nii = nib.Nifti1Image(data, np.eye(4))
    nib.save(nii, file_path)

returns list of files from the given dirName

def getListOfFiles(dirName):
    # creating a list of file and sub directories 
    listOfFile = os.listdir(dirName)
    allFiles = []
# Iterate over entries
    for entry in listOfFile:
        # Creating full path
        fullPath = os.path.join(dirName, entry)
        if os.path.isdir(fullPath):
            allFiles = allFiles + getListOfFiles(fullPath)
        else:
            allFiles.append(fullPath)
    return allFiles

Takes list of files and creates 1 dimension and 2 dimension data from MR data

def main():
    listOfFiles = getListOfFiles(dirName)
    twod_data = []# to save two dimensional data
    oned_data = []
    listOfFiles = []
    for (dirpath, dirnames, filenames) in os.walk(dirName):
        listOfFiles += [os.path.join(dirpath, file) for file in filenames]
       # listOfFiles_under.append(listOfFiles_under)
        #print(listOfFiles_under)
    for element in listOfFiles:
        print(element)
        red_files = FileRead(element)
        twod_array = Nifti3Dto2D(red_files)
        twod_data.append(twod_array)
        #print(twod_unsam)
        oned_array= Nifti2Dto1D(twod_array)
        oned_data.append(oned_array)
    oned_data = np.asarray(oned_data, dtype=np.float64, order='C')
    twod_data = np.asarray(twod_data, dtype=np.float64, order='C')
    return oned_data, twod_data

oned_data, twod_data = main()

for visualization

data_vis = Nifti2Dto3D(Nifti1Dto2D(oned_data[0,:], 256))
#plt.imshow(data_vis[20,:,:], cmap='gray')
plt.imshow(data_vis[:,:,20], cmap='gray')

Output for one slice image

Comments

[nouha-mejri]

good job , but what does this line refers to data_vis[20,:,:] , thanx