How to load nib to pytorch

1

for medical images format Nifty ( nii.gz) how to load the image using data loader in pytorch and then plot them, my code blow try to do but seems there is something wrong with the numby shape since the image show as lines instead of normal medical images

bs = 2
num_epochs = 100
learning_rate = 1e-3
mom  = 0.9
import torch
import torch.nn as nn
import numpy as np
import pandas as pd
import torchvision
import torchvision.transforms as tfms
import torch.utils.data as data
import matplotlib.pyplot as plt
import torch.nn.functional as F
import scipy as sc
import os
import PIL
import PIL.Image as Image
import seaborn as sns
import warnings
import nibabel as nib#http://nipy.org/nibabel/gettingstarted.html
class Dataloder_img(data.Dataset):
    def __init__(self,root_dir,seg_dir,transforms ):
        self.root_dir = root_dir
        self.seg_dir = seg_dir
        self.transforms = transforms
        self.files = os.listdir(self.root_dir)
        self.lables = os.listdir(self.seg_dir)
        print(self.files)
    
    def __len__(self):
        return len(self.files)
    
    def __getitem__(self,idx):
        img_name = self.files[idx]
        label_name = self.lables[idx]
        img = nib.load(os.path.join(self.root_dir,img_name)) #!Image.open(os.path.join(self.root_dir,img_name))
        #change to numpy
        img = np.array(img.dataobj)
        #change to PIL 
        img = Image.fromarray(img.astype('uint8'), 'RGB')
        
        print(img.size)
        
        label = nib.load(os.path.join(self.seg_dir,label_name))#!Image.open(os.path.join(self.seg_dir,label_name))
        #change to numpy
        label = np.array(label.dataobj)
        #change to PIL 
        label = Image.fromarray(label.astype('uint8'), 'RGB')
        
        print(label.size)
        
        if self.transforms:
            img = self.transforms(img)
            label = self.transforms(label)
            return img,label
        else:
            return img, label
full_dataset = Dataloder_img(' image ',
                                     ' labels ',tfms.Compose([tfms.RandomRotation(180),tfms.ToTensor()
                                                            ]))#
                                   

train_size = int(0.8 * len(full_dataset))
val_size = len(full_dataset) - train_size
train_dataset, val_dataset = torch.utils.data.random_split(full_dataset, [train_size, val_size])
train_loader = data.DataLoader(train_dataset,shuffle=False,batch_size=bs)
val_loader = data.DataLoader(val_dataset,shuffle=False,batch_size=bs)

test_img, test_lb = next(iter(full_dataset))
print(test_img[0].shape)
plt.imshow(test_img[0])
plt.show()
2

Hi,
Your img is 4-Dimensional data. This can be checked using:

img.header.get_data_shape()

For my MRI data the resulting shape is (height, width, depth, sequence). Nibabel recommends accessing the data as an array using:

image_data = img.get_fdata()

This will still result in 4D data. If you want to process your images in 2D or 3D you can slice into this. Alternatively, you can used the 4D tensor from the dataset and use the dataloader to create a 5D tensor as discussed here pytorch.

Hope this helps.
James.

3 
data_dir='#image directory"
img=nib.load(os.path.join(data_dir,'ADNI_941_S_1311_MR_MPR__GradWarp__B1_Correction__N3__Scaled_Br_20080703170241434_S51039_I112538.nii'))                           #loading the image
img_data=img.get_data()                                                     #accessing image array
multi_slice_viewer(img_data)
plt.show()

it will load your nii file using nib.load
4

Let me know about your dataloading method from folder or directory which you used in your code.
after that it will clarify what you really want

5

Hello Aliktk
I am trying to load MRI NIFTI volume images from a folder, and pass it to dataloader for training and validation.

7

hello modupe, did you find a way to do that? Right now I have 295 .nii images and I am trying to get them by nibabel library and feed it to pretrained classification models. How can I do that?