I am trying to do brain tumor segmentation. I need to load my data. How to make customized dataset of a brain images ( .nii)?

vision
1

I am trying to build the customized dataset for brain image. I have two folders

  1. HGG
  2. LGG
    In each folder we have 5 MRI images including Flair, t1, t1c, t2 and a labeled image.
    How to create custom dataset. Your help in this situation is highly appreciable. Thank you
2

This tutorial might be hellful.
In case you get stuck, could you add more information about your data and how each sample would be loaded?

3

You can create your own dataset class by subclassing a dataset like so (minimal example):

class Lazy_Dataset(torch.utils.data.Dataset):

    def __init__(self, path):
        self.steps = [function that figures out the size of your dataset located in the path]
        self.path = path

    def __len__(self):
        return self.steps
    
    def __getitem__(self, idx):
        array = [function that loads your nifti file as a numpy array]
        data = torch.Tensor(array)
        label = [function that figures out your label]
        return data, label, idx

You can later create an instance of that dataset (and a dataloader) by calling:

mydataset = Lazy_Dataset(mypath)
mydataloader = torch.utils.data.DataLoader(mydataset, batch_size=24, shuffle=True)
4

Thank you for your kind response. Here What i did so far. Can you please highlight the mistake.
Here is my code and folder organization.
(upload://pr9pSRiQKjIGy9X03dr2BAlfY4F.png) ![Presentation4|690x388]

patient_dir = “D:\Brats2020\”
l_patnt =os.listdir(patient_dir)
#print(l_patnt)

class Brats2022(Dataset):
def init(self,patient_dir):
self.patient_dir = patient_dir

def __len__(self):
    return len(self.patient_dir)

def __getitem__(self, index):
    #return self.patient_dir[index]
    patient_dir = self.patient_dir[index]
    volumes = []
    patient_id = os.path.split(patient_dir)[-1]
    volume_path = os.path.join(patient_dir,patient_id + "_" + "seg"+".nii.gz")
    volume_seg = nib.load(volume_path)
    volu_im = volume_seg.get_fdata()
    np_arr = np.array(volu_im)
    data_transform = transforms.ToTensor()
    return data_transform(np_arr)