Loader

The Loader class is used to Load data for various types of tasks such as classification, image generation and image segmentation.
The Loader class has Classification DataLoader for classification tasks and Segmentation DataLoader for image-to-image (image generation) and image-to-labels (segmentation) tasks.

Classification DataLoader

For Classification DataLoader the class label can be provided to the model in the form of a csv file, which contains a filename label and classname label.
The input file is written as input_image.nii.gz
The output file is written as output_labels.csv

Parameters:

• root (str,required): This is path to the directory where the data is stored
• csv_dir (str,required): This is the csv file which contains the subject and its corresponding label
• transform (torchvision.transforms,required) : The transforms here is the transforms from torchvision library
• target_transform (torchvision.transforms,required): The target tranforms is the same as transforms parameter but the transformation is done on the test data
• loader (function,required): The type of loader to be used, current loader uses nipy.
• is_valid_file (function,required): List of type of files to be considered to be loaded as an input data.
• filename_label (str,required): The filename label in the CSV file.
• class_label (str,required): The class label in the CSV file.
• file_type (tuple,default = ('nii.gz','.nii')): The file types to be considered as an input for the model. The resulting files are stacked along channels in the input tensor.
• common (int,default = 64): The common is the size of the 3D Tensor.
• demographic (list,default = []): The list demographic factors to be used for classification.

Usage:

from niftytorch.loader.dataloader import ImageFolder
from torchvision import transforms
data_transforms = transforms.Compose([transforms.ToTensor()])
data_folder = "../data/train/"
data_csv = "../data.csv"
filename_label = "Subject"
class_label = "labels"
image_scale = 32
file_type = ('t1w.nii.gz','flair.nii.gz')
demographic = ['factor1','factor2']
image_datasets = ImageFolder('classification',root = data,data_csv = data_csv,transforms = data_transforms,target_transforms = data_transforms,loader = ,filename_label = filename_label,class_label = class_label,common = image_scale,file_type = file_type,demographic = demographic)

Segmentation DataLoader

The Segmentation DataLoader is used for image segmentation and image generation.
The input file is written as input_image.nii.gz
The output file is written as seg.nii.gz

Parameters:

• root (str,required): This is path to the directory where the data is stored.
• transform (torchvision.transforms,required) : The transforms here is the transforms from torchvision library
• target_transform (torchvision.transforms,required): The target tranforms is the same as transforms parameter but the transformation is done on the test data
• loader (function,required): The type of loader to be used, current loader uses nipy.
• is_valid_file (function,required): List of type of files to be considered to be loaded as an input data.
• common (int,default = 64): The common is the size of the 3D Tensor.

Usage:

from niftytorch.Loader.dataloader import ImageFolder
from torchvision import transforms
data_transforms = transforms.Compose([transforms.ToTensor()])
data_folder = "../data/train/"
image_scale = 32
image_datasets = ImageFolder('segmentation',root = data,data_csv = data_csv,transforms = data_transforms,target_transforms = data_transforms,loader = ,filename_label = filename_label,class_label = class_label,common = image_scale)

Paired DataLoader

For Paired DataLoader the class label can be provided to the model in the form of a csv file, which contains a filename label and classname label.
The input file is written as input_image.nii.gz
The output file is written as output_labels.csv

Parameters:

• root (str,required): This is path to the directory where the data is stored.
• transform (torchvision.transforms,required) : The transforms here is the transforms from torchvision library
• target_transform (torchvision.transforms,required): The target tranforms is the same as transforms parameter but the transformation is done on the test data
• loader (function,required): The type of loader to be used, current loader uses nipy.
• is_valid_file (function,required): List of type of files to be considered to be loaded as an input data.
• common (int,default = 64): The common is the size of the 3D Tensor.
• negative_examples (int, default = None): The number of negative example to be considered for each positive examples

Usage:

from niftytorch.Loader.dataloader import ImageFolder
from torchvision import transforms
data_transforms = transforms.Compose([transforms.ToTensor()])
data_folder = "../data/train/"
image_scale = 32
image_datasets = ImageFolder('paired',root = data,data_csv = data_csv,transforms = data_transforms,target_transforms = data_transforms,loader = ,filename_label = filename_label,class_label = class_label,common = image_scale,negative_examples = 4)