Preprocess Volumes for Deep Learning

Read Volumetric Data

Supported file formats for volumetric image data include MAT-files, Digital Imaging and Communications in Medicine (DICOM) files, and Neuroimaging Informatics Technology Initiative (NIfTI) files.

Read volumetric image data into an ImageDatastore. Read volumetric pixel label data into a PixelLabelDatastore (Computer Vision Toolbox). For more information, see Datastores for Deep Learning.

The table shows typical usages of imageDatastore and pixelLabelDatastore for each of the supported file formats. When you create the datastore, specify the 'FileExtensions' argument as the file extensions of your data. Specify the ReadFcn property as a function handle that reads data of the file format. The filepath argument specifies the path to the files or folder containing image data. For pixel label images, the additional classNames and pixelLabelID arguments specify the mapping of voxel label values to class names.

volds = imageDatastore(filepath, ...
   'FileExtensions','.mat','ReadFcn',@(x) fcn(x));

pxds = pixelLabelDatastore(filepath,classNames,pixelLabelID, ...
    'FileExtensions','.mat','ReadFcn',@(x) fcn(x));

function data = matRead(filename)
    inp = load(filename);
    f = fields(inp);
    data = inp.(f{1});
end

volds = imageDatastore(filepath, ...
   'FileExtensions','.dcm','ReadFcn',@(x) dicomread(x));

pxds = pixelLabelDatastore(filepath,classNames,pixelLabelID, ...
   'FileExtensions','.dcm','ReadFcn',@(x) dicomread(x));

volds = imageDatastore(filepath, ...
   'FileExtensions','.nii','ReadFcn',@(x) niftiread(x));

pxds = pixelLabelDatastore(filepath,classNames,pixelLabelID, ...
   'FileExtensions','.nii','ReadFcn',@(x) niftiread(x));

Associate Image and Label Data

To associate volumetric image and label data for semantic segmentation, or two volumetric image datastores for regression, use a randomPatchExtractionDatastore (Image Processing Toolbox). A random patch extraction datastore extracts corresponding randomly-positioned patches from two datastores. Patching is a common technique to prevent running out of memory when training with arbitrarily large volumes. Specify a patch size that matches the input size of the network and, for memory efficiency, is smaller than the full size of the volume, such as 64-by-64-by-64 voxels.

You can also use the combine function to associate two datastores. However, associating two datastores using a randomPatchExtractionDatastore has several benefits over combine.

Preprocess Volumetric Data

Deep learning frequently requires the data to be preprocessed and augmented. For example, you may want to normalize image intensities, enhance image contrast, or add randomized affine transformations to prevent overfitting.

To preprocess volumetric data, use the transform function. transform creates an altered form of a datastore, called an underlying datastore, by transforming the data read by the underlying datastore according to the set of operations you define in a custom function. Image Processing Toolbox™ provides several functions that accept volumetric input. For a full list of functions, see 3-D Volumetric Image Processing (Image Processing Toolbox). You can also preprocess volumetric images using functions in MATLAB^® that work on multidimensional arrays.

The custom transformation function must accept data in the format returned by the read function of the underlying datastore.

RandomPatchExtractionDatastore does not support the DataAugmentation property for volumetric data. To apply random affine transformations to volumetric data, you must use transform.

The transform function must return data that matches the input size of the network. The transform function does not support one-to-many observation mappings.

dsTrain = transform(volds,@(x) preprocessVolumetricIMDS(x,inputSize));

function dataOut = preprocessVolumetricIMDS(data,inputSize)
 
numRows = size(data,1);
dataOut = cell(numRows,1);
 
for idx = 1:numRows
    
    % Perform randomized 90 degree rotation about the z-axis
    data = imrotate3(data{idx,1},90*(randi(4)-1),[0 0 1]);

    % Resize the volume to the size expected by the network
    dataClean = imresize(data,inputSize);
    
    % Add zero-mean Gaussian noise with a normalized variance of 0.01
    dataNoisy = imnoise(dataClean,'gaussian',0.01);

    % Return the preprocessed data
    dataOut(idx) = dataNoisy;
    
end
end

dsTrain = transform(volds,@preprocessVolumetricPatchDS);

function dataOut = preprocessVolumetricPatchDS(data)

img = data(1);
resp = data(2);

% 5 augmentations: nil,rot90,fliplr,flipud,rot90(fliplr)
augType = {@(x) x,@rot90,@fliplr,@flipud,@(x) rot90(fliplr(x))};

rndIdx = randi(5,1);
imgOut = augType{rndIdx}(img);
respOut = augType{rndIdx}(resp);

% Return the preprocessed data
dataOut = table(imgOut,respOut};

end