virtualscanner.server.recon.drunck

Deep learning Reconstruction of Undersampled Cartesian K-space (DRUNCK) is a convolutional neural network that reconstructs 4x undersampled Cartesian acquisitions. Front end integration coming soon!

nifti2npy.py

load_dataset_from_nifti(nifti_path: pathlib.Path, img_size: int = 128) → numpy.ndarray

Make dataset by reading NIFTI files from nifti_path and resizing each image to img_size x img_size.

Parameters

  • nifti_path (str) – Path to folder containing NIFTI files.

  • img_size (int, optional) – Desired size of images in dataset. Images read from NIFTI files will be resized. Default value is 128.

Returns

dataset – ndarray of images converted from NIFTI files.

Return type

numpy.ndarray

main(nifti_path: pathlib.Path, img_size: int, low_freq_pc: float, save_path: pathlib.Path, reduction_factor: int, plot_flag: bool = True)

  1. Load NIFTI data as numpy.ndarray and resize each image to img_size x img_size.

  2. Undersample by skip_factor and add low_freq_pc low-frequency k-space values.

  3. Save files to disk.

Parameters

  • nifti_path (str) – Path to folder containing NIFTI files.

  • img_size (int) – Desired size of images in dataset. Images read from NIFTI files will be resized.

  • low_freq_pc (float) – Percentage of low-frequency fully-sampled k-space values to add to undersampled k-space.

  • save_path (str) – Path to save ndarray files.

  • reduction_factor (int) – Undersampling factor.

  • plot_flag (bool, optional) – Boolean flag to plot img_ind sample of dataset, undersampled dataset and corresponding k-space. Default value is True.

normalise_dataset(dataset: numpy.ndarray) → numpy.ndarray

Normalised dataset to [0, 1].

Parameters

dataset (numpy.ndarray) – ndarray of dataset samples.

Returns

normalised_dataset – ndarray of dataset normalised to [0, 1].

Return type

numpy.ndarray

plot(dataset: numpy.ndarray, ft_dataset: numpy.ndarray, dataset_undersampled: numpy.ndarray, ft_dataset_undersampled: numpy.ndarray, img_ind: int)

Plot fully-sampled and under-sampled data sample at img_ind position with corresponding k-space data.

Parameters

  • dataset (numpy.ndarray) – ndarray of dataset samples.

  • ft_dataset (numpy.ndarray) – ndarray of Fourier Transform of dataset.

  • dataset_undersampled (numpy.ndarray) – ndarray of undersampled dataset.

  • ft_dataset_undersampled (numpy.ndarray) – Fourier Transform of dataset_undersampled.

  • img_ind (int) – Index of data sample to plot.

save2disk(filename1: str, file1: numpy.ndarray, filename2: str, file2: numpy.ndarray, save_path: pathlib.Path)

Save file1 and file2 to disk as filename1 and filename2 at save_path.

Parameters

  • filename1 (str) – Filename of file1.

  • file1 (numpy.ndarray) – ndarray to be saved to disk.

  • filename2 (str) – Filename of file2.

  • file2 (numpy.ndarray) – ndarray to be saved to disk.

  • save_path (str) – Path to save ndarray files to.

undersample(dataset: numpy.ndarray, low_freq_pc: float, reduction_factor: int) -> (<class 'numpy.ndarray'>, <class 'numpy.ndarray'>, <class 'numpy.ndarray'>)

Undersample dataset as follows: 1. Obtain ft_dataset - Fourier Transform of dataset 2. Obtain ft_dataset_undersampled - undersample of ft_dataset 3. Add low_freq_pc low-frequency k-space components from ft_dataset to ft_dataset_undersampled. 4. Obtain dataset_undersampled - Inverse Fourier Transform of ft_dataset_undersampled.

Parameters

  • dataset (numpy.ndarray) – ndarray of dataset samples.

  • low_freq_pc (float) – Percentage of low-frequency fully-sampled k-space values to add to undersampled k-space.

  • reduction_factor (int) – Undersampling factor.

Returns

  • ft_dataset (numpy.ndarray) – ndarray of Fourier Transform of dataset.

  • dataset_undersampled (numpy.ndarray) – ndarray of undersampled dataset.

  • ft_dataset_undersampled (numpy.ndarray) – Fourier Transform of dataset_undersampled.

reconstruct.py

main(img_path: str, img_type: str) → tuple

Perform inference on pre-trained network, compute and display time to perform inference and plot results.

Parameters

  • model_path (keras.models.Model) – Path to load pre-trained Keras model.

  • test_data_path (str) – Path to folder containing input.npy and ground_truth.npy.

  • reduction_factor (int) – Undersampling factor of the dataset that the model was pre-trained on.

  • num_pred (int) – Number of test samples to perform inference on.

Returns

output_image – ndarray containing num_pred number of reconstructed samples.

Return type

numpy.ndarray

utest_reconstruct.py

class MyTestCase(methodName='runTest')

Bases: unittest.case.TestCase

test_drunck_recon()

Unit test DRUNCK reconstruction of a 256x256 undersampled image.