Alzheimer's disease skull reconstruction

Repository for reconstruction of simulated skull CT data for AD project with files for reconstruction simulated skull CT data (KTH/SSF AD project)

Usage

Installation

To close the repository, run

$ git clone https://github.com/davlars/ad-skull-reconstruction
$ cd ad-skull-reconstruction

To install the package adutils, run (in the root folder of this package)

$ pip install -e .

Some dependencies are not currently available on pip and you'll get an error message if you do not have them installed.

To install the latest version of ODL, run

$ pip install git+git://github.com/odlgroup/odl

To install astra, run

$ conda install -c astra-toolbox astra-toolbox

Load data to disc

In order to use this repo, you first need to copy the data to your local machine. This can be done by:

$ python -c "import adutils; adutils.load_data_from_nas('Z:/')"

where 'Z:/' should be replaced with your local path to the "REFERENCE" data drive of the NAS (i.e 'Z:/' on windows, or '/mnt/imagingnas/Reference/' on linux). This takes quite some time (~10 minutes) to run at first, but makes subsequent reconstructions much faster. Note that it uses ~6GB of disk space in a sub-folder to this project.

Note that the above autaomtically loads only one of the simulated phantoms (70100644). If you want to load any of the othe ones, specify this with a phantom_number argument like:

$ python -c "import adutils; adutils.load_data_from_nas('Z:/',phantom_number='70162244')"

For linux users, instructions on how to properly mount the nas cand be found here

Files to read and reconstruct the data is given in this repository (the easiest example is given by a FBP reconstruction in FBP_reco_skullCT.py. Most of the data handling is however hidden in adutils.py. To make use of these, simply run the following in your script

import adutils

Rebin data

To avoid heavy computations, a suggestion is to use downsampled rebinned data. To rebin data, run the following after you've downloaded data to your local drive:

$ python -c "import adutils; adutils.rebin_data(10)"

with desired (rebin_factor) (default is set to 10). Once you've done this, simply load your data using the rebin data flag, as per below:

rebin_factor = 10

# Discretization
reco_space = adutils.get_discretization()

#Forward operator
A = adutils.get_ray_trafo(reco_space, use_rebin=True, rebin_factor=rebin_factor)

# Data
rhs = adutils.get_data(A, use_rebin=True, rebin_factor=rebin_factor)

Save image

To save data in a format that the clinical can review (typically nifti), use the adutils.save_image utility, with x being your reconstruction

fileName = /my/path/myFile
adutils.save_image(x, fileName, as_nii=True, as_npy=True)

Visualize data and compare to ground truth phantom

To visualize data, simply call built-in odl functionalities like my_reconstruction.show(), or alternatively use adutils.plot_data(reco) to get pre-defined cuts of clinical importance.

Also, using adutils.get_phantom the ground-truth phantom for the simulated data set can be retrieved. An example of such is given in the 2D-FBP, where the corresponding 2D-slice is loaded on the lines of:

# Compare to phantom
phantom = reco_space.element(adutils.get_phantom(use_2D=True))

phantom.show()

Alternatively, you can load the entire 3D dataset as a label map. The phantom is given with attenuation values expected for the spectrum used for the simualted dataset (120kVp). Note that if wanted the phantom can be loaded as labelled with tissue flags such that:

  0 - Air/Background
  1 - CSF/Soft tissue
  2 - Grey matter
  3 - White matter
  4 - Bone

for that, call

phantom = reco_space.element(adutils.get_phantom(use_2D=True, get_Flags=True))

phantom.show()

2D data set

To be able to work with a smaller dataset (to e.g. try out different reconstruction parameters), a 2D fanbeam dataset has been generated, consiting of a mid transversal slice of the skull. To use this dataset, simply use the flag use_2D = True in the appropriate calls to adutils. This means e.g.:

# Discretization
reco_space = adutils.get_discretization(use_2D=True)

#Forward operator
A = adutils.get_ray_trafo(reco_space, use_2D=True)

# Data
rhs = adutils.get_data(A, use_2D=True)

Exampes of such is given for FBP, CGLS, and TV.

Note: A 2D dataset only exists for phantom_number='70100644. If wanting to do quick tests on any of the other datasets, simply use use_rebin or use_subset.

Raw data

All simulated data is available on the lcrnas. For high dose (150 mGy) skullCT (following settings used at KI Geriatrics), data can be found in:

/lcrnas/data/Simulated/120kV

The simulated data consists of 23 spiral turns, and is divded in 23 separate files. The data is given as:

HelicalSkullCT_70100644Phantom_no_bed_Dose150mGy_Turn_{0,1,2,3,...,22}.data.npy

with corresponding geometry pickled in

HelicalSkullCT_70100644Phantom_no_bed_Turn_{0,1,2,3,...,22}.geometry.p

Note: Data with lower dosage is available upon request. All simulated data is given in /lcrnas/data/Simulated/120kV/raw/. DO NOT change the data in this repository. If unsure, ask.