Clinically Applied Artificial Intelligence
Rigshospitalet, Copenhagen University

DeepAC

DeepAC is a deep learning network for obtaining attenuation correction umaps. The network currently works with input images Dixon-VIBE (VE11P) and UTE (VE11P)

DeepAC is a deep learning network for obtaining attenuation correction umaps. The network currently works with input images:
  • Dixon-VIBE (VE11P)
  • UTE (VE11P)
Future version will also include T1w-MPRAGE and VB20P versions of all three. If you are in need already now, please contact claes.noehr.ladefoged@regionh.dk

At some point, a more smooth sharing will be implemented, but for now, please just download the main python scripts and keras models.


Requirements

First, if you need to use GPU, you need to install the GPU, CUDA and cuDNN toolkits. See https://www.tensorflow.org/install/gpu.
Then, download the DeepAC files:
DeepDixon script (version August 20 2019)
DeepUTE script (version March_12_2019)
Models for DeepUTE and DeepDixon (1 GB)
Requirements file for installation

To install the required software tools, run:
pip3 install -r requirements.txt

Please note - to use DeepDixon, you further need to install FSL.

The best performance will be obtained with GPU support (about 4 seconds, depending on GPU type), but CPU can also be used. Please install tensorflow rather than tensorflow-gpu (in requirements). The running time is about 15-20 minutes for CPU.

Running the script

Change the root folder in the main script (on line 23) to where you put the main script.
Run: python3 process_DeepUTE.py ﹤path to DICOM data﹥

Optionally, you can supply different models or change the output directory. See within top of file for explanations.
The output will be a folder called DeepUTE within the DICOM data folder.

Contact

Claes Ladefoged, Rigshospitalet, Copenhagen, Denmark
claes.noehr.ladefoged@regionh.dk

Citation

The publication has been submitted for review - please contact claes.noehr.ladefoged@regionh.dk for details on citations in the mean time

Recent Publications