Skip to content

Latest commit

 

History

History
60 lines (43 loc) · 2.94 KB

README.md

File metadata and controls

60 lines (43 loc) · 2.94 KB
Raw

EIT Image Reconstruction Algorithm

This is a submission for the Kuopio Tomography Challenge.

Authors

  • Amal Mohammed A Alghamdi (DTU), Denmark
  • Martin Sæbye Carøe (DTU), Denmark
  • Jasper Marijn Everink (DTU), Denmark
  • Jakob Sauer Jørgensen (DTU), Denmark
  • Kim Knudsen (DTU), Denmark
  • Jakob Tore Kammeyer Nielsen (DTU), Denmark
  • Aksel Kaastrup Rasmussen (DTU), Denmark
  • Rasmus Kleist Hørlyck Sørensen (DTU), Denmark
  • Chao Zhang (DTU), Denmark

Addresses

DTU: Technical University of Denmark, Department of Applied Mathematics and Computer Science Richard Petersens Plads Building 324 2800 Kgs. Lyngby Denmark

Description of the algorithm

We built a complete electrode model CEM by extending the model provided in the linked BSc. thesis. We also built the framework to solve a nonlinear optimization problem of inferring the conductivity $\sigma$ using a least-square optimization approach. We added two regularization terms. One is TV, based on the linked tutorial, and the other one is Tikhonov, based on the regularization class SMPrior provided by the competition. We used scipy L-BFGS-B to solve the optimization problem and segment with Chan-Vese segmentation method from scikit-image. Our implementation uses FEniCS, a finite element method library.

Installation instructions

To run our EIT image reconstruction algorithm, you will need:

  • Python 3.x
  • Required Python libraries (listed in requirements.txt)
  • Access to the provided dataset (not included in this repository)

Note that you will need to install FEniCS 2019.1.0. One way to do it is to follow Anaconda installation instructions in this link.

Usage instructions

python main.py path/to/input/files path/to/output/files difficulty

Examples

Phantom Ref Level 1 Level 4 Level 7
a
b
c
d

Scores for each phantom and difficulty 1,4 and 7:

Phantom Level 1 Level 4 Level 7
a 0.730 0.587 0.246
b 0.591 0.331 0.243
c 0.682 0.671 0.150
d 0.622 0.676 0.593

Scores have been computed using our own implementation of the scoring function based on scikit learn.

License

All files in the repository come with the Apache-v2.0 license unless differently specified.