CIL-HTC2022-Algo3

Authors:

• Gemma Fardell (STFC), United Kingdom
• Jakob Sauer Jørgensen (DTU), Denmark
• Laura Murgatroyd (STFC), United Kingdom
• Evangelos Papoutsellis (STFC, Finden), United Kingdom
• Edoardo Pasca (STFC), United Kingdom

Addresses

STFC: United Kingdom Research and Innovation Scientific Computing Department of Science Technology Facilities Council Rutherford Appleton Laboratory Harwell Campus Didcot OX11 0QX

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

Finden: Building R71, Rutherford Appleton Laboratory, Harwell, Oxford, OX11 0QX

Description of the algorithm

This is an entry for the HTC2022 competition. The algorithm in algo.py is developed using CIL, a toolkit for tomographic imaging and optimisation. The main steps of the algorithm is:

1. Pre-processing: renormalisation, single material beam hardening correction, zero padding
2. Generation of pixelwise lower and upper bound circular masks (mask is fitted to the provided data)
3. Regularised iterative reconstruction algorithm using tools from CIL: L1Norm data fidelity with TV and L1 regularisation
4. Post-processing: segmentation of the reconstruction with multi-Otsu threshold, followed by a morphological closing and opening on the reconstruction

Installation instructions

Installation instructions, including any requirements.

conda env create --file environment.yml

Examples

Examples of reconstructing the example datasets, where we have limited the angles to 90, 60 and 30 degree ranges. The 'ref' column is the given segmented result from the full dataset.

	Ref	90	60	30
ta
tb
tc
td

Scores for each sample and angle:

	90	60	30
ta	0.988	0.977	0.899
tb	0.969	0.941	0.816
tc	0.946	0.880	0.798
td	0.958	0.952	0.924

Repository content

• utils.py
• algo.py
• main.py
• environment.yml
• README.md
• recalc_score.py
• test_data
  • htc2022_ta_full_recon_fbp_seg.png
  • htc2022_ta_sparse_example.mat
• results
  • htc2022_ta_sparse_example.png
  • AR30
    • htc2022_ta_full.png
    • htc2022_tb_full.png
    • htc2022_tc_full.png
    • htc2022_td_full.png
  • AR60
    • htc2022_ta_full.png
    • htc2022_tb_full.png
    • htc2022_tc_full.png
    • htc2022_td_full.png
  • AR90
    • htc2022_ta_full.png
    • htc2022_tb_full.png
    • htc2022_tc_full.png
    • htc2022_td_full.png

License

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