The DL-sparse-view CT Challenge has concluded. The Challenge provided an opportunity for investigators in deep-learning CT image reconstruction to compete with their colleagues on the accuracy of their methodology for solving the inverse problem associated with sparse-view CT acquisition.
The top five performing individuals and teams are:
| Username | Team Name | Members | Institutions(s) | RMSE Score |
|---|---|---|---|---|
| Max | Robust-and-stable | Martin Genzel Jan Macdonald Maximillian März |
Utrecht University Technical University of Berlin |
6.37 x 10^(-6) |
| TUM | YM & RH | Youssef Mansour Reinhard Heckel |
Technical University of Munich Rice University |
3.99 x 10^(-5) |
| cebel67 | DEEP_UL | Cédric Bélanger |
Université Laval | 1.29 x 10^(-4) |
| deepx | Yading Yuan | Icahn School of Medicine at Mount Sinai |
1.59 x 10^(-4) | |
| Haimiao | HBB | Haimiao Zhang Bin Dong Baodong Liu |
Beijing Information Science and Technology University Peking University Chinese Academy of Sciences |
1.81 x 10^(-4) |
For more information see the Challenge website which will remain open: dl-sparse-view-ct-challenge.eastus.cloudapp.azure.com/competitions/1
The validation phase, which included about 60 active participants, and the final test phase included 25 submissions. Each submission consisted of an algorithm report along with predictions on 100 images from sparse view data. Final score was the mean root-mean-square-error (RMSE) calculated in comparison with the truth images. RMSE values for the top five appear in the table. For full ranking for validation and test phases, please see the ‘Results’ tab of the Challenge website.
A full Challenge report will be forthcoming at the end of August.