¶
Leaderboard - First round¶
Based on the median rank profile for SSIM scores, the top five submissions from each task were assessed by a panel of radiographers and radiologists. The median rank profile of these observer quality scores (QS) determines the final ranking (QS-rank) for each task. The definition of the quality scores can be found under Evaluation.
Test Results - Task 1:¶
| Place | Team | SSIM | SSIM-Rank | QS | QS-Rank |
|---|---|---|---|---|---|
| 1st | SubtleMedical | 0.1513 (Details) | 2 | 3.7625 (Details) | 3 |
| 2nd | UVABME | 0.2523 (Details) | 3 | 2.875 (Details) | 2 |
| 3rd | Di Fan Team | 0.0953 (Details) | 1 | 1.075 (Details) | 1 |
The columns SSIM and QS show average values across all test subjects. A file containing individual values is linked.
Test Results - Task 2:¶
| Place | Team | SSIM | SSIM-Rank | QS | QS-Rank |
|---|---|---|---|---|---|
| 1st | SubtleMedical | 0.3050 (Details) | 2 | 3.2375 (Details) | 3 |
| 2nd | UVABME | 0.3666 (Details) | 3 | 2.975 (Details) | 2 |
| 3rd | Di Fan Team | 0.2805 (Details) | 1 | 2.55 (Details) | 1 |
Participating Teams:¶
| Team Name | Participants | Description | Details |
| Di_Fan_Team | Di Fan | Deep de-aliasaing generative adversarial network (DAGAN), SSIM loss | MICCAI 2013 grand challenge and fastMRI dataset used for pre-training |
| SubtleMedical | Kevin Blansit, Zhehao Zhu, Ben Duffy, Keshav Datta |
Deep convolutional neural network (DCNN), combined SSIM and L1 loss | fastMRI dataset used for training |
| UVABME | Quan Dou, Xue Feng, Craig Meyer |
U-Net, combined SSIM and L1 loss |
fastMRI dataset used for training |