v26_6_2 results on Brats2024-glioma
Baseline = nnUNet
Dice (↑) — baseline vs v26_6_2 (fold 0)
Note: baseline converged, but v26_6_2 didn't, running for another 300 epochs (total=1300).
| contrast |
model |
ET |
NCR |
RC |
SNFH |
| t1n (in-domain) |
baseline |
0.364 |
0.249 |
0.579 |
0.716 |
|
v26_6_2 |
0.345 |
0.226 |
0.594 |
0.706 |
| t1c |
baseline |
0.160 |
0.092 |
0.523 |
0.654 |
|
v26_6_2 |
0.361 |
0.266 |
0.552 |
0.655 |
| t2f |
baseline |
0.009 |
0.002 |
0.117 |
0.020 |
|
v26_6_2 |
0.219 |
0.184 |
0.502 |
0.626 |
| t2w |
baseline |
0.001 |
0.000 |
0.000 |
0.066 |
|
v26_6_2 |
0.238 |
0.176 |
0.514 |
0.644 |
HD95 mm (↓) — baseline vs v26_6_2 (fold 0)
| contrast |
model |
ET |
NCR |
RC |
SNFH |
| t1n |
baseline |
15.5 |
15.3 |
15.0 |
12.9 |
|
v26_6_2 |
14.7 |
16.0 |
12.2 |
10.6 |
| t1c |
baseline |
26.8 |
29.6 |
17.3 |
18.1 |
|
v26_6_2 |
21.6 |
20.2 |
13.7 |
13.7 |
| t2f |
baseline |
66.1 |
60.4 |
46.3 |
67.6 |
|
v26_6_2 |
22.1 |
17.7 |
13.8 |
20.0 |
| t2w |
baseline |
72.9 |
64.9 |
81.4 |
66.8 |
|
v26_6_2 |
22.0 |
16.4 |
11.9 |
14.0 |
In-domain it's close to a tie (but note that v26_6_2 actually hasn't covered yet): dice score is superior for baseline and HD95 is mostly superior for our method. Out of domain, our method is completely crushing the baseline.
v26_6_2 results on Brats2024-glioma
Baseline = nnUNet
Dice (↑) — baseline vs v26_6_2 (fold 0)
Note: baseline converged, but v26_6_2 didn't, running for another 300 epochs (total=1300).
HD95 mm (↓) — baseline vs v26_6_2 (fold 0)
In-domain it's close to a tie (but note that v26_6_2 actually hasn't covered yet): dice score is superior for baseline and HD95 is mostly superior for our method. Out of domain, our method is completely crushing the baseline.