Closed uzaymacar closed 2 years ago
The linked PR currently compares our lesion segmentation model to sct_deepseg_lesion
. However, due to the vast differences between the mp2rage contrast and the t2 contrast (which the sct_deepseg_lesion
model is trained on), we don't get a fair comparison. Therefore, we can focus our efforts on visually comparing the lesion segmentation predictions to the two raters' annotations as shown in the linked PR.
Similar to #28, the goal is to have a benchmark on lesion segmenation wherein we compare our newly trained
ivadomed
lesion segmentation model to alternatives.