Team members: Joana Grilo
Affiliation: Mediaire Research
The second submission of joana based on the original but including the following modifications:
- Only positive voxels closer than 4 voxels to the center of the aneurysm were considered for training. The remaining positive voxels were excluded from training. This was performed to reduce FP located in the edge of bigger vessels.
- Weight decay of 1e-4 was added.
- All data was used for training (in the first submission 13 were left out).
- Final threshold of 0.75 to binarize probabilities.