TotalSpineSeg-r20241115

TotalSpineSeg-r20241115

TotalSpineSeg-r20241005

TotalSpineSeg-r20241005

Sacrum segmentation

This release includes the nnunet model trained to segment the sacrums on T1w and T2w images.

train-with-vert-landmarks-variants

This release includes models trained with vertebrae landmarks (C2, T1, T12, Sacrum) and disc landmarks.

Here's a description of each configuration in the release:

• nnUNetTrainer__nnUNetPlans: Default nnUNet trainer configuration.
• nnUNetTrainer__nnUNetPlans_small: Default nnUNet trainer configuration with small patch size (128x96x96).
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetPlans: nnUNet trainer with no mirroring data augmentation and nearest-neighbor segmentation interpolation (Instead of the default linear).
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetPlans_small: Same as above but with small patch size (128x96x96).
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetResEncUNetLPlans: nnUNet trainer with no mirroring data augmentation and nearest-neighbor segmentation interpolation (Instead of the default linear), using residual encoder plans.
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetResEncUNetLPlans_small: Same as above but with small patch size (128x96x96).

train-with-disc-only-landmarks-variants

This release includes models trained with only discs landmarks (C2C3, C7T1, T12L1, L5S).

Here's a description of each configuration in the release:

• nnUNetTrainer__nnUNetPlans: Default nnUNet trainer configuration.
• nnUNetTrainer__nnUNetPlans_small: Default nnUNet trainer configuration with small patch size (128x96x96).
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetPlans: nnUNet trainer with no mirroring data augmentation and nearest-neighbor segmentation interpolation (Instead of the default linear).
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetPlans_small: Same as above but with small patch size (128x96x96).
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetResEncUNetLPlans: nnUNet trainer with no mirroring data augmentation and nearest-neighbor segmentation interpolation (Instead of the default linear), using residual encoder plans.
• nnUNetTrainer_DASegOrd0_NoMirroring__nnUNetResEncUNetLPlans_small: Same as above but with small patch size (128x96x96).

TotalSpineSeg-r20240826

TotalSpineSeg-r20240826

TotalSpineSeg-r20240628

for release r20240628

TotalSpineSeg-labels

Temporary release to store labels (untill all labels will be pushed into the repositories).

TotalSpineSeg-r20240601

Training done on this commit: e35a764

TotalSpineSeg-r20240220

Initial model weights