TSXR-V2-R001

This dataset comprises models that extend the modality support of the TotalSegmentator2D tool (TS2D) to X-ray images. 

The model checkpoints have been trained on the v2.0.1 TotalSegmentator dataset using DiffDRR to synthesize X-ray images and constitute the initial release (R001). The checkpoints were generated using nnU-Net, which is also used within TS2D for inference.

Each model was trained on a specific subset of labels from the DiffDRR generated synthetic TotalSegmentator dataset.
This release includes:

• Cardiac: 18 cardiovascular structures, including heart and major arteries.
• Muscles: 23 musculoskeletal structures, including the brain and spinal-cord, and also multiple skeletal structures.
• Organs: 24 organ structures, including lungs, liver and stomach.
• Ribs: 26 ribs, including sternum and costal-cartilages.
• Vertebrae: 26 vertebrae, including the sacrum.

Synthesized data was created at three attenuation levels (1, 3 and 6) to allow for a more robust training of each respective model.