Try cervical dorsal model on whole spine image

[valosekj]

When we tried the cervical model (r20240523) segmenting both dorsal and ventral rootlets on a whole spine image, the model predicted some false positive rootlets in the Th/L area; details in this comment. It would also be interesting to see prediction from the dorsal model, which is part of the SCT (sct_deepseg -task seg_spinal_rootlets_t2w).
@KaterinaKrejci231054, can you please run the model on the same subject to see the prediction? Thank you!

[KaterinaKrejci231054]

When we tried the cervical model (r20240523) segmenting both dorsal and ventral rootlets on a whole spine image, the model predicted some false positive rootlets in the Th/L area; details in this comment. It would also be interesting to see prediction from the dorsal model, which is part of the SCT (sct_deepseg -task seg_spinal_rootlets_t2w). @KaterinaKrejci231054, can you please run the model on the same subject to see the prediction? Thank you!

I've tried the model for dorsal rootlets segmentation (sct_deepseg -task seg_spinal_rootlets_t2w) on a composed image - there were no false positive rootlets (no rootlets segmented below C8 rootlets) - see the examples below:

- Example of dorsal only spinal rootlets segmentation:

• Result: without false positive rootlets:

- Example of ventral and dorsal rootlets segmentation:

• Result: with false positive rootlets:

[valosekj]

Thank you for running the model! It's great to see that sct_deepseg -task seg_spinal_rootlets_t2w does not segment any false positive rootlets below the C8 rootlets! @joshuacwnewton, do you know if sct_deepseg applies any post-processing after running the inference? For example, removing small objects, etc.?

[joshuacwnewton]

Yes! We've just added support for using the existing sct_deepseg postprocessing options (-remove-small, -largest) on monai/nnunetv2 models.

(Before, they only worked for ivadomed models)