While I’m now attempting to follow the workflow described in this paper, I’m still stymied by the poor fitness of my baseline dataset (a SPECT of my spine) for segmentation of soft tissue (or it may be a limitation of the TotalSegmentator module. What suggestions to any of you have for filling in the missing intervertebral discs? Am I just stuck working to manually segment those structures slice by slice?
Example of both the dataset and segments attached.
There is no soft tissue visible in the image you shared. it is all noise.
What is the scalar range of the volume? If you try optimizing window/level values can you achieve a less noisy visualization? The segmentation seems quite good to me btw.
Technically the volume ranges from -3024 to 2324, but checking the histogram the practical range is approximately from -1035 to 2110. I’ll keep playing with the window/level values, but so far I’m not having luck improving the contrast between the IVDs and the adjacent soft tissue.
