lucaszamprogno/CT2US
0
ct2us
This tool is intended to automate the generation of simulated ultrasound image and label pairs from ct volumes (.nii/.nii.gz).
Purpose
Intended to be capable of supplementing datasets for ultrasound image labeling.
Expandability
Image generation process is very dependant on tissue attenuation, so specialized US renderers would be necessary/ideal to expand this tool to work on other body parts. For this purpose, much of the following code has hence been designed with modularity as a core goal, so that new methods can be added/replaced, as for example the segmentation quality or speed could have a significant impact on overall results.
Current use:
Further goals:
- code for two alternate optimized segmentation pipelines is still being developed
- one focusing on avoiding internal totalsegmentator steps being saved to memory
- another further optimizes by properly using gpu and cpu acceleration.
- Improved version of the totalsegmentator nnunet is still WIP. Once that is taken care of, pluging this in the pipeline with the stacked assemble should yield a significant speed up. [ ]