I finally have sorted all my raw imaging and labels (only took 3 months of coding!). Since then, I’ve been running some feature extractions on approximately 200 image sets.
However after getting the basic’s running I thought it’d be best to read the literature to get a better idea of how best to approach feature extraction and selection.
One of the things I’ve already come across is how image sets might have different grey levels ranging typically from 32 to 256. Given my imaging set is spread across 5 years I suspect different image sets have different grey levels. I can’t see evidence that pyradiomics resamples imaging to same grey level.
What’s the best approach to resampling grey levels ?
Right now my data approach is
- Extract raw data from Eclipse (radiotherapy planning system) - images come out as DICOM
- Image resampling using plastimatch and sitk - images returned as nrrd
- Feed resampled images into pyradiomics
I suspect there is a sitk solution - but haven’t done a deep dive. Wanted to see what people did to normalize their image sets.