We are reaching out from the Physiology Department at the University of Córdoba (Spain) to seek guidance or suggestions on processing head PET/CT studies using 3D Slicer.
Attached below are some images showing the output when opening a folder with 3D Slicer.
After reviewing the available information, it appears that performing a Brain Parcellation of different brain areas would be an ideal approach. This would allow for the independent quantification of the PET signal within these regions.
Once these data are obtained, team members will analyze the data matrices to identify specific changes across different brain areas throughout the study. These changes will also be correlated with biochemical data and/or the final classification of subjects (patients with mild cognitive impairment) after the study.
Regarding CT scans: SynthSR does a decent job with CT ! The only caveat is that the dynamic range of CT is very different to that of MRI, so they need to be clipped to [0, 80] Hounsfield units. You can use the --ct flag to do this, as long as your image volume is in Hounsfield units. If not, you will have to clip to the Hounsfield equivalent yourself (and not use --ct).
Thank you very much, Fernando, for your response. Please ignore the email I just sent you :-). Although I was trying to do something with 3D Slicer, since I feel more comfortable having worked with it before.
We have several examples of segmentation models that have been integrated with Slicer but not SynthSeg. In theory it should be possible to follow existing examples and plug it in, but someone with the right experience would need to go through the process. It would be great if someone has the bandwidth and wants to take this on (an AI coding buddy might be able to help).
Thank you all very much for the responses. Following some of @lassoan suggestions, we are using the MONAIAuto3DSeg extension and the AI model: “Brain and intracranial hemorrhage (ICrH)” for segment the main areas of the brain (images attached). However, it does not segment as we expected (would like); the main lobes are what we would like.
The Brain and intracranial hemorrhage (ICrH) model was trained for hemorrhage analysis. We used corrected SynthSeg labels to train this model.
Would you be interested in collaborating to train a brain parcellation model using Auto3DSeg on CT images? We could also use SynthSeg labels as the GT.
Great idea @diazandr3s - training Auto3DSeg using SynthSeg ground truth would be very cool. The resulting model would be more maintainable and easy to just drop into Slicer.
Thank you very much, Andres, for your message, but I believe my background is far from training models for 3D Slicer. We are happy to collaborate with anyone who can help us quantify PET specifically by large brain regions.