Hi,
- Chord Length & Activity Modeling: When using an annular phantom, is it essential to explicitly account for the geometric path length (the intersection chord thickness through the annular active volume) for each LOR?
- Self-Attenuation: Does the calculation of normalization factors need to incorporate the self-attenuation of the annular phantom (attenuation of 511 keV photons through water and the acrylic walls)?
I would say it is required to take into account at least the intersection chord with the phantom, as your annular phantom still has some finite thickness. I would not expect those to be the reason of your high-activity at the edge of the reconstructed image though.
Edge Divergence: Is our naive 11+count formula the primary cause of the 1021 edge explosion (e.g., due to the lack of forward-projection weighting and sensitivity backprojection boundary issues)?
Your formula implies that any LORs without count is still sensitive. This is probably not the case for edge LORs but they are still taken into account in the generation of the sensitivity image. This could be why the edge voxels become very sensitive and cause the edge divergence.
Reference Workflow/Tools: Does CASToR have an official recommended pipeline, script, or tool (such as
castor-normor forward-projecting an analytical phantom image) to compute proper normalization factors from experimental annular phantom data? If so, are there any reference documentation or papers you would recommend for this geometry?
Zero-count LOR Handling: What is the recommended practice in CASToR for handling zero-count LORs in high-granularity systems to prevent numerical instabilities during reconstruction?
castor-norm is so far the main tool provided by CASToR to help compute normalization. It is also based on direct estimation and thus has the disadvantage of requiring high statistics as it is very sensitive to noise (more details here) in comparison with other approaches such as component based norm. According to your acquisition data (3 billion coincidences events for 1.6 billion LORs) the low-count LORs measurements will most likely be dominated by noise.
Specific questions regarding performance:
- Given our high-statistics dataset (2.5B prompts vs. 1.6B possible LORs), would converting and reconstructing in histogram mode yield a significant speedup compared to list-mode mode? Roughly what order of magnitude improvement can we expect?
- If real-time or dynamic list-mode reconstruction is not strictly required, is histogram-based reconstruction generally the recommended standard in CASToR for static high-count acquisitions?
I would expect an improvement in computing time but it’s complicated to estimate an order of magnitude given the sensitivity is computed differently in list-mode (beforehand) and histogram (recomputed for each subset, though lightweight as backprojection is just performed once) and could also vary if your system includes TOF or some level of compression (span/mash), but you don’t mention it so I assume this is not your case.
→ Essentially I would try to address the normalization issue in priority, modify the formula and check how noise affects your normalization by trying to reconstruct with normalization dataset of varying statistic, ideally generate a substantially higher-statistics normalization dataset. If you have some simulation data it could clarify what level of statistic would be required for your system with this experimental setup using a direct normalization approach.
Best,
Thibaut