.png){kind=icon}
.png){kind=icon}
Return
ComptonNet: A direct reconstruction model for Compton camera
.png){kind=icon}
PRE
AI
DOI：10.1063/5.0213950.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A Compton camera is a gamma-ray imaging device, especially in the sub-mega-electron volt to higher than mega-electron volt range. Compton cameras have recently attracted attention as an environmental survey tool. However, owing to their limited sensitivity, Compton camera images often suffer from various artifacts, especially when the event statistics are low. To address this challenge, several deep learning models have been proposed to enhance the quality of reconstructed images with limited statistics. However, during the event selection phase of a typical Compton camera image reconstruction, a significant number of events that potentially reflect the source distribution are generally discarded. Effective utilization of these discarded events has the potential to estimate an accurate source distribution from limited statistical data. Thus, we initially developed ComptonNet-v1, a framework designed to directly estimate source distribution by integrating all measured events into a single model. To explicitly implement the difference in contribution between events that interact solely with scatterers, solely with absorbers, or with both, we developed ComptonNet-v2, which integrates these events individually. Consequently, our proposed models exhibited superior performance in both quantitative and qualitative assessments compared with existing models, even under low event statistics. In the future, we plan to implement a more memory-efficient model to estimate the distribution of complex source shapes.
Keywords:
GAMMA-RAY
SINGLE-CRYSTAL
TELESCOPE