Assessing Local Inductive Bias in Deep Learning-Based Three-Gamma Event Classification
Résumé
In the context of the Xenon Medical Imaging System (XEMIS) project-developing the first liquid xenon (LXe) Compton camera for functional three-gamma medical imaging-we explore the application of deep learning methods for event classification based on the number of observed gamma-rays per event. Accurately identifying the number of gamma rays interacting with the liquid xenon (LXe) target per event is crucial for enabling event-byevent targeted data treatment and image reconstruction techniques. To this end, we evaluated the classification accuracy of two deep learning models with or without local inductive bias (Sparse Tensor Network and PointNet). Both models process unstructured raw spatial and light detection data from Geant4 simulated events, aiming to distinguish between single-, double-, and triple-gamma ray interactions. Our results offer insights into the feasibility and performance of sparse tensor architectures in handling complex data typical of Compton imaging systems.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |