#StudentSeminar: Enhancing Compton Camera Imaging with Neural Networks

by Javier Pérez Curbelo (IFIC (CSIC-UV))

Monday 29 Sep 2025, 15:00 → 16:30 Europe/Madrid

zoom-0-0 - zoom (Virtual)

Neural networks are increasingly applied in medical physics to enhance imaging and treatment verification. This work presents three studies where neural networks address major challenges in Compton camera imaging.

The first study investigates event selection for multi-energy radioactive sources. A Compton camera prototype (MACACO III), built with monolithic LaBr3_33​crystals coupled to silicon photomultiplier arrays, was used to collect experimental data from a circular array of 22^{22}22Na sources. Gate v8.0 simulations provided additional training data. Neural networks trained on these datasets effectively discriminated useful events from background, enabling higher-quality image reconstruction compared to conventional energy thresholds.

The second study focuses on proton range verification in hadron therapy. Data collected at a clinical cyclotron with the MACACO III prototype, combined with simulations, were used to train a neural network for signal selection. Compared with energy-cut methods, the neural network significantly increased the fraction of signal events and improved sensitivity to millimetric shifts in depth–dose profiles, a key requirement for accurate monitoring of proton therapy.

The third study employs convolutional neural networks to predict photon interaction positions within monolithic LaBr3_33​crystals. Using simulated data, the network accurately determined 2D and 3D interaction positions, including depth of interaction, surpassing analytical methods in spatial resolution and error metrics.

Together, these studies demonstrate the potential of neural networks to enhance Compton camera performance, enabling improved medical imaging, treatment verification, and ultimately patient care.



ONLY BY ZOOM MEETING: https://cern.zoom.us/j/64257849955?pwd=wcOVGtDl6FyxiiKfjmpsidDaK6PQQS.1 

SeminarUV.pdf