Related Experiment Video
Updated: Jun 25, 2025
06:28
Visualization of Low-Level Gamma Radiation Sources Using a Low-Cost, High-Sensitivity, Omnidirectional Compton Camera
Published on: January 30, 2020
12.5K
Precise positioning of gamma ray interactions in multiplexed pixelated scintillators using artificial neural networks
P M M Correia1, B Cruzeiro1, J Dias2,3
1Institute for Nanostructures, Nanomodelling and Nanofabrication (i3N), University of Aveiro, Campus Universitário de Santiago, 3810-193, Aveiro, Portugal.
Biomedical Physics & Engineering Express
|May 23, 2024
Summary
Artificial neural networks (NN) offer improved gamma-ray positioning in positron emission tomography (PET) detectors compared to traditional Anger logic. Both multiple-class and binary classification NNs achieved over 85% accuracy, demonstrating their feasibility for future PET systems.
Area of Science:
- Medical Physics
- Instrumentation
- Nuclear Medicine
Background:
- Positron Emission Tomography (PET) detector positioning traditionally relies on Anger logic flood histograms.
- Machine learning, using signal waveform features, shows promise in advancing PET instrumentation.
- Artificial neural networks (NNs) are explored for enhanced gamma-ray interaction positioning.
Purpose of the Study:
- To evaluate the efficacy of artificial neural networks (NNs) for gamma-ray interaction positioning.
- To compare the performance of NNs against traditional Anger logic in pixelated PET detectors.
Main Methods:
- An experimental setup utilized 16 Cerium-doped Lutetium-based (LYSO) crystal pixels coupled to silicon photomultipliers (SiPMs).
- Data from 160,000 events irradiated by 511 keV gamma rays from a Sodium-22 source were recorded.
- Two NN models were trained and tested: a single multiple-class neural network (mcNN) and 16 binary classification neural networks (bNN).
Main Results:
- Both mcNN and bNN models demonstrated a mean positioning accuracy exceeding 85% on the evaluation dataset.
- The mcNN model exhibited faster training times compared to the bNN model.
- Electronic collimation was employed to mitigate misclassified events from neighboring crystals.
Conclusions:
- Artificial neural networks (mcNN and bNN) can outperform traditional Anger logic for gamma-ray positioning.
- These NN models show feasibility for positioning procedures in future multiplexed detector systems.
- Further improvements may be achieved with advanced configurations like light-sharing.
