Concept Development of an On-Chip PET System.

Area of Science:

• Medical Imaging
• Microdevice Technology
• Nuclear Instrumentation

Background:

• Organs-on-Chips (OOCs) are increasingly used for disease modeling and drug discovery.
• High-resolution imaging is crucial for OOC applications, but current Positron Emission Tomography (PET) systems lack the necessary spatial resolution.
• There is a growing demand for advanced imaging techniques to support OOC research.

Purpose of the Study:

• To propose and demonstrate an On-Chip PET system for high-resolution imaging of OOCs.
• To develop a novel method for predicting gamma-ray interaction positions using Convolutional Neural Networks (CNNs).
• To achieve sub-millimeter spatial resolution for OOC imaging.

Main Methods:

• Designed an On-Chip PET system with four detectors, each comprising monolithic Lutetium-yttrium oxyorthosilicate (LYSO) crystals and Silicon photomultipliers (SiPMs).
• Utilized a CNN trained with Monte Carlo Simulation (MCS) data to predict gamma-ray interaction points from SiPM light patterns.
• Employed Simultaneous Algebraic Reconstruction Technique (SART) for image reconstruction using predicted Line of Responses (LORs).

Main Results:

• The CNN achieved a mean average prediction error of 0.78 mm.
• A mean spatial resolution of 0.53 mm was obtained when imaging a grid of 21 point sources.
• Demonstrated the feasibility of achieving nearly 0.5 mm spatial resolution in a PET system using monolithic LYSO crystals and CNN-based scintillation position prediction.

Conclusions:

• The proposed On-Chip PET system, leveraging CNNs and monolithic LYSO detectors, can achieve high spatial resolution suitable for OOC imaging.
• CNNs, particularly ResNet architectures, show superior performance for predicting scintillation positions compared to EfficientNet.
• Specific detector surfaces provide more informative patterns for accurate scintillation-point prediction.