Abstract
P1467
Introduction: Organs-on-Chips (OOCs) have gained popularity in radiopharmaceutical research as they provide a more accurate representation of organ physiology. Positron Emission Tomography (PET) can offer an advanced level of analysis of OOCs by imaging physiological processes, yet its spatial resolution is not sufficient for this purpose. To overcome this limitation, we proposed a dedicated On-Chip PET scanner using Monte-Carlo Simulation at the SNMMI 2022 Annual Meeting. In this study, we present improved performance results and showcase the scanner's ability to dynamically image a realistic OOC phantom, demonstrating its potential for pharmacokinetic analysis of OOCs.
Methods: We developed a Monte-Carlo Simulation of our proposed system using GATE [1] to model its interaction with an F18-positron source. The system is equipped with Silicon Photomultipliers (SiPM) on every surface except the front one, which capture the photons emitted from the scintillation. Utilizing the light pattern images generated by the SiPMs, we trained a Convolutional Neural Network (CNN) to predict the gamma-ray interaction positions. These positions were then used to reconstruct the insides of the detector using Ordinary-Poisson OSEM [2]. For more information on the individual steps, we refer the reader to our previous work [3]. Additionally, we incorporated a realistic OOC phantom into our Monte-Carlo Simulation, modelled after a commercially available device consisting of multiple compartments of varying diameters connected by microfluidic channels.
Results: We investigated various design options for crystal thickness, SiPM size, and CNN architecture in order to optimize the performance of our scanner. We found that a combination of 13 mm thick crystals, 3 mm x 3 mm SiPMs, and the ConvNeXt Tiny architecture yielded the best results, with an interaction-position prediction Mean Absolute Error (MAE) of 0.80 mm on the test dataset. The scanner had a sensitivity of 34.81% and a spatial resolution of 0.55 mm, as determined by the mean of the FWHM values in x-, y-, and z-direction of a grid of 7 x 3 point-sources. A qualitative analysis of a realistic OOC phantom demonstrated the feasibility of pharmacokinetic analyses, and a quantitative analysis revealed that 25,000 Line-of-Responses are needed for adequate image quality with a Signal-to-Noise Ratio above 16 dB in the dynamic sequence for source volumes with radii between 0.4 mm and 0.7 mm.
Conclusions: In this study, we showed that it is feasible to attain a high spatial resolution of almost 0.5 mm in a PET system composed of multiple monolithic LYSO crystals. We accomplished this by using a CNN to directly predict the scintillation position from light pattern images captured by SiPMs. Our system has the potential to advance the study of 3D models in radiopharmaceutical research and provide a valuable tool for radio pharmacists in the development of radiotheranostics. In future work, we plan to improve the spatial resolution further by modelling the positron range effect using a deep learning-based approach.
[1] D. Sarrut et al., "Advanced Monte Carlo simulations of emission tomography imaging systems with GATE," Physics in Medicine & Biology, vol. 66, no. 10, p. 10TR03, May 2021, doi: 10.1088/1361-6560/abf276.[2] Comtat et al., "OSEM-3D Reconstruction Strategies for the ECAT HRRT," in IEEE Symposium Conference Record Nuclear Science 2004., October 2004, doi: 10.1109/NSSMIC.2004.1466639.[3] Clement et al., "Concept development of an on-chip PET system," EJNMMI Physics, vol. 9, no. 1, pp. 1–24, May 2022, doi: 10.1186/s40658-022-00467-x.[4] Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, T. Darrell, and S. Xie, "A ConvNet for the 2020s," in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, doi: 10.48550/arXiv.2201.03545.