Online program Overview Session: MIC-16

Abstract

Positron-electron non-collinearity of the 511 keV annihilation-photon has been derived from several experiments. In the one hand, of the (anti-)correlation between Doppler broadening induced by the transverse momentum component of the annihilating pair, and photon non-collinearity. Direct non-colinearity measurements with pin-hole or slits provide also noncolinearity information. In the past, discrepancies in the non-collinearity derived from both kinds of experiments have arised, and the non-collinearity value generally accepted has been solely derived from Doppler broadening experiments. To better understand non-collinearity, we review  the correlation between the  kinematics of the initial electron-positron pair and the emitted non-collinear photon. We then use a detailed 3D simulation of the kinematics annihilation of the e-e+ pair, including the motion of the initial pair of leptons, which in turns introduces the energy broadening of the 511 keV peak, and the non-collinearity of the final photons,  linked to the underlying electron kinetic-energy distribution. To this end, initial electron kinetic energies are sampled from a semi-Gaussian distribution, while initial electron directions are sampled isotropically. We fit the initial kinetic energy distribution of the electron-positron pairs to reproduce, after the detailed 3D simulation, the observables in each kind of experiments, either Doppler broadening or actual non-collinearity in slit experiments. We find that the same initial electron kinetic energy distribution explains both Doppler broadening and accurate slit experiments in water at room temperature. Finally, we compare our extraction for non-collinearity distribution, and found it similar to the one presented by Tuoissant et al in 2024, confirming that non-collinearity is being underestimated in most MC simulations of PET scanners.

Abstract

Positron Annihilation Lifetime has previously been demonstrated to provide useful information in PET imaging. Due to the fact that the annihilation time depends on the tissue where the annihilation occurs, to disentangle the different times for heterogeneous tumors is extremely challenging. Geant4, the toolkit for the simulation of the passage of particles through matter developed at CERN, has been shown to be a good candidate for medical physics simulations. The main drawback is that Geant4 code does not include Positron Annihilation Lifetime information in the source code. The IRIS group at IFIC (Valencia) has developed PAL-G4, an unofficial version of Geant4 that introduces the possibility to add measured Positron Annihilation Lifetimes for different materials in order to be able to simulate realistic scenarios. In order to test the PAL-G4 code, the BrainPET-7T insert for magnetic resonance at the PET-physics group at the Forschungszentrum Jülich GmbH was simulated. Two cases were simulated, one in which the annihilation took place in a homogeneous medium, and another in a heterogeneous medium. The simulation results fit the expected behavior in terms of the detected annihilation times in both cases, thereby showing the suitability of using PAL-G4 for real case studies. Further measurements of annihilation times for various sources and tissues of interest in medical applications will be measured with the BrainPET-7T and used as input parameters for the PAL-G4 code. The simulations results will be compared with the measured data in all cases.

Abstract

Accurate system response modeling is essential for quantitative SPECT, but system response functions (SRFs) generated from nominal collimator geometry can deviate from the response of the actual system due to aperture imperfections, mechanical misalignment, and septal penetration. This mismatch is especially significant for a non-conventional aperture. We propose a data-driven SRF modeling method that reconstructs a directional penetration map from sparse point-source scans. In the proposed aperture-domain representation, each source-detector measurement is indexed by its aperture crossing position and ray direction, allowing measurements from different source locations and detector pixels to be pooled into a shared position-direction map. The map represents the effective material path length, which is estimated by maximum likelihood under a Poisson model with latent detector-depth sublayers, and the resulting problem is solved using an expectation-maximization (EM) algorithm. SRFs for arbitrary source locations can then be generated through the ray-tracing method by combining geometric factors with the learned path length and the photon-energy-dependent attenuation coefficient. Preliminary Geant4 validation using a 16-aperture micro-ring CZT-based SPECT model with controlled collimator tilting demonstrates that scan-based SRFs recover aperture-level response features missed by nominal-design SRFs and agree more closely with the ground-truth tilted-collimator response at non-scanned source positions. Experimental validation on the clinical AlphaSPECT system and image-domain evaluation will be presented at the conference.

Abstract

Scintillation detector inverse problems are limited by how sensitively photon distributions change with interaction or detector conditions. We propose a likelihood-free Wasserstein-geometric framework for quantifying this distributional sensitivity using a pretrained generative model for scintillation photon transport. By representing conditional photon distributions as pushforwards of a latent distribution, the Jacobian of the transport map induces a Wasserstein pullback metric on the condition space. We use the trace of this metric as a local sensitivity measure and its detector-level average as the expected distributional sensitivity (EDS). The proposed sensitivity captures DOI and crystal-dependent timing trends, correlates monotonically with interaction localization error across crystal dimensions (Spearman ρ=-0.88 for rough Teflon and -0.90 for polished ESR), and supports differentiable detector optimization. These results suggest that Wasserstein sensitivity can provide an estimator-independent measure of inverse-problem difficulty and a tool for detector characterization and optimization.

Abstract

Mechanical detector misalignments in positron emission tomography (PET) scanners, caused by manufacturing tolerances as well as mechanical and thermal stress, lead to inaccurate line of response (LOR) positioning and consequently degrade image quality and quantitative accuracy. Existing calibration methods typically rely on dedicated measurements with known tracer distributions, requiring either calibration phantoms or precisely positioned point sources. These procedures are time-consuming, operationally demanding, and difficult to integrate into PET-MRI systems due to magnetic field constraints. This work proposes an unsupervised detector alignment calibration method that estimates detector misalignments from point-source measurements with unknown source positions. The proposed approach estimates source locations using an iterative likelihood-based optimization and derives detector alignment parameters from the resulting LOR geometry, eliminating the need for externally calibrated source positioning. The method is evaluated on a simulated PET scanner and achieves translational and rotational alignment accuracies of 0.95mm and 0.41°, respectively, while the source positions are recovered with an accuracy of 0.61mm using only 3×10⁵ LORs. By avoiding specialized calibration hardware and requiring only limited acquisition data, the proposed method enables rapid, flexible, and easily deployable PET calibration. Its low operational complexity and short acquisition time make it particularly promising for routine clinical recalibration and for hybrid imaging systems such as PET-MRI.

Abstract

Positronium-based imaging [1-3] requires accurate Monte Carlo modelling of positronium decay in matter.  Existing implementations in the Geant4 Application for Tomographic Emission simulation framework [4-5] are limited to at most two annihilation channels, which is insufficient to represent the multi-component lifetime structures observed in biological tissues and complex materials. We present a new modular positronium decay source model implemented in both GATE versions 9.4 and 10 [5], enabling the simultaneous definition of an arbitrary number of decay channels. Each channel is independently characterised by its mean lifetime, branching fraction, annihilation multiplicity into two or three photons, and optional prompt photon emission, as required for radionuclides such as scandium-44, gallium-68, and iodine-124. The model is validated through analytical and numerical benchmarks. Branching fractions are correctly reproduced within statistical uncertainties across all tested configurations. Practical applicability is demonstrated through simulations of a NEMA Image Quality phantom filled with tissue-mimicking positronium sources representing water, fat, muscle, and bone. The implementation is fully compatible with standard simulation workflows, providing the community with the first general-purpose, multi-channel positronium decay model integrated into a widely adopted medical physics simulation framework.