Online program Overview Session: M-LT-02

Abstract

Monolithic scintillators are a promising alternative to standard pixelated crystals in positron emission tomography, as they can be more cost-effective, reduce dead space, and capture an unobstructed light distribution, which can be used for depth of interaction estimation. We focus on the improvements seen by using an array of integrated corner reflectors (ICR) machined directly into the entrance face of the monolith, which we previously showed via simulation to improve light collection and narrow the light distribution. In this study, we built a physical prototype of the device using a CNC machining process. We compared this prototype with other monolithic block designs. We show that the ICR monolith with the CNN positioning algorithm has potential for high-resolution imaging with ~1 mm resolution at 12 mm thickness. Using GATE simulations, we constructed a full scanner with the performance of the monolithic blocks and tested the capabilities compared to existing systems. We show the improved spatial resolution and feasibility for a non-cylindrical design for axillary lymph node scanning.

Abstract

Pixelated crystal arrays are the standard in commercial PET scanners. However, these detectors usually lack DOI information, essential to achieve a homogenous spatial resolution along the entire FOV. Yet, studies have been already performed with pixelated detectors with DOI capabilities but at the cost of using extra hardware material and/or crystal assembling. Inspired by semi‑monolithic crystal geometries, pseudo-slabs were proposed as an alternative. This geometry consists of an N×N pixel array optically isolated only in one direction, allowing light crosstalk in the orthogonal one. The N pixels sharing light in this direction form a pseudo-slab. The light sharing within a pseudo-slab is enough to enable DOI retrieval, while the physical separation between pixels confines the light more than in a pure slab geometry, yielding better time resolution.

In this work we evaluated the DOI resolution of a BGO block formed by 8×8 pixels, each measuring around 3×3×15 mm³. Two different readout configurations were employed and compared. First, all the 64 channels corresponding to the TOFPET2 ASIC channels were digitized. Then, a multiplexing circuit was utilized, reducing the signals to 8+8 by summing rows and columns. The DOI was obtained using neural networks trained on experimental data.

Abstract

High-sensitivity, high-resolution brain-dedicated SPECT systems are critically important for advancing diagnosis and surgical planning in complex neurological disorders such as epilepsy, Parkinson's disease, and Alzheimer's disease. In this paper, we present SurfaceCam, a close-proximity multi-pinhole SPECT architecture with a radially adaptable rotating gantry, designed specifically for clinical brain imaging. The system comprises 15 detector-collimator panels distributed across three detector heads in a cylindrical arrangement, operating at a 1:8 minification geometry with a 28 cm bore. Each panel integrates 16 high-performance CZT sensors in a compact 2×8 configuration. The three detector heads translate radially to accommodate varying patient head sizes while maintaining close-proximity geometry, and the rotating gantry enables micro-rotation between acquisitions to facilitate angular super-sampling strategies. Pinholes are organized into groups of four with a twisted geometry, each subtending partial coverage of the 20 cm spherical FOV, while eliminating inter-pinhole projection multiplexing on the detector surface. Analytical comparison against a conventional 1:11 minification compound-eye reference design demonstrates approximately 1.4-fold higher geometric sensitivity and approximately 27% improvement in effective pinhole spatial resolution. The front-end electronics and data acquisition systems have been fully developed and validated, with Monte Carlo simulation studies and experimental phantom validation currently underway.

Abstract

Photon-counting CT (PC-CT) is a powerful spectral imaging technique that identifies materials by counting individual X-ray photons. While direct-conversion systems are entering clinical practice, we are developing a cost-effective, large-area scintillator-based PC-CT system using YGAG:Ce coupled with silicon photomultipliers (SiPMs). PC-CT excels at identifying high-atomic-number elements like gold (Au) by utilizing its K-edge (80.7 keV). Gold nanoparticles (AuNPs) are promising for drug delivery, as their biodistribution can be controlled by particle size, enabling selective tumor accumulation via the enhanced permeability and retention (EPR) effect.

In this study, we evaluated the in vivo biodistribution of AuNPs in tumor-bearing mice using our SiPM-based PC-CT. Two sizes of AuNPs (1.9 nm and 15 nm) were intravenously administered, and their accumulation was quantitatively assessed using multi-energy CT concentration estimation. Results showed distinct size-dependent patterns: 1.9 nm AuNPs exhibited rapid renal clearance and bladder accumulation, while 15 nm AuNPs showed prolonged circulation and significant hepatic accumulation. Tumor accumulation was confirmed for both sizes, demonstrating the potential of our SiPM-based PC-CT for quantitative in vivo nanoparticle imaging.This presentation reports on the
experimental setup, the analysis methods, and the results obtained.

Abstract

X-ray computed tomography (CT) is essential nondestructive internal imaging. Photon-counting computed tomography (PC-CT) acquires X-ray energy information to enable material identification and quantitative density evaluation. Combining PC-CT energy information with magnetic resonance imaging (MRI) data is expected to provide a novel diagnostic approach to visualize inflammation through tissue water fraction estimation. We developed a PC-CT system combining high-speed scintillators (YGAG:Ce) and silicon photomultipliers (SiPMs), specifically utilizing multi-pixel photon counters, MPPCs, which are optical semiconductor devices. This study evaluated healthy and fatty liver rat models (fed a high-fat diet for two or four weeks). Liver fat fractions, estimated from energy-dependent linear attenuation coefficients, showed good agreement with MRI and pathological results. Furthermore, we performed a detailed pathological evaluation focusing on inflammation induced changes in tissue water content. By integrating high reliability MRI fat fraction data with PC-CT data, we improved the estimation accuracy of water content and quantitatively visualized tissue ”water-richness” (edema). Increased water proportions in fatty liver groups suggested inflammation progression. These findings demonstrate that PC-CT is highly valuable for quantitative evaluation of liver disease within a multi-modality imaging approach.

Abstract

The reconstructed spatial resolution of a PET image is determined by both intrinsic scanner properties and the chosen reconstruction algorithm. Standard NEMA measurements use point sources reconstructed with FBP and  therefore mainly characterize the scanner.  They are thus suboptimal for predicting the resolution of images reconstructed with  MLEM and point-spread-function modeling, as is common in clinical practice.

We propose a 3D-printed resolution cube with an off-center inner compartment which, combined with a mid-acquisition phantom flip and a single reconstruction, produces sharp edges at finite (2:1) contrast along three orthogonal directions. Position-, direction- and contrast-dependent edge and line spread functions (ESFs, LSFs) are then extracted using smoothing-spline fits.  As a proof of concept, the method is applied to the NeuroExplorer brain PET system at the isocenter and 10 cm off-center, with and without depth-of-interaction (DOI) modeling.

The LSFs are non-Gaussian with pronounced negative side lobes.  With DOI, the resolution is approximately 2.5 mm FWHM in all directions at both positions; without DOI it degrades radially to about 4 mm at 10 cm off-center.  Notably, differences between finite and infinite contrast are small in most configurations.

The method provides a position-, direction- and contrast-dependent characterization  of PET resolution from clinically reconstructed images and complements NEMA point-source values for image quality comparison, harmonization and partial volume correction.

Abstract

Monte Carlo scatter correction is well established as a highly accurate approach for scatter estimation in PET/CT imaging. It demonstrates superior robustness compared with model based scatter estimation methods, particularly under extreme variations in activity distribution and in the presence of patient motion, conditions under which model based approaches often produce artifacts. Despite these advantages, Monte Carlo scatter correction is computationally intensive, and does not inherently account for out-of-field scatter. In this work, we address both limitations and propose an implementation suitable for integration into routine clinical workflows. A multi-GPU implementation, as well as parallel processing with efficient data streaming significantly reduce the processing time. To handle out-of-field scatter, we separately handle regions where information about out-of-field distribution is known, and can be used for a highly precise estimation. In regions where out-of-field information is inaccessible, the effect on scatter is approximated by the measured data, while ensuring a smooth transition in which the high-accuracy region governs the final estimate. Our approach is scalable and robust across system designs ranging from a standard axial field-of-view to a long axial field-of-view, each presenting distinct considerations for scatter estimation. We demonstrate the effectiveness of our method using phantom and clinical data.

Abstract

Accurate motion correction is critical for dynamic brain PET, particularly on the high-resolution NeuroEXPLORER (NX), where residual head motion can blur focal targets and bias time-activity curves (TACs). We evaluated multiple motion correction (MC) strategies using 17 multi-tracer human NX PET scans, including no motion correction (NMC), post-reconstruction registration (PRR), registration before attenuation correction (RBA), UMT-based frame-averaged correction (AVG), and UMT-based event-by-event correction (EBE). Performance was assessed using qualitative image comparison and ROI-based residual TAC variability (%SD_res) from 20-70 min. All MC methods reduced %SD_res compared with NMC. Across all tracers, subjects, and ROIs, EBE achieved the lowest residual variability (1.40 ± 1.27%), followed by PRR (1.79 ± 1.49%), RBA (1.94 ± 1.64%), AVG (2.01 ± 1.63%), and NMC (4.68 ± 4.26%). Poisson noise-floor analysis further showed that residual motion effects were tracer- and count-statistics-dependent. Overall, UMT-based EBE provided the most effective reduction of residual motion effects in dynamic NX brain PET.

Abstract

Recent increases in computational power as well as new tools being more readily available have led to increased interest in Monte Carlo scatter estimation. However, Monte Carlo scatter simulation is still computationally expensive. This is exacerbated by the fact that model-based scatter estimation requires multiple iterations of scatter simulation and image reconstruction. We propose to combine the current clinical standard (analytic single scatter simulation and tail-fitting) with Monte Carlo scatter simulation.

Abstract

The spatial resolution of PET imaging is fundamentally limited by the positron range, whose extent depends on both the emission energy of the radionuclide and the electron density of the surrounding medium. In integrated PET/MRI systems, the static magnetic field exerts a Lorentz force on the positrons, confining their trajectories in the transverse plane to the main magnetic field, potentially improving image resolution. Two complementary experiments were performed using a dedicated monolithic-crystal based PET insert operated at 0T and inside a 9.4T MRI: capillary line sources embedded in three tissue-equivalent phantoms of increasing density (EPF, PLA, and Rigid10k resin), and a microDerenzo phantom. Both tests were carried out with three positron emitters (¹⁸F, ⁸⁹Zr, and ⁶⁸Ga) eventually use in pre-clinical imaging. The results demonstrate that the magnetic field significantly enhances transverse spatial resolution for high-energy positron emitters in medium- and high-density materials, with FWTM reductions of ~60% for ⁶⁸Ga and ~18% for ⁸⁹Zr, while the axial direction remains largely unaffected. Rod resolvability improved at 9.4T from 1.5 mm to ~1.0 mm and from 1.2 mm to 0.9 mm for ⁶⁸Ga and  ⁸⁹Zr, respectively. . These findings highlight the relevance of both positron energy and tissue-equivalent density in determining the magnitude of magnetic-field-induced resolution improvements in preclinical PET/MRI.