Online program Overview Session: MIC-20

Abstract

This study presents the design and preliminary performance evaluation of a long-axial field-of-view (LA-FOV) preclinical SPECT system for small-animal total-body imaging (TBI) [1]. The proposed full scanner consists of 240 CZT detectors distributed across 10 panels and coupled with a multi-pinhole collimator. The system offers a FOV of 74 mm in diameter, ideal for rat imaging, a high-sensitivity central region of 37 mm in diameter tailored for mice, and an axial coverage up to 20 cm. 

System performance was characterized through simulations to compute the system response function, 3D sensitivity map, and reconstructed image quality, assessed using a digital long-axial Derenzo phantom under both noiseless and Poisson-noise conditions. A 50% checkerboard sparse configuration was compared against the full configuration [2-3]. While the sparse configuration yielded a proportional 50% reduction in sensitivity, the volumetric uniformity was preserved. Both configurations achieve similar spatial resolution. In noiseless conditions, both can resolve hot rods as small as 1.62 mm with average peak-to-valley ratios (PVR) of 4.35 (full) and 5.32 (sparse). Under noisy conditions, both configurations can fully resolve the 1.93-mm hot rods with average PVR of 7.22 (full) and 7.58 (sparse). Notably, this performance is consistent along the 20 cm axial extent of the phantom.

Future work will evaluate spiral sparse geometries at sparsity levels of 25%, 33%, 66%, and 75% to characterize the degradation of image quality, with quantitative analysis in terms of contrast-to-noise ratio and contrast recovery coefficient. This work supports a pathway toward cost-effective LA-FOV preclinical SPECT instrumentation suitable for TBI.

Abstract

Multiplexed PET (mPET) enables simultaneous multi-tracer imaging by detecting prompt gamma in positron plus prompt gamma emitters for triple-photon coincidence tagging, thereby distinguishing them from pure positron emitters. ⁶⁸Ga is a commonly used PET isotope and emits a prompt gamma at 1077 keV in ~1% of its positron decays. In this study, we assessed the feasibility of ⁶⁸Ga/¹⁸F mPET on the GE Omni 128 cm total-body PET/CT using Monte Carlo simulation.

The scanner model was validated against NU2-2018 sensitivity and NECR measurements. A NU2-2018 image-quality phantom was simulated with a 1:1 ⁶⁸Ga/¹⁸F background (2.65 kBq/mL per isotope) and 4:1 hot-to-background ratio, with the two largest spheres assigned to ¹⁸F and the four smaller spheres to ⁶⁸Ga to represent a challenging case for triple-coincidence detection. Isotope unmixing employed an “alternating-update joint EM (AU-EM) algorithm” that we developed with a prompt-gamma detection efficiency map simulated from a ⁶⁸Ga point-source array.

The simulation yielded 3.46 billion double and 57 million triple coincidences; triple events were dominated by randoms (55.6%), with ⁶⁸Ga true triples comprising 7.1%. Using true events only, AU-EM reconstructions successfully separated the ⁶⁸Ga and ¹⁸F distributions and achieved percent contrast comparable to OSEM reconstructions from ground-truth events, although percent background variability was degraded. These results demonstrate the feasibility of ⁶⁸Ga/¹⁸F mPET on a total-body PET scanner and motivate future work on random triple-coincidence estimation, variance reduction methods, and experimental validation.

Abstract

Positronium lifetime imaging (PLI) enables probing tissue microstructure via the lifetime of ortho-positronium (o-Ps). Robust voxel-wise lifetime estimation remains limited even for high-sensitivity long axial field-of-view (LAFOV) positron emission tomography (PET) systems due to a combination of low detection sensitivity for triple-coincidences and variability in o-Ps lifetime (o-PS-L) extraction methods. In this work, three established approaches for o-Ps-L estimation are presented: least-squares fitting (LS), Bayesian fitting (B), and exponential fitting of the late decay tail (LT). These methods were systematically evaluated using inorganic samples (Quartz, wet and dry Amberlite XAD-4) prepared with I-124. Data were acquired on a Siemens Healthineers Biograph Vision Quadra system in single event acquisition mode and analyzed using time difference distributions (TDDs). The o-PS-L for the Quartz sample was measured to be 1.572 | 1.573 | 1.574 ns (LS | B | LT), wet XAD-4 2.381 | 2.381 | 2.798 ns and dry XAD-4 2.719 | 2.719 | 2.915 ns. LS and B approaches showed consistent results across all samples, while the LT method exhibited systematic deviations under reduced o-Ps contribution. A subsampling analysis identified a minimum of 2500 events per TDD for stable o-PS-L estimation within 5% deviation from the full-statistics result. Based on this constraint, voxel-wise lifetime imaging was demonstrated in a metastatic thyroid cancer patient following injection of 37 MBq I-124 using a 15 min acquisition. To satisfy required statistics, a voxel size of 32 mm was selected, and lifetime estimates were restricted to regions with sufficient counts. These results identify count statistics as primary limiting factor for clinically viable PLI and define a practical threshold for stable voxel-wise lifetime imaging. Future work will focus on extending acquisition times to enable higher spatial resolution and improve robustness for clinical applications.

Abstract

The Patlak plot is a widely used kinetic modeling method for estimating the tracer net influx rate K_i in dynamic PET imaging. However, the requirement for a full time course input function limits its routine clinical use. The Relative Patlak (RP) plot mitigates this constraint effectively by reducing the required acquisition time. Nevertheless, the quantified RP slope K_i' introduces a time-dependent and subject-specific scaling factor relative to K_i, reducing the reliability for inter-subject analysis. In this study, we propose an Intercept Graphical (IG) plot with an associated normalization that requires the same imaging duration as the RP method. The quantified parameter K_IG has a time-independent and approximately subject-independent scaling factor for estimating K_i. We evaluated the proposed method using total-body dynamic PET images of 12 healthy and 10 cancer subjects. To evaluate inter-subject quantification performance, multi-organ and lesion K_i' and K_IG were generated using the RP and IG methods, respectively, and then rescaled using population-mean scaling factors to recover K_i. Using the Patlak K_i as the reference standard, the organ or lesion K_i estimated by the population-rescaled IG had an error that was 50–70% lower than that of the population-rescaled RP. Furthermore, group analysis indicates the K_i yielded by the population-rescaled IG exhibits statistical behavior more similar to that of the Patlak K_i and outperforms the population-rescaled RP K_i. Overall, the proposed IG approach with normalization retains the low imaging time requirement and improves inter-subject quantitative robustness. It may enable more accurate K_i quantification and improve the clinical feasibility of PET parametric imaging.

Abstract

The selection of compartmental models for dynamic PET analysis is compromised by three key scenarios—the unknown kinetics of novel tracers, mixed kinetics resulting from tissue heterogeneity and spillover, and the diverse multi-organ kinetics in total-body PET—where conventional model selection using nonlinear least squares and Akaike Information Criterion can fail. For this, we augmented our previous framework--kinetic parameter posterior estimation using generative consistency models--to perform Bayesian model selection. The objective is to simultaneously estimate both the model probability and the posterior distributions of the kinetic parameters across six commonly used kinetic models. The method involves training the consistency model on 6,000,000 simulated time-activity curves. Evaluation against a conventional method demonstrates significant improvements; across 6,000 simulated time-activity curves, the proposed method achieved an overall model selection accuracy of 86%, substantially outperforming the 34% accuracy of the conventional method. Furthermore, application to clinical total-body positron emission tomography data from a patient with metastatic cancer yielded physiologically plausible model selection maps that successfully differentiated complex underlying regional kinetics, such as healthy liver tissue versus tumour. Our results indicate that this proposed framework is useful in kinetic model selection in dynamic total-body PET and can have potential for evaluating new radiotracers in clinical practice.