The research team, led by Laura Providência and including Adriaan Lammertsma and Charalampos Tsoumpas, set out to answer a deceptively simple question: when you move brain imaging from a conventional short axial field-of-view (SAFOV) scanner to a LAFOV system, do the numbers actually stay the same? LAFOV scanners, exemplified by the Siemens Biograph Vision Quadra with its extraordinary 106-centimeter axial field of view, capture far more of the body at once and offer substantially higher sensitivity than the 26-centimeter Biograph Vision. But higher sensitivity comes with geometric consequences, including larger acceptance angles that increase scatter contributions and depth-of-interaction effects from oblique lines of response. Before clinics rewrite decades of established protocols, someone had to verify that quantitative accuracy survives the transition.

To do this, the researchers turned to a classic workhorse of PET physics: the Hoffman 3D brain phantom, a physical model filled with a solution of 2-deoxy-2-[F]fluoro-D-glucose ([F]FDG) that mimics the uptake patterns of grey and white matter in a real brain. The phantom was scanned on both the Vision and the Quadra under carefully matched conditions, with acquisition times adjusted to account for radioactive decay so that each scan contained approximately the same total number of decay events. In a second experiment, cylindrical germanium sources containing roughly 73 MBq of activity were placed in direct contact with the phantom to simulate background activity originating from the body, testing whether activity outside the brain would contaminate LAFOV brain images.

The results were strikingly reassuring. Grey matter recovery coefficients, a measure of how faithfully the scanner reproduces true tracer concentrations, came out at 0.96 and 0.97 for the Vision without and with external sources, and 0.93 for all Quadra configurations. Contrast values, which should ideally reach 4 for the Hoffman phantom, were highly consistent at 3.9 across every scanner and acquisition condition. The left-to-right hemisphere ratio, which should be 1 to confirm symmetrical hardware and software performance, ranged from 0.98 to 0.99. Cold-spot recovery, measured in a plastic region simulating a low-uptake midbrain area, was identical at 0.04 everywhere. Estimated image resolution varied by no more than 0.1 millimeters, from 3.3 to 3.4 millimeters.

Noise characteristics told a similar story. The coefficient of variation (COV), an indicator of image homogeneity measured across five spherical regions in the phantom’s white matter, remained stable across all configurations. The Vision produced the highest COV values at 11.3 percent without and 10.7 percent with external sources, while the Quadra, reconstructed with a maximum ring difference of 322, achieved the lowest at 9.6 and 9.8 percent. All values fell comfortably within the 15 percent threshold suggested in the literature, even though no harmonization was performed. The presence of external background activity led to only minor increases in COV, about 0.2 percent for the MRD322 reconstruction, and a marginal 0.1-millimeter increase in estimated resolution, indicating that the potentially higher scatter contribution in LAFOV systems does not meaningfully degrade Hoffman phantom image quality.

One small discrepancy did emerge. The ratio between image-derived activity concentration and dose-calibrator activity was 0.96 and 0.95 for Vision acquisitions and 0.92 and 0.93 for Quadra acquisitions, both meeting the 10 percent acceptable error level of EARL criteria. The approximately 3 percent lower grey matter recovery coefficient observed for the Quadra appears attributable to calibration differences between the systems rather than intrinsic image quality differences. When the researchers recalculated recovery coefficients using the image-derived activity instead of the dose-calibrator activity, the differences vanished entirely, with all values around 1.01, falling within the range reported for non-harmonized brain PET images.

The sensitivity measurements revealed where the real advantage lies. When the Hoffman phantom was positioned at the first detector ring of the Quadra, reflecting common clinical configurations where the brain is imaged alongside other anatomical regions, effective brain-specific sensitivity was 1.2-fold higher than the Vision with MRD85 and 1.4-fold with MRD322. But when the phantom was moved to the axial centre of the 106-centimeter bore, the gain jumped to 1.4-fold and, with MRD322, up to 3.4-fold relative to the conventional scanner. This magnitude is consistent with theoretical predictions estimating a 2.5- to 3.5-fold sensitivity gain for LAFOV systems in single-organ imaging, and notably lower than the 5- to 10-fold increases reported for whole-body acquisitions, because the brain is already fully encompassed within a SAFOV scanner and therefore benefits less from extended axial coverage.

The clinical implications are substantial. With a 3.4-fold sensitivity gain, the same image quality could be achieved with roughly one-third of the injected activity, or the same activity could be imaged in about one-third of the time. For a typical [F]FDG brain scan, current guidelines recommend 37 to 74 MBq, delivering an effective dose of 0.8 to 1.5 millisieverts when combined with an ultra-low-dose CT protocol. Applying the sensitivity gain exclusively to activity reduction would bring that dose down proportionally. Similar reductions apply to amyloid tracers like [F]florbetapir, currently recommended at 109 MBq with an effective dose of 2.1 millisieverts, and tau tracers like [F]florotaucipir at 54 MBq and 1.4 millisieverts. Lower doses would make longitudinal studies safer, reduce cumulative exposure in patients requiring multiple tracers, and open the door to populations currently excluded from brain PET research, including children, healthy volunteers, pregnant women, and intensive care patients.

Shorter acquisition times carry their own benefits, particularly for patients with dementia or movement disorders who struggle to remain still during prolonged scans. Reduced scan duration improves patient comfort, decreases motion-related artifacts, and enhances overall examination feasibility in vulnerable populations. Alternatively, the sensitivity gain could be spent on image quality rather than speed or dose, enabling more robust kinetic modelling approaches such as high-temporal-resolution dynamic imaging, parametric imaging, and reference tissue models that demand cleaner data than conventional protocols typically provide.

The authors caution that these are theoretical estimates derived from phantom measurements, and validation in patient studies is required before clinical implementation. The study also has limitations: image quality was assessed only with the phantom in the first detector ring, sensitivity was evaluated at a limited number of axial positions, all experiments were based on single measurements without repeats, and the Hoffman phantom represents a simplified model of the human brain. Still, the two positions evaluated represent the extreme limits of clinically relevant configurations, and the researchers do not expect additional positions to change the main conclusion. The core message stands: LAFOV PET systems maintain quantitative consistency with SAFOV scanners while offering higher sensitivity, supporting the development of optimized brain PET protocols that reduce injected activity and radiation exposure, improve safety, and enable broader inclusion of vulnerable populations in brain PET studies.

Subject of Research: Long vs. short axial field-of-view PET scanners for brain imaging: a phantom study

Article Title: Long vs. short axial field-of-view PET scanners for brain imaging: a phantom study

Article References: Long vs. short axial field-of-view PET scanners for brain imaging: a phantom study. (n.d.). https://doi.org/10.3389/fnume.2026.1856651

Image Credits: AI Generated

DOI: 10.3389/fnume.2026.1856651

Keywords: Long, short, axial, field-of-view, scanners, brain, imaging, phantom, scientific research