Jagiellonian Positron Emission Tomography

• Jagiellonian Positron Emission Tomography is a PET system that replaces traditional crystal detectors with long plastic scintillator strips arranged in modular cylindrical layers for cost-effective, large field-of-view imaging and positronium studies.
• The technique employs dual-end readout and precise time sampling, enabling high temporal and spatial resolution essential for both diagnostic imaging and discrete-symmetry tests.
• J-PET integrates dedicated reconstruction software and calibration methods to support advanced positron annihilation lifetime spectroscopy and multi-photon imaging, enhancing PET performance.

Jagiellonian Positron Emission Tomography (J-PET) is a positron emission tomography technology developed at the Jagiellonian University that replaces conventional inorganic crystal detectors with long plastic scintillator strips arranged axially in cylindrical layers. It has been developed simultaneously as a cost-effective large-field-of-view or whole-body PET concept and as a detector for positronium lifetime spectroscopy, discrete-symmetry tests, and photon-polarization studies, with reconstruction driven primarily by timing rather than by photopeak-based energy measurement (Niedźwiecki et al., 2017, Czerwiński et al., 2017, Moskal et al., 2016).

1. Detector concept and architecture

J-PET is built from long organic scintillator bars rather than short high-$Z$ crystals. In the first full-scale prototype, 192 detection modules are arranged axially in three non-overlapping cylindrical rings, forming a diagnostic chamber with inner diameter $85\ \text{cm}$ and axial field-of-view $50\ \text{cm}$. The strips are made of EJ-230 plastic scintillator, with dimensions $7 \times 19 \times 500\ \text{mm}^3$, and are read out at both ends by Hamamatsu R9800 photomultipliers. An earlier 24-module prototype used BC-420 strips of $5 \times 19 \times 300\ \text{mm}^3$ and R4998 photomultipliers (Niedźwiecki et al., 2017).

The basic detector module was characterized independently as a BC-420 strip of $5 \times 19 \times 300~\text{mm}^3$ read out at both ends by Hamamatsu R5320 photomultipliers. In that implementation, the long strip functions simultaneously as scintillator and lightguide, exploiting the fast rise and decay times of plastic scintillation and the long attenuation length of the material (Moskal et al., 2014).

The same architectural principle has been extended to other geometries. Simulation studies of preclinical total-body J-PET for rodents use axially oriented EJ-230 strips read out at both ends by silicon photomultipliers, with a mouse design of $23\ \text{cm}$ axial field-of-view and $11\ \text{cm}$ diameter, and a rat design of $30\ \text{cm}$ axial field-of-view and $16\ \text{cm}$ diameter. These systems retain the defining J-PET features of long strips, dual-end readout, modular rings, and timing-based localization, while adding wavelength-shifting arrays for improved axial information (Dadgar et al., 2024).

2. Timing-centric readout, calibration, and detector response

J-PET readout is organized around precise time sampling. In the 192-strip detector, each scintillator is read out at both ends, and the photomultiplier signals are sampled at four amplitude thresholds; for each threshold, both the leading and trailing edges are recorded, yielding up to eight time measurements per hit. The corresponding Time-over-Threshold (TOT) values serve as a proxy for deposited energy and enable discrimination between annihilation photons and the $85\ \text{cm}$0 de-excitation photon from $85\ \text{cm}$1 (Czerwiński et al., 2017).

The reconstruction formalism is explicitly timing-based. For a strip hit, the position along the strip is obtained from the time difference of the two end signals,

$85\ \text{cm}$2

where $85\ \text{cm}$3 is the effective propagation speed of light in the scintillator. The hit time is reconstructed from the average of the end times. At the coincidence level, J-PET uses the time-of-flight difference

$85\ \text{cm}$4

and the annihilation point displacement along the line of response is

$85\ \text{cm}$5

The overall TOT observable in the full-scale prototype is expressed as

$85\ \text{cm}$6

These relations are central both for imaging and for positronium studies (Niedźwiecki et al., 2017).

Precise synchronization of all timing channels is therefore indispensable. A dedicated calibration method irradiates each module with a $85\ \text{cm}$7 source and a small plastic reference detector, determining first the relative offset between the two ends of a strip and then the common reference time for all modules in a layer. In that formulation,

$85\ \text{cm}$8

which yields

$85\ \text{cm}$9

For inter-layer synchronization the additional flight times were calculated as $50\ \text{cm}$0 and $50\ \text{cm}$1 (Skurzok et al., 2017).

At the module level, a single BC-420 strip achieved interaction-time resolution better than $50\ \text{cm}$2 ($50\ \text{cm}$3) for a single-level discrimination, spatial resolution along the strip of about $50\ \text{cm}$4 ($50\ \text{cm}$5), and fractional energy resolution

$50\ \text{cm}$6

corresponding to $50\ \text{cm}$7 at the Compton edge (Moskal et al., 2014). At the system level, the 192-module prototype reported a preliminary Coincidence Resolving Time of about $50\ \text{cm}$8 ($50\ \text{cm}$9) using only single-threshold timing, whereas simulations of matrix SiPM readout predicted $7 \times 19 \times 500\ \text{mm}^3$0 CRT for $7 \times 19 \times 500\ \text{mm}^3$1 axial field-of-view and $7 \times 19 \times 500\ \text{mm}^3$2 for $7 \times 19 \times 500\ \text{mm}^3$3 axial field-of-view (Niedźwiecki et al., 2017, Moskal et al., 2016).

3. Reconstruction methods and software ecosystem

J-PET reconstruction is supported by a dedicated software stack. The J-PET Framework is an open-source C++ library based on ROOT that provides a common environment for reconstruction, calibration, filtering, and user-level PET data analysis. It accepts low-level acquisition data, oscilloscope data, high-level tomography structures such as sinograms and lists of lines-of-response, and interfaces with Monte Carlo packages such as GEANT and GATE (Krzemien et al., 2020).

A central algorithmic theme is library-based signal comparison. In the Mahalanobis-distance method, threshold-crossing times from a measured signal are compared with synchronized model signals stored in a library, and the similarity measure is

$7 \times 19 \times 500\ \text{mm}^3$4

with $7 \times 19 \times 500\ \text{mm}^3$5 the difference vector between measured and model timing coordinates and $7 \times 19 \times 500\ \text{mm}^3$6 the covariance matrix constructed for a given library entry. In the two-threshold realization studied in the double-strip prototype, the timing vector was $7 \times 19 \times 500\ \text{mm}^3$7, and the achievable time and position resolution depended on the number and values of the thresholds (Sharma et al., 2015).

The framework has also been extended toward positronium spectroscopy and imaging. A dedicated positron annihilation lifetime spectroscopy analysis code compatible with the J-PET software uses iterative fitting of multi-exponential spectra convolved with Gaussian resolution functions and was shown to perform as well as LT 9.0 on hexane data (Dulski et al., 2017). For voxel-wise positronium imaging, a multi-photon TOF MLEM method computes system-matrix elements on the fly for coincidences containing two annihilation photons and one de-excitation photon, accumulates time-difference histograms per voxel, and estimates mean ortho-positronium lifetimes from those histograms (Shopa et al., 2023).

4. Imaging performance, scanner variants, and quantitative system studies

Representative configurations and reported results span single modules, whole-body-oriented prototypes, preclinical scanners, and normalization studies.

Configuration	Key geometry	Representative result
Single detection module	BC-420, $7 \times 19 \times 500\ \text{mm}^3$8, dual PMT readout	better than $7 \times 19 \times 500\ \text{mm}^3$9 time resolution ($5 \times 19 \times 300\ \text{mm}^3$0); $5 \times 19 \times 300\ \text{mm}^3$1 position resolution ($5 \times 19 \times 300\ \text{mm}^3$2)
192-module prototype	192 EJ-230 strips in three layers, $5 \times 19 \times 300\ \text{mm}^3$3 AFOV	preliminary CRT $5 \times 19 \times 300\ \text{mm}^3$4 ($5 \times 19 \times 300\ \text{mm}^3$5)
Mouse TB J-PET (simulation)	AFOV $5 \times 19 \times 300\ \text{mm}^3$6, diameter $5 \times 19 \times 300\ \text{mm}^3$7, $5 \times 19 \times 300\ \text{mm}^3$8 strips	sensitivity $5 \times 19 \times 300\ \text{mm}^3$9; volumetric spatial resolution $5 \times 19 \times 300~\text{mm}^3$0
Rat TB J-PET (simulation)	AFOV $5 \times 19 \times 300~\text{mm}^3$1, diameter $5 \times 19 \times 300~\text{mm}^3$2, $5 \times 19 \times 300~\text{mm}^3$3 strips	sensitivity $5 \times 19 \times 300~\text{mm}^3$4; volumetric spatial resolution $5 \times 19 \times 300~\text{mm}^3$5
Modular large-FOV scanner normalization	24 modules, 13 strips per module, $5 \times 19 \times 300~\text{mm}^3$6 strips	geometrical corrections reduce SD/mean intensity ratio to $5 \times 19 \times 300~\text{mm}^3$7–$5 \times 19 \times 300~\text{mm}^3$8

The single-module values are reported in (Moskal et al., 2014), the 192-module prototype in (Niedźwiecki et al., 2017), the mouse and rat total-body simulations in (Dadgar et al., 2024), and the normalization result in (Coussat et al., 2024).

The preclinical simulation study concluded that the mouse and rat scanners are almost parallax-free systems and that a $5 \times 19 \times 300~\text{mm}^3$9 acceptance-angle cut yields only small changes in radial, tangential, and axial full widths at half maximum, implying that the usual acceptance-resolution tradeoff is weak in that geometry (Dadgar et al., 2024). In a separate large-field-of-view modular system study, component-based normalization factors derived from Monte Carlo simulations compensated the non-uniformities of a uniform cylindrical phantom and produced satisfactory correction of axial and radial artifacts, with a computation scheme described as straightforward to parallelize and appropriate for long axial field-of-view scanners, including total-body J-PET under development (Coussat et al., 2024).

5. Positronium spectroscopy, multi-photon reconstruction, and symmetry tests

A distinctive feature of J-PET is its use as a positronium detector. Because positronium is a purely leptonic bound state and an eigenstate of charge conjugation and parity, it provides a laboratory for tests of discrete symmetries whose precision is, in principle, limited by weak-interaction effects at the level of $23\ \text{cm}$0 and photon-photon interaction effects at the level of $23\ \text{cm}$1. J-PET is designed to determine the spin of ortho-positronium, photon momenta, and photon polarization-related observables, enabling measurements of symmetry-odd operators such as

$23\ \text{cm}$2

and

$23\ \text{cm}$3

as well as operators involving $23\ \text{cm}$4, $23\ \text{cm}$5, and $23\ \text{cm}$6 more generally (Moskal et al., 2016, Raj et al., 2018).

Commissioning studies established practical event-selection strategies for positronium decays. In one dataset, the selection required exactly three hits in a $23\ \text{cm}$7 time window and TOT values corresponding to two $23\ \text{cm}$8-like annihilation photons plus one $23\ \text{cm}$9-like prompt photon. Background rejection used angular correlations ordered so that $11\ \text{cm}$0, with true $11\ \text{cm}$1-Ps $11\ \text{cm}$2 events populating the region $11\ \text{cm}$3, while $11\ \text{cm}$4 events with scattering clustered near $11\ \text{cm}$5 or near the origin. The distance between the three-hit plane and the source center provided an additional discriminator. In that context the expected sensitivity to CP-violation observables was stated to be at the order of $11\ \text{cm}$6 (Czerwiński et al., 2017).

J-PET has also been used to develop searches for charge-conjugation-forbidden decays. Simulations and experimental selections demonstrated that the detector covers almost the whole phase space of $11\ \text{cm}$7-Ps $11\ \text{cm}$8 and can isolate likely three-photon events using TOT and the kinematic condition that the sum of the two smallest relative azimuthal angles satisfies $11\ \text{cm}$9. A planned next step is separation of allowed $30\ \text{cm}$0-Ps $30\ \text{cm}$1 from forbidden $30\ \text{cm}$2-Ps $30\ \text{cm}$3 by positronium lifetime spectra (Chhokar, 2019).

The ability to access photon polarization through Compton scattering extends J-PET beyond momentum-only analyses. In the time-reversal feasibility study, the polarization direction of an annihilation photon was estimated as

$30\ \text{cm}$4

where $30\ \text{cm}$5 and $30\ \text{cm}$6 are the photon momenta before and after scattering, and the T-odd observable $30\ \text{cm}$7 was proposed for event-by-event evaluation. With an upgraded detector, the rate of $30\ \text{cm}$8-Ps $30\ \text{cm}$9 events accompanied by one additional scattered photon was estimated at about $16\ \text{cm}$0, corresponding to a projected sample of order $16\ \text{cm}$1 events during one year of data collection and an expected order-of-magnitude improvement in sensitivity with respect to previous analyses (Raj et al., 2018).

These methodologies have been exercised in data. In precision studies of ortho-positronium decays in XAD-4, J-PET reported an experimentally determined mean lifetime

$16\ \text{cm}$2

in agreement with the literature value $16\ \text{cm}$3, and measured the $16\ \text{cm}$4-Ps $16\ \text{cm}$5 fraction as $16\ \text{cm}$6, consistent with the Monte Carlo value $16\ \text{cm}$7 (Dulski et al., 2020).

6. Integration of PET, PALS, and positronium imaging

The convergence of PET imaging and positron annihilation lifetime spectroscopy is a recurring theme in J-PET development. A dedicated lifetime-spectrum analysis program compatible with the J-PET framework models a spectrum as a sum of exponential components convolved with one or more Gaussian resolution functions and was benchmarked against LT 9.0 using hexane data, with the conclusion that it performs as good as LT 9.0 (Dulski et al., 2017).

Detector commissioning in view of positronium lifetime spectroscopy then showed that J-PET is capable of performing simultaneous imaging of the density distribution of annihilation points as well as positron annihilation lifetime spectroscopy. In measurements with a $16\ \text{cm}$8 source surrounded by porous XAD4, multi-component lifetime fits yielded ortho-positronium lifetimes in agreement with standard PALS, while the effective resolution function was described by two Gaussian components (Dulski et al., 2018).

This integration depends on calibration infrastructure. For the 24-module prototype, beam-profile studies of a dedicated collimator system showed that two lead bricks of $16\ \text{cm}$9 form a beam of annihilation quanta with a Gaussian profile characterized by $85\ \text{cm}$00 FWHM. The beam-profile characterization was described as essential for time and energy calibration and for constructing the library of model signals used in hit-time and hit-position reconstruction (Kubicz et al., 2016).

More recent reconstruction studies extend the same logic into voxel-wise positronium imaging. In the multi-photon TOF MLEM formulation, the estimated time differences between annihilation and positron emission are accumulated into histograms for each voxel and updated by the activity weights reconstructed by TOF MLEM. Under the simplified model of ortho-positronium decays into back-to-back photons, the estimated mean lifetimes were consistent with simulation and distributed quasi-uniformly at high MLEM iterations, while the authors explicitly noted that the technique can be upgraded to include correction factors and more realistic decay models (Shopa et al., 2023).