Speaker
Description
Nowadays, the bulk of the cost in TBPET systems comes from the scintillator, followed by SiPMs and electronics. Commercial TBPET scanners typically use LYSO, which is very expensive and relies on rare-earth elements whose production and manufacturing are geographically concentrated and supply-limited. Thus, the first priority of any putative low-cost TBPET is to find significantly cheaper alternatives to LYSO.
In this work, we propose CRYSP, a novel TBPET scanner based on large monolithic crystals of pure cesium iodide (CsI) operating at cryogenic temperatures ($\sim$100 K). Cooling pure CsI from room temperature increases its light yield by a factor of 20, reaching ~105 photons/keV, though its decay time also increases from 15 ns to 800 ns. The resulting energy resolution, below 7% at 511 keV, combined with the monolithic form factor and its inherent depth-of-interaction capability, allows millimeter-scale spatial resolution in all three dimensions, minimizing parallax error. These features more than compensate for the lower sensitivity and lack of time-of-flight, yielding performance comparable to state-of-the-art TBPET scanners at significantly reduced cost.
The slow scintillation signal makes conventional commercial PET readout solutions, such as PETsys, unsuitable for cryogenic CsI, as the integration window is too short to capture the full scintillation pulse. To overcome this, an alternative front-end based on the CAEN CITIROC ASIC was adopted. Unlike PETsys, CITIROC provides peak-hold and pulse-height analysis measurements well suited for slow scintillators, allowing reliable extraction of deposited energy. A dedicated study evaluated its performance across various sources and crystals, comparing it with the available commercial solutions, PETsys and a waveform digitizer. This demonstration of CsI’s intrinsic performance motivates the use of monolithic geometries coupled to SiPM arrays. Unlike pixelated arrays, a monolithic crystal preserves the continuous light distribution, enabling event localization in all three dimensions. A dedicated deep-learning algorithm was developed to further improve position reconstruction. Trained in PyTorch on simulated Geant4 events, the algorithm is able to identify in Monte Carlo data up to two interaction vertices per event, achieving millimeter-scale resolution in all three dimensions, including depth-of-interaction. Moreover, the algorithm will be also validated against experimental data.
As a first experimental milestone, the group is currently assembling a small-animal PET prototype composed of 18 monolithic CsI(Tl) crystals arranged in three rings, operating at room temperature. This proof-of-concept system will validate the full acquisition and reconstruction pipeline before proceeding to a fully cryogenic implementation.
| Track | TBPET |
|---|---|
| Presentation type | Oral |