Online program Overview Session: MIC-09

Abstract

The coincidence time resolution of scintillation detectors plays a critical role in time-of-flight positron emission tomography (TOF-PET) and it is fundamentally limited by the emission, transport, and detection processes of scintillation photons. Although previous analytical and Monte Carlo simulation studies have suggested that exploiting multiple timing information of scintillation photons could approach the theoretical timing limits, direct experimental access to all detected photon timestamps remains unavailable with standard PET detectors using analog silicon photomultipliers (SiPMs). To address this challenge, we previously proposed a deep learning-based approach to recover photon timestamps on an event-by-event basis by removing the SiPM degradation factor from detector waveforms within an unsupervised learning framework. Our experimental evaluations demonstrated that the recovered photon behavior is consistent with previous findings, although the absence of ground-truth timestamps inevitably limited a rigorous and quantitative assessment of its validity and reliability. In this study, we evaluate the proposed method using Monte Carlo simulation data with fully known photon-level information. The recovered photon timestamps were quantitatively compared with the true timestamps in terms of timing bias, timing uncertainty, and total photon count fidelity. The results showed that the proposed method achieves timing biases within approximately 10 ps and maintains acceptable timing precision across photons while accurately preserving the number of detected photons per event. With the optimized parameter settings, we will further improve temporal precision in timestamp recovery using experimental data.

Abstract

In positron emission tomography (PET), inter-crystal scattering (ICS) is a major source of uncertainty contributing to spatial resolution. While rejecting ICS events can improve the resolution, it comes at the cost of sensitivity. Accurately recovering the first interaction position is therefore crucial for maintaining both high sensitivity and spatial resolution.
In this work, we apply ICS recovery to experimental data from semi-monolithic detectors, with a 2:1 slab-to-SiPM ratio using both analytical and deep-learning (DL) approaches. We developed a GATE Monte Carlo simulation framework tuned to reproduce experimental signals for data augmentation. A multilayer perceptron (MLP) achieved ~78% exact layer prediction, with an additional ~15% misassigned by only one slab, leaving just ~7% exceeding 1.5 mm (one slab thickness). This outperforms analytical methods, particularly for ICS-only events, where the MLP is 16 percentage points more accurate, and the analytical method fails entirely at edge slabs. We show simulated signals closely match experimental data; additional tuning will enable their use in training networks for experimental application. 
This study takes a first step toward combining simulated and experimental data for DL-based ICS recovery, which could bypass individual detector array calibration.

Abstract

High-resolution SPECT imaging requires detectors with high intrinsic spatial resolution, good energy resolution, and high detection efficiency. Pixelated scintillator arrays are commonly used to improve the intrinsic spatial resolution, but sub-millimeter pixels increase fabrication complexity and cost, and inter-crystal gaps may reduce detection efficiency.

In this work, we developed and evaluated a semi-monolithic GAGG SPECT detector composed of 29 stacked continuous crystal slabs. Each slab measured 25.8 × 0.8 × 6 mm³, and adjacent slabs were separated by 0.08 mm-thick BaSO₄ reflectors. The detector was coupled to a silicon photomultiplier (SiPM) array (Hamamatsu MPPC S14161-3050HS) and read out using a PETsys TOFPET2 ASIC. A tungsten parallel-slit collimator with 13 slits, a 2 mm pitch, and a 0.8 mm slit width was placed directly in front of the detector entrance surface for flood image, energy resolution (ER), and intrinsic spatial resolution (ISR) measurements.

Most pseudo-pixels (defined as discrete areas on the flood image based on the crystal look-up table) were clearly resolved in the flood image, except for several edge crystal pixels, with average peak-to-valley ratios of 5.4 ± 2.1 and 7.4 ± 3.9 in the X and Y directions, respectively. Within a 110–170 keV energy window, DOI-based energy correction improved the average ER from 15.37% ± 1.75% to 11.07% ± 0.77% at 140 keV. The average ISR were 1.02 ± 0.36 mm and 0.42 ± 0.29 mm in the X and Y directions, respectively.

In conclusion, these results demonstrate that the proposed semi-monolithic GAGG detector can achieve good energy resolution and sub-millimeter-level intrinsic spatial resolution, showing potential for high-resolution SPECT imaging.

Abstract

Recent applications of Positron Emission Tomography (PET) in dosimetry such as Pair Production Tomography (P2T) or Proton Range Verification (PRV) require a detector count rate capacity far beyond the conventional range for diagnostic-oriented PET. Most commonly used PET detectors use scintillation crystal coupled to silicon photomultipliers (SiPMs). For high count rate situations the scintillation decay time in the range of tens of nanoseconds can lead to undesired pile-up effects.

In this work, we explore the feasibility of using Cherenkov-only emitter based on PbF₂ coupled to SiPMs to overcome the limitation posed by the long scintillation tails. While Cherenkov-only emitters have been proposed and tested for ultra-fast timing in time-of-flight PET (TOF-PET), the goal of this work is to evaluate the feasibility of using them in readily scalable setup.

We used 3×3×3-20 mm³ PbF₂ crystals coupled to Broadcom NUV-MT SiPMs (4×4 mm²) and custom low-noise high-frequency (LN-HF) readout electronics. SiPM over-voltages from +2 to +8 V and different trigger levels were evaluated to determine the optimal performance for this type of application, focusing on the true coincidence fraction (TCF), coincidence ratio and coincidence timing resolution (CTR), in order to assess if the Cherenkov signals are discernible from background noise. Additionally, we tested the same setup read out by an ASIC (FastIC evaluation board) to understand the impact of an integrated system-applicable readout. Using LN-HF-readout electronics and an excess bias voltage of +4 V we measure a TCF of 81% and a CTR of 240±2 ps. Using the FastIC ASIC, the best results were a TCF of 81% and a CTR of 506±6 ps. Our results demonstrate the potential of Cherenkov-only radiators for high count rate PET applications.

Abstract

A prototype stationary tomographic ultra-fast Molecular Breast Imaging system (UFMBI) has been developed and characterised using phantom studies. The system combines ultra-high resolution CZT detectors with depth‑of‑interaction (DOI) sensitivity and wide‑angle sampling from densely packed multi‑pinhole (MPH) collimators, together with novel de‑multiplexing reconstruction to mitigate multiplexing artefacts. This approach enables substantial reductions in scan time (∼40 min) and radiation dose (∼2.4 mSv) compared with conventional MBI, supporting wider adoption as a supplemental screening tool for women with dense breast tissue.
A dual-head prototype with a field-of-view of 132mm x 88mm has been designed and built. Each head consists of a 6x4 array of 5 mm thick CZT detectors with 11x11 grid of 2 mm pixels. Raw detector data are processed using a machine learning model providing 10x10 sub-pixelisation and DOI sensitivity. Image reconstruction uses OSEM algorithm incorporating attenuation, collimator blurring, septal penetration and multiplexing effects.
Performance was evaluated using uniform and spherical lesion phantoms (5–9 mm) at tumour‑to‑background ratios of 10–20:1, with comparison to a clinical planar MBI system by imaging ^99mTc-infused gel phantoms. Dose/time reduction factors (DRF) were evaluated by resampling measured data relative to the reference equivalent to current clinical practice. Lesion detectability was assessed using contrast‑to‑noise ratio (CNR) and the Rose criterion (CNR>3). Sphere studies demonstrated detectability of 5–7 mm lesions at DRF = 8. Gel phantom experiments showed that UFMBI achieved ~2.9× higher CNR than the clinical system at reference conditions and ~2× higher at DRF = 6.
These results demonstrate the feasibility of UFMBI as a high‑throughput tomographic extension of conventional MBI, with potential application as a supplemental screening tool for women with dense breasts.

Abstract

The ultrafast nature of Cherenkov photon emission offers an avenue to improve the coincidence time resolution (CTR) of time-of-flight positron emission tomography (TOF-PET) detectors. Bismuth germanate (BGO) provides a higher Cherenkov photon yield than conventional lutetium-based crystals, but its slow scintillation component induces large variations in signal shape. Measuring the rise time of the signals, obtained from the time difference between two threshold crossings, enables classification of events by Cherenkov yield and time-offset correction to improve the CTR. While this principle has been demonstrated with dedicated fast electronics, scalable implementation remains a challenge. Here, we exploit the dual-threshold capability of the new FastIC+ ASIC to extract rise-time information in BGO-based detectors and study the influence of bias voltage, settings of thresholds, and rise-time correction methods on CTR performance. We obtained a 32% CTR improvement, from 577 to 395 ps FWHM, with 3×3×15 mm³ polished BGO coupled to 3×3 mm² FBK NUV-MT SiPMs biased at 49 V, and we show trade-offs between noise rejection and rise-time correction efficiency as a function of the two selected thresholds. This work marks a concrete step toward a scalable approach to exploiting BGO's mixed Cherenkov–scintillation emission in TOF-PET imaging.

Abstract

Simultaneous PET and SPECT imaging enables concurrent characterization of multiple biological processes, but conventional multi-modality systems typically require separate detector subsystems. We previously proposed a scintillator-as-collimator design for a PET&SPECT shared detector, and demonstrated PET, SPECT, and simultaneous PET&SPECT imaging capabilities. Nevertheless, the scintillator-as-collimator volume (SCV) lacked depth-of-interaction (DOI) discrimination in PET mode, and the thickness of the SPECT Detector volume (SD) was not optimized, compromising its time-of-flight (TOF) PET performance. To address these limitations, we developed an optimized two-layer PET&SPECT detector with a top 16×16 LYSO array (1.53×1.53×10 mm³) and a bottom 8×8 LYSO array (3×3×10 mm³), separated by two 2-mm light guides. The block was coupled to an MPPC array and read out by a PETsys TOFPET2 ASIC. To evaluate the effect of hole drilling on performance, two prototype detectors were fabricated: a drilled detector with holes in the top LYSO array and light guides, and a non-drilled reference. Both detectors used unpolished crystal sides with BaSO₄ reflectors. Flood images, energy resolution (ER), DOI resolution, and coincidence time resolution (CTR) were evaluated.
​​​​​​​The drilled detector achieved effective DOI discrimination in both SCV and SD. After excluding ICS events, DOI resolutions of 3.60 ± 0.43 mm and 2.18 ± 0.15 mm were obtained in the top and bottom layers, with a bottom-to-bottom CTR of 259 ps. The DOI and CTR are comparable to those of the non-drilled reference detector.
In conclusion, these results validate the optimized two-layer scintillator-as-collimator design and demonstrate its potential for a  PET&SPECT shared detector with high DOI and TOF performance. Future work will evaluate the capability of the SD region in detecting 140 keV gamma photons and develop a prototype system for imaging validation.

Abstract

Time-of-Flight Positron Emission Tomography requires readout electronics capable of extracting precise timing information. The Positron Emission Tomography Integrated Circuit (PoETIC) has been developed to bridge the performance gap between the intrinsic timing resolution of the detector and that achievable with integrated readout solutions, while remaining scalable in channel count and power consumption. PoETIC introduces a novel channel architecture based on two fully independent analog front-ends per channel, separately optimized for timing and energy measurements, and thereby relaxing conflicting design requirements at the input stage. Fabricated in 65 nm CMOS technology, the PoETIC die integrates fast low-noise front ends, dual-threshold discriminators and a TDC with 24.4 ps time binning. Four 16-channel dies are combined into a 64-channel 2.5D Integrated Circuit. A custom high-speed serial interface supports daisy-chained data and trigger distribution, significantly reducing interconnect complexity. The power consumption is 10 mW per channel. Measurements of a first prototype show a single-photon time resolution of 57 ± 1 ps FWHM and 85 ± 1 ps FWHM using FBK NUV-HD-MT SiPMs of 1×1 mm² and 3×3 mm², respectively. Preliminary measurements yield a coincidence time resolution of 74 ± 1 ps FWHM with the 3×3 mm² SiPMs and LGSO crystals of 2×2×3 mm³. To the best of our knowledge, these results represent state-of-the-art timing performance for an ASIC and suggest that PoETIC approaches the intrinsic timing performance of scintillator crystals while offering a flexible and scalable readout solution for next-generation PET systems and other fast-timing applications.