Online program Overview Session: MIC-03

Abstract

Conventional photon-counting systems rely on dead-time models [1], [2] that saturate at high flux, introduce systematic bias, and require per-channel calibration that scales poorly to large pixel arrays [3]. We present LUCID (Linear Unbiased Count-rate from Interval Detection), which estimates photon flux by measuring the silence between pulses rather than counting individual arrivals, eliminating dead-time modelling entirely. The method assumes only that photon arrivals follow a Poisson process — a condition satisfied by all common detector types at constant source output — and requires no detector-specific dead-time model.

For a Poisson arrival process, the memoryless property of the exponential distribution ensures that gaps exceeding a configurable minimum duration θ_gap retain the full statistical information about the true rate λ, regardless of pile-up below that threshold. The unbiased estimator λ̂ = (N−1)/T_excess requires only two hardware-accumulated quantities — valid gap count N and cumulative excess gap duration T_excess — and a single division. No knowledge of pulse shape, dead-time, or detector material is required. The estimator variance Var(λ̂) ≈ λ²/(N_total·exp(−λθ_gap)) provides closed-form uncertainty quantification with graceful degradation at extreme rates rather than a hard saturation limit.

LUCID has been validated from Monte Carlo simulation to commercial deployment, with no per-channel calibration. Simulation demonstrates sub-0.3% RMSE per pixel to 100 Mcps and continued accuracy to 300 Mcps. Eliminating count-rate non-linearity directly improves CT number accuracy, beam-hardening correction, and material decomposition fidelity in photon-counting CT. The architecture requires no high-speed ADC and is suited to FPGA, ASIC, or analogue comparator-integrator realisation [4]. We discuss current work extending the framework toward spectral recovery.

Abstract

Spectral photon counting is widely seen as a revolution in medical imaging, however applications will be limited by the spectral performance of the x-ray sensor. Primarily these sensors are made of semiconductors such as CdTe, however in recent years there has been growing interest in whether scintillators could be used as an alternative. These materials lack the high energy resolution of the semiconductors, however they are not affected by the extensive low energy tailing seen in small pixelated semiconductor devices. As spectral photon counting is performed with coarse energy bins rather than fine spectral resolution, this raises the question of which factor is more important for spectral fidelity in energy binned tasks: a narrow intrinsic energy resolution (semiconductors) or the absence of charge sharing induced low energy tailing (scintillators)? This work sets out to compare the two sensor types to answer this question. To do this we use a new simulation tool called PC-SIM (Photon Counting Spectral Imaging Model), which is a modular framework, designed for flexible modelling of a range of spectral photon counting scanner designs. We combine this tool with Monte Carlo simulations performed in GATE to quantify the spectral fidelity of scintillator- and semiconductor-based detectors under a range of conditions, including various numbers of energy bins, pixel sizes, and differing beam spectra (polychromatic and monochromatic sources). This presentation will introduce PC-SIM, outlining its design philosophy, module options and how it was used to compare the scintillator and semiconductor systems as fairly as possible. Results will show under what conditions scintillators might provide comparable photon-counting performance compared with the current gold standard of semiconductor based detectors, but at a cheaper price-point. Limitations and assumptions of this work will be discussed, along with implications of the results.

Abstract

We propose single-energy material decomposition (SEMD), a method that achieves quantitative basis material imaging from a single X-ray grating interferometry (XGI) acquisition without dual-energy scans. XGI inherently provides two independent spectra from a single acquisition, yielding the spectral separation required for material decomposition. SEMD is particularly valuable for dark-field CT, enabling material decomposition directly from the existing acquisition without requiring dual-energy hardware, greatly enhancing the clinical utility of these systems. Experimental results demonstrate that SEMD achieves decomposition accuracy comparable to conventional dual-energy material decomposition (DEMD), providing absorption, dark-field, and basis material images simultaneously from a single scan.

Abstract

Photon-counting computed tomography (PC-CT) shows remarkable potential in low-dose spectral imaging by binning X-ray photons into discrete energy windows. This study presents a novel PC-CT system integrating scintillation detectors and multi-voltage threshold (MVT) digitizers, extends its application to 3D in vivo spectral imaging, and specifically evaluates the low-dose imaging performance. Validation results show that the system achieves a detection efficiency of 94.78%, a dark count rate of 0.08 cps/pixel, and a peak count rate of 0.85 Mcps/pixel. Dosimetric evaluations at 120 kVp reveal a central CTDI₁₀₀ of 0.15–0.375 mGy, yielding a dose-normalized figure of merit (FOM) nearly tenfold higher than that of conventional micro-CT. Furthermore, in vivo experiments validate the system's capability for high-fidelity 3D material decomposition, even under photon-starved conditions. These results substantiate the potential of the YSO/SiPM-based PC-CT system for precise in vivo material decomposition at low radiation doses.

Abstract

Conventional absorption-based dental CT often fails to detect early tooth microcracks due to limited resolution and attenuation contrast. While X-ray dark-field imaging is highly sensitive to sub-pixel structures, traditional systems lack spectral capabilities. To address this, our study utilized a photon-counting grating-interferometry micro-CT system to perform energy-resolved CT imaging of an ex vivo human tooth. Using a 4.8 μm Talbot-Lau interferometer and dual-threshold detection (15 and 30 keV), absorption and dark-field signals were acquired across total-energy (TE), low-energy (LE), and high-energy (HE) channels. Results confirm the complementary nature of the two modalities: absorption imaging better differentiates enamel and dentin, whereas dark-field imaging reveals microcracks with significantly higher conspicuity.  Regarding spectral performance, although increasing energy leads to higher overall image noise and decreased clarity (LE > TE > HE), microcrack contrast exhibits the exact opposite trend (HE > LE > TE), peaking in the HE channel with the maximum crack contrast. These findings highlight the advantages of photon-counting spectral dark-field micro-CT for highly sensitive dental crack detection.

Abstract

Next-generation X-ray sources, including 4th-generation synchrotrons and Free-Electron Lasers (FELs) such as LCLS-II, deliver photon rates that saturate conventional photon-counting detectors. ePixUHR-100k addresses this gap with a hybrid X-ray imager providing continuous 100 kfps readout with 104 dB dynamic range and single-photon sensitivity down to 2 keV, capturing single photon and > 1000-photon events with low cross-talk. Power-and area-efficient SAR A/D converters distributed across the pixel matrix enable massively parallel digitization, while a charge-integrating front-end with auto-gain switching sustains fluxes beyond 109 photons/s/pixel. Fabricated in a 130 nm CMOS process, the full-reticle 19.4 × 18.0 mm^2 ASIC is bump-bonded to a 144 × 192 pixel Si sensor and achieves 123 e- RMS noise. Validation of a prototype camera system with a 109Cd radioactive source and at the LCLS X-ray FEL confirms state-of-the-art continuous frame-rate and power-efficiency performance among reported X-ray detectors, including SPAD-based designs. Full ASIC architecture, A/D converter characterization, dynamic range evaluation, gain-switching performance, and comparison with state-of-the-art detectors will be presented at the conference. Scalability toward large-area X-ray cameras will also be discussed through tiled 2 × 2 and 2 × 3 ASIC configurations bump-bonded to large-format Si sensors, enabling systems from 0.5 to 16 Mpixels.

Abstract

Photon-counting detectors offer high spatial resolution and energy discrimination for advanced X-ray imaging. However, their performance is limited due to pulse pile-up and charge sharing, which distort the observed current waveforms.
In this study, we propose a Bayesian inverse analysis framework to estimate photon events directly from observed current waveforms. The detector response is modeled by the drift-diffusion equation. Photon events are parameterized by their energy, arrival time, and interaction position. The event parameters are estimated using Bayesian inference. Prior probability distributions of the number of events and interaction depths are incorporated to obtain physically plausible solutions.
Numerical simulation results demonstrated that the proposed method accurately estimated photon events under pulse pile-up and charge sharing conditions. Furthermore, the reconstructed energy spectra showed reduced distortion compared with the conventional method.

Abstract

All components of a first Time-of-flight CT scanner prototype are now assembled at Université de Sherbrooke where both timing and energy measurements enabled to create first line profiles. Previous simulation studies have shown that, by using these measurements, the contribution of scattered photons to an image can be estimated and removed without an using anti-scatter grid. The scanner prototype can now explore and confirm the measurement and scatter correction methods used in simulation studies. The experimental setup is composed of a scintillator coupled to an SiPM readout and a Weerock front end board. This board includes a Radioroc ASIC along with a picoTDC. An FPGA harvest data and send them to a PC for signal processing. From these data, primary and scatter counts in each pixel can be recovered through the use of an iterative deconvolution algorithm. The experiment successfully produced a small photon counting projection image of an ACR phantom and produced time-of-flight distribution measurements for each pixel with around 550~ps FWHM resolution for the full energy spectrum (30 to 120 keV). Scatter correction lead to a contrast resolution improvement of 195\% in the irradiated bone-water region of the phantom. Its accuracy was partially validated with an analytical calculation of the X-ray's penetration through a part of the phantom with a 5.2\% penetration calculated, and a 10.8\% and 5.7\% penetration before and after scatter correction respectively.