Online program Overview Session: RTSD-07

Abstract

High-resolution 3D Compton imaging is essential for quantitative environmental monitoring. Traditional Maximum Likelihood Expectation Maximization (MLEM) suffers from voxel discretization penalties. Without ground-truth references, MLEM's required early stopping creates a dilemma between severe depth-elongation and late-stage noise deterioration. Meanwhile, modern deep learning models suppress noise but lack zero-shot generalization. To bridge this gap, we present a zero-shot, physics-informed 3D Gaussian Splatting (3DGS) framework. Although likelihood computation utilizes differentiable voxelization, the fundamentally continuous nature of 3DGS introduces implicit spatial regularization, bypassing rigid voxel constraints. It rigorously decouples the morphological probability density from overall source activity via an analytical optimal scaling mechanism. In simulations, 3DGS eliminates background artifacts, yielding up to 5.7 dB Peak Signal-to-Noise Ratio (PSNR) improvement over MLEM in 4-view phantom reconstructions. Validated experimentally on an H3D M400 detector within a 0.125 m³ workspace, 3DGS overcomes MLEM's depth-elongation and noise deterioration without arbitrary early stopping. It achieves an angular resolution of ~9° and excels in practical multi-isotope identification while strictly preserving relative activity linearity. This training-free paradigm offers a robust, quantitative tool for radiation monitoring.

Abstract

H3DD_UM ASIC has been developed at the University of Michigan since 2017. This paper summarizes the performance of the most recent version H3DD_UMv5 ASIC. The electronic noise was measured to be 1.2 keV FWHM at 700-keV dynamic range with a 20×20×15 mm³ CdZnTe detector attached but not biased. Such low noise enables an excellent energy resolution of 0.37% FWHM at 662 keV for single-pixel events. For all events, the measured energy resolution is 0.52 % FWHM. The low-energy threshold can be as low as 2.0 keV using the 200 keV dynamic range. The nonlinearity of the AISC is around 0.6% and it can be calibrated using the internal test pulse.

Abstract

We report recent progress in the development of CdZnTe (CZT) and CdZnTeSe (CZTS) position-sensitive virtual Frisch-grid (VFG) gamma-ray detectors and detector arrays. The high 3D position sensitivity enabled by the VFG design allows correction of response non-uniformities caused by crystal defects. A key challenge in current large-volume CZT/CZTS detectors is the lateral displacement of charge carriers, which may exceed 1-2 mm over a >1 cm drift distance, strongly affecting the imaging capabilities of gamma cameras. These displacements are closely tied to crystal defects, such as dislocations and sub-grain boundaries, which are always present even in the highest-quality crystals. To address this issue, we propose using detectors with smaller-than-typical cross-sectional areas (4×4 or 5×5 mm²), which match the size of the basic element in a coded-aperture imaging system. Improving position and energy resolution through the application of neural network algorithms is critical for the development of high-performance and reliable detector systems. We demonstrate that the neural network approach yields better results and requires significantly less calibration data because it handles the continuous distribution of events within the detector, whereas the conventional correction-matrix approach relies on discretized values resulting from detector segmentation. We present the results of testing these devices and discuss their advantages and shortcomings.

Abstract

Spectrum unfolding, or deconvolution, is a critical preprocessing step in X-ray image formation, particularly for spectral photon-counting detectors (PCDs). While an ideal detector records the full energy of an incident photon at a single pixel, realistic semiconductor detectors are governed by complex physical phenomena. Charge sharing, K-fluorescence escape, and pulse pile-up are the dominant factors that distort the recorded energy of the recorded signal. Thes first two effects are significantly exacerbated as pixel sizes decrease, leading to spectral degradation that compromises material discrimination and CNR, challenging the multi-energy imaging system.
This study presents a comprehensive characterization of the energy response function (ERF) for a PCD module developed by AMS-OSRAM, utilizing a CdZnTe (CZT) crystal with a native pixel pitch of 341 µm × 365 µm coupled to three ultra-low noise ASICs. The work was conducted at the HEX-27 (High Energy X-ray Scattering) beamline at Brookhaven National Laboratory. The detector was exposed to monochromatic X-ray beams from 30 keV to 140 keV at a flux of 9 Mcps/mm². The complete spectral response was mapped at high resolution, by sweeping the detector’s five energy thresholds. The measurements enabled the derivation of an empirical model describing the ERF of the AS5920M/CZT module, accurately capturing inherent spectral features and tailing effects. To verify this empirical model, the derived response was convoluted with ideal spectra generated via SpekCalc. The resulting reconstructed distorted spectra matched well the measured detector output.
Finally, we propose a correction methodology applied to individual energy bins. By compensating for spectral distortions, enhanced contrast-to-noise ratio (CNR) can be achieved, thereby significantly improving the accuracy of material separation.

Abstract

Future hybrid pixel detectors for particle tracking and imaging require dense, low-inductance interconnections that can distribute clean power, sustain multi-gigabit data transmission, and minimize non-sensitive area. Conventional wire-bonding constrains input and output access to the die periphery, and results in long internal power routing as well as dead area which prevents close tiling of chips. This contribution presents a three-dimensional integration approach for the readout of hybrid pixel detectors based on Through-Silicon Vias (TSVs) fabricated after complementary metal-oxide-semiconductor processing. Timepix4 readout chips are used as a test vehicle. Vias are etched from the back side of thinned chips, insulated from the substrate, metallized, and connected to a back-side redistribution layer that routes power and signals to an array of bonding pads on which solder balls can be deposited. The resulting interconnects replace wire bonds and allow the chip to be soldered directly to a carrier printed circuit board. Several processed Timepix4 batches have been evaluated. A custom-designed test system allows measurement of interconnect quality for signal transmission and power delivery. TSV-processed chips have also been co-integrated together with silicon-photonics into module demonstrators. The tests demonstrate successful operation of Timepix4 after TSV processing, including 5 Gbit/s data transmission and stable power delivery through the back-side interconnect stack. The presentation will discuss the processing flow, test system design choices, assembly and test results, and the suitability of solder ball-grid arrays and anisotropic conductive paste for permanent chip-to-board attachment.

Abstract

Gallium arsenide (GaAs) has broad application prospects as a material for photon counting detectors (PCDs). In this study, we investigate carrier dynamics and polarization in Cr-compensated GaAs (GaAs:Cr) via infrared transient current analysis and clinical X-ray temporal response experiments. The transient current results indicate that while massive carrier generation under high-intensity illumination can induce localized polarization, the deep energy levels introduced by Cr doping effectively preserve dynamic charge neutrality. By capturing and recombining excess carriers, this Cr-compensation successfully mitigates polarization effects, significantly improving detector signal stability up to an ultra-high clinical flux of 2.7 Gcps/mm^2.

Abstract

Advances in the fabrication of detector-grade GaAs semiconductors through chromium compensation have enabled the development of imaging sensors with high detection efficiency, energy resolution, high-flux tolerance, and enhanced radiation hardness, providing improved performance in the mid-energy X-ray range compared to conventional Si and Cd(Zn)Te sensors. Detection efficiency in semiconductor sensors strongly depends on sensor thickness, as the active detection volume scales directly with it. High absorption efficiency up to 80 keV is required in medical computed tomography (CT) and security imaging, which can be achieved using 2 mm thick GaAs sensors.

GaAs sensors with thicknesses from 0.3 mm to 2 mm manufactured by Advafab were developed, characterized, and evaluated with different photon-counting readout electronics. High-resistivity GaAs wafers were processed using Cr compensation, followed by lithography, metallization, dicing, and flip-chip bonding to various readout ASICs. The sensors were evaluated under X-ray irradiation to investigate detection efficiency, count-rate uniformity, charge transport, stability, and imaging performance. The 2 mm thick sensors demonstrated excellent image uniformity and operation at extreme X-ray fluxes up to 1100 Mcnt/s/mm2 with stability better than 2% over 30 s. Sidewall etching of diced GaAs improved sensor performance by reducing leakage current and enhancing electric field uniformity. The results highlight GaAs:Cr as an effective material for photon-counting X-ray imaging in medical, synchrotron, and industrial applications.