In this paper we present a study of machine learning (ML) algorithms to simplify the computation of the planar scintillation coordinates in Anger Cameras for emission tomography applications. Two ML-based techniques for data inference and one technique for speed-up the training procedure are explored within the framework of a multimodal SPECT scanner. Firstly, the use of Principal Component Analysis (PCA), a dimensionality reduction algorithm, is explored to reduce the computational complexity of maximum-likelihood statistical estimation method. The analysis indicates a 3-fold reduction of computational complexity for typical Anger Camera architectures (with 72 channels). Secondly, the estimation of the scintillation coordinates is formulated as a classification problem, addressed by means of a Decision Tree (DT) classifier. No degradation of the achievable intrinsic spatial resolution ( 1.2:mm FWHM) of the detection module was observed when applying PCA (reducing from 72 to 25 components). The DT classifier was trained on experimental data obtained using a parallel-hole collimator: again no degradation of spatial resolution is observed and the computation cost is reduced by more than two orders of magnitude. Finally, in order to overcome the limits of a cumbersome training procedure involving the translation of the collimator, data augmentation was successfully leveraged for the generation of artificial data.
18 July 2024 in IEEE Transactions on Nuclear Science
Charge sharing between pixels is a significant concern in X-ray spectroscopy detectors based on monolithic arrays of Silicon Drift Detectors (SDDs) when a photon is absorbed near a pixel edge. Traditional mechanical collimation mitigates charge sharing but decreases the active area of the detector. This is particularly true for small-pixel detectors, where the mechanical collimator shields a significant portion of the active area and, moreover, becomes quite complex to be manufactured for small apertures. In this work, we carry out a study of charge sharing in SDDs and introduce active collimation as an alternative to mechanical collimation, leveraging key parameters of signals at the output of the charge sensitive amplifier (CSA), to identify charge-sharing events occurring within a defined coincidence window in neighboring pixels. To validate the technique, we first conducted a quantitative assessment of the impact of charge sharing on a 16-element monolithic SDD module with 2-mm-side square pixels using a focused pulsed laser. We then tested an active collimation algorithm on this detector configuration during a spectroscopic measurement with an uncollimated 55 Fe source, demonstrating its capability in recovering charge-sharing events and reconstructing them as Mn-Kpeak events. This active collimation approach, in contrast to mechanical collimation, not only enhances the effective active area while achieving a substantial reduction in the background continuum of the spectrum, but also recovers information inevitably lost in uncollimated detectors due to intrinsic charge sharing effects.
22 September 2025 in Journal of Synchrotron Radiation
Among the many factors that determine data acquisition efficiency in synchrotron radiation experiments, detector throughput is a primary bottleneck. As modern synchrotron sources provide increasingly intense X-ray beams, the signal processing time of individual detector channels often becomes the limiting factor for throughput. State-of-the-art digital pulse processors (DPPs) can mitigate this bottleneck by minimizing the processing time per channel, thereby improving overall event throughput. In this work, we evaluate the performance of the ARDESIA-16 detection system, based on a 16-channel monolithic silicon drift detector array, when coupled with two advanced DPPs – DANTE+ and FalconX – which employ different signal processing architectures. Using a standard 55 Fe source, we systematically characterize and compare output count rate, energy resolution, dead time and spectral quality across a broad range of input count rates, from 40 kcps to 2.7 Mcps. Our results demonstrate the excellent performance of ARDESIA-16 with both DPP systems while highlighting a key trade-off between energy resolution and event throughput that requires thorough optimization for high-rate synchrotron applications.
X-ray spectrometers with high count rate capability are becoming essential to provide higher speeds in synchrotron measurements, like in X-ray fluorescence mapping (XFM). Monolithic multichannel silicon drift detectors (SDDs) enable the creation of dense and high-performance matrices to reach this goal. In this paper we show the results of the development of ASCANIO, a 16-channel backscattering X-ray spectrometer based on SDD monolithic modules. The solid angle and count-rate uniformity in X-ray detection have been optimized thanks to a `tilted' configuration of the SDD units. ASCANIO underwent preliminary testing in the laboratory, where the system achieved around −42°C with an internal pressure of 5.1 × 10 −6 mbar. During further verification of its spectroscopic performance, the spectrometer showed resolutions as low as 134 eV FWHM at the Mn K α peak. After preliminary testing the full system was commissioned on beamline P06 at PETRA III (DESY). Here the effects of the tilted configuration could be observed, and the system achieved a 20 Mcps cumulative output count rate with a dead time of 20%. As a demonstration of ASCANIO's usability, an example use case is demonstrated, where ASCANIO was used for an XFM analysis of a biological sample.
Monolithic arrays of Silicon Drift Detectors (SDDs) have advanced X‐ray spectroscopy by providing excellent energy resolution and high rates at near ambient temperatures. Despite these benefits, they are affected by charge sharing (CS), where photogenerated charge clouds spread across adjacent channels. This effect causes partial charge collection by neighboring pixels and impacts spectroscopic performance. To address this issue, we developed CASPER‐AI (Charge‐sharing Analysis and Spectroscopic Performance Enhancement via Reconstruction through AI), an innovative approach that leverages Machine Learning (ML) and particularly Decision Trees (DT) for effective CS mitigation. CASPER‐AI combines DT classification with CS event reconstruction, thus enhancing spectral quality and recovering useful signals that would otherwise be discarded. Experimental characterization with a monolithic SDD‐based 16‐channel detection module demonstrated significant improvements in peak‐to‐tail ratio and full‐energy event recovery.