Fast Maximum Likelihood Positioning for a Staggered Layer Scintillation PET Detector

C. Lerche, W. Bi, M. Schöneck, D. Niekämper, Q. Liu, E. Pfaehler, L. Tellmann, J. J. Scheins, N. J. Shah · 2025

Read Fast Maximum Likelihood Positioning for a Staggered Layer Scintillation PET Detector by C. Lerche, W. Bi, M. Schöneck, D. Niekämper, Q. Liu, E. Pfaehler, L. Tellmann, J. J. Scheins, N. J. Shah (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-13723.txt.

Introduction

Christoph Lerche and co-authors published an engineering paper in 2025 on fast positioning for a layered scanner used in brain imaging. The scanner must find which tiny crystal pixel lit up and how much energy arrived. The layered layout lets the team estimate depth too. Their method avoids repeated guessing steps.

Calibration runs supplied the expected light counts the method needs. With parallel tricks and careful memory use, the system handled tens of millions of events per second on a multi-processor machine. A small share of detector blocks needed hand fixes after auto-calibration missed a few pixels. Final energy sharpness came close to that of single pixels, showing the method picks the right pixel.

Catalog record

CREATORC. Lerche, W. Bi, M. Schöneck, D. Niekämper, Q. Liu, E. Pfaehler, L. Tellmann, J. J. Scheins, N. J. Shah
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-13723

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Fast Maximum Likelihood Positioning for a Staggered Layer Scintillation PET Detector C. Lerche, W. Bi, M. Schöneck, D. Niekämper, Q. Liu, E. Pfaehler, L. Tellmann, J. J. Scheins, N. J. Shah Abstract In this study, we propose a fast implementation of a Maximum Likelihood Positioning (MLP) algorithm to estimate the energy and identify the active scintillator pixel in staggered layer scintillation detectors for PET. The staggered layer design with pixelated scintillators enables the determination of the gamma’s depth of interaction and facilitates an iteration-free formulation of the MLP algorithm. The efficacy of the algorithm optimization was tested on a scintillation detector block designed for an ultra-high field BrainPET 7T, comprising three scintillator pixel layers. The three layers contain 24 × 24, 24 × 23 and 23 × 22 scintillator pixels, respectively, with a pixel pitch of 2 mm in both directions and layer thicknesses of 9, 8 and 7 mm. Calibration measurements, in combination with an automated calibration script, were used to obtain the expected counts of scintillation photons required in the MLP algorithm. Using Single-Instruction-Multiple-Data parallelization, multi-threading and optimized cache lines, a maximum processing speed of approximately 22.5 million singles per second was achieved on a platform with four Intel Xeon Platinum 8168 CPUs and 60 threads, encompassing all required processing steps (i.e. input/output, unpacking, timestamp extraction and calibration, single event clustering, MLP, storage). The automatic calibration failed for 1 to 15 individual scintillator pixels in approximately 10% of the 120 scintillation detector blocks, necessitating manual correction. After applying the energy correction to the positioned single events, an energy resolution of Δ ​ E = 12 % ± 2 % ​ FWHM \Delta E=12\%\pm 2\%\,\mathrm{FWHM} was obtained for the entire scintillation block. This value is very close to the energy resolutions measured for the individual scintillator pixels, proving that the MLP accurately identifies the scintillating pixel and that the energy correction method effectively compensates for the light collection variations of the SiPM array. Journal: Phys. Med. Biol. \DeclareDelimFormat newunitpunct \DeclareDelimFormat [bib]nameyeardelim \DeclareDelimFormat [bib]nametitledelim \DeclareDelimFormat [bib]newunitpunct Keywords : Position determination, Maximum Likelihood, Energy correction, Scintillation detector 1 Introduction Scintillation detectors are key components of gamma cameras and positron emission tomographs. A large variety of possible realizations using photomultiplier tubes (PMTs), avalanche photodiodes (APDs) or silicon photomultipliers (SiPMs) in combination with monolithic, semi-monolithic or pixelated scintillators of different materials have been proposed over the years. The output signals from the photodetectors are fed into an attached electronic and computing device to extract the photo conversion position, the energy of the converted gamma photon and the time of photo conversion. As with the design of scintillation detectors, there is a wide variety of methods for extracting these parameters, including centroid-based algorithms, also known as Anger logic, least-square methods and methods from statistical estimation theory, such as Maximum Likelihood Estimation (MLE) ( Joung et al., 2001a ; Barrett et al., 2009 ; Van et al., 2011 ; Liu & Goertzen, 2013 ) . Perhaps the most important property of all these signal processing algorithms is processing speed since the number of scintillation events can easily go beyond 1 million counts per second per detector block in modern PET scanners. Accuracy, precision and robustness are also essential. Compared to other scintillation event positioning methods, statistically-based methods, such as MLE, provide a correct statistical approach which also allows the integration of relevant physical aspects. In this way, simplifying and idealizing assumptions often made to achieve practical solutions can be omitted. Apart from the inherent consideration of the Poisson random nature of scintillation light detection, MLE allows the integration of models for scintillation light propagation, missing signals, Compton scattering and many other effects exhibited by real scintillation detectors ( Barrett et al., 2009 ) . Moreover, the likelihood value can be used as a measure of the estimate quality ( Lodomez, 2012 ; Lerche et al., 2014 ) . Ignoring the real-world effects in the positioning algorithm generally leads to spatial bias and inaccuracies in the determined scintillation event position and energy, necessitating further correction efforts. To address these challenges, maximum likelihood positioning (MLP) has been proposed continuously by various groups since the 1970s, especially for scintillation detectors with monolithic scintillators ( Gray & Macovski, 1976 ; Fessler et al., 1991 ; Gagnon et al., 1993 ; Joung et al., 2000 ; Joung et al., 2001 ; Wenze et al., 2007 ; Miyaoka et al., 2008 ; Lerche et al., 2009 ; Hesterman et al., 2010 ) . In addition, MLP implementations for pixelated scintillation detectors have also been proposed by ( Lerche et al., 2011 ; Lerche et al., 2016 ; Gross-Weege et al., 2016 ; Berker & Schulz, 2018 ; Berker & Schulz, 2018 ) . However, all mentioned MLP methods for determining the energy and position of the scintillation events are based on iterative algorithms, making them very time-consuming and thus potentially impractical for processing scintillation events in a typical PET scan with limited computational resources in a reasonable time. An iteration-free formulation of the MLP algorithm is also of great interest for implementations in Field Programmable Gate Arrays (FPGA) as it takes advantage of accelerated data processing with dedicated circuitry and massively parallelized signal processing pipelines. Several FPGA implementations of MLP algorithms have been proposed ( DeWitt et al., 2010 ; Johnson-Williams et al., 2011 ; Wang et al., 2016 ) . In this work, we present an iteration-free MLP algorithm for pixelated scintillation detectors with depth of interaction (DOI) detection capabilities based on a staggered scintillation layer design ( Ito et al., 2010 ) and the algorithm previously presented in ( Lerche et al., 2016 ) . As in the original implementation, the result of the position determination is the scintillator pixel ID together with the calibrated gamma photon energy. Thus, the additional association of the continuous x and y centroid (also called Anger) positions with these scintillator pixel IDs using Voronoi tessellations or similar approaches ( Du et al., 1999 ) can be avoided during normal data acquisition. Although the implementation of the presented MLP algorithm in FPGAs is not the focus of this work, all special requirements of the specific FPGA hardware relevant for such an implementation have been taken into account ( Lerche et al., 2019 ) . 2 Methods 2.1 Iteration-free ML implementation The log-likelihood of a gamma photon interacting in a single elementary process, i.e., photoelectric effect or Compton effect, and depositing the energy, E_{\rm d} , in the scintillator pixel, i , given that the scintillation photons {\bf q}=(q_{1}\mathord{\mathchar 59\relax}q_{2}\mathord{\mathchar 59\relax}\ldots\mathord{\mathchar 59\relax}q_{N^{PD}}) were detected on the photodetectors (1\mathord{\mathchar 59\relax}2\mathord{\mathchar 59\relax}\ldots\mathord{\mathchar 59\relax}{N^{PD}}) , is given by \mathcal{L}_{i}\left(i\mathord{\mathchar 59\relax}E_{\rm d}|{\bf q}\right)=\ln\prod_{j}^{N_{PD}}\frac{e^{-\bar{q}_{i\mathord{\mathchar 59\relax}j}}(\bar{q}_{i\mathord{\mathchar 59\relax}j})^{q_{j}}}{q_{j}!}\mathchar 59\relax (1) as the detection process is a Poisson process. In eq. 1 , N_{PD} denotes the total number of individual photodetectors or photodetector elements j of the photodetector array. The total amount Q_{E}=\sum_{j}^{N_{PD}}q_{j} of scintillation photons detected in all photodetectors j is proportional to the energy deposition E_{\rm d} , the gain or photon detection efficiency (PDE) of the photodetector, and the light yield of the scintillator. The latter two are assumed to be constant during any measurement. \bar{q}_{i\mathord{\mathchar 59\relax}j} is the expected average amount of detected scintillation photons, i.e. avalanches, in the

Read the complete text