Beyond Mechanics: Maximum-Likelihooddriven Pet Detector Alignment

Abstract

Mechanical inaccuracies during PET detector assembly pose a major challenge to quantitative accuracy, as true crystal positions often deviate from design due to manufacturing tolerances and installation errors. High-resolution PET systems, in particular, are highly sensitive to misalignments, leading to distorted lines of response, reduced signal-to-noise ratio, and significant degradation in image quality and quantification. Conventional calibration methods rely on regression models, single-point sources, and precise motion systems, which are often inflexible and costly. We propose a statistical optimization framework that directly estimates scanner alignment parameters from measured list-mode coincidence data using stochastic gradient descent. Our method maximizes the likelihood of observed coincidences given only a known tracer distribution. This formulation supports arbitrary tracer distributions, simplifying the calibration process and increasing its adaptability. We validate our approach on simulated and real PET systems, including phantoms with complex activity distributions and a real scanner with artificially introduced misalignments. Our method achieves sub-millimeter positional accuracy of$\sim 45 \mu ~\mathrm{m}$and angular precision of$\sim 0.2^{\circ}$, closely matching the accuracy of a mechanically verified blueprint. The precision of recovered alignment parameters improves with finer crystal binning, directly translating into sub-millimeter spatial resolution in reconstructed images. By enabling softwarebased, in-system recalibration, this approach improves imaging stability in mobile and mechanically unstable environments, while relaxing manufacturing constraints without sacrificing LOR accuracy, making it a powerful tool for modern PET scanner calibration where quantification fidelity is critical.

Authors 4

1. Julian Thull Aachen Institute of Imaging & Computer Vision

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Institute of Imaging and Computer Vision,Aachen,North Rhine-Westphalia,Germany

2. T. Freye Aachen Institute of Imaging & Computer Vision

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Institute of Imaging and Computer Vision,Aachen,North Rhine-Westphalia,Germany

3. Yannick Kuhl Aachen Institute of Imaging & Computer Vision

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Institute of Imaging and Computer Vision,Aachen,North Rhine-Westphalia,Germany

4. Volkmar Schulz Aachen Institute of Imaging & Computer Vision

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Institute of Imaging and Computer Vision,Aachen,North Rhine-Westphalia,Germany