Beyond mechanics: Maximum-likelihood-driven PET detector alignment calibration | JoVE Visualize

Area of Science:

Medical Imaging Physics
Nuclear Medicine Technology
Computational Imaging

Background:

Positron-emission tomography (PET) relies on accurate quantification, which is often compromised by mechanical inaccuracies in detector assembly.
High-resolution PET systems are particularly sensitive to minor misalignments (<0.5 mm or 0.5°), leading to distorted lines of response and reduced quantitative reliability.

Purpose of the Study:

To develop a calibration strategy for PET systems that eliminates the need for precise physical assembly or movable calibration setups.
To create a data-driven framework for estimating detector alignment parameters directly from PET measurements, enabling robust recalibration across diverse scanner geometries and designs.

Main Methods:

A statistical optimization framework using gradient-based optimization on time-of-flight list-mode coincidence data.
Maximum likelihood estimation conditioned on a known 3D tracer distribution to maximize Poisson likelihood of measured coincidence data.
A software-based recalibration approach supporting complex tracer distributions and generic scanner geometries, eliminating the need for movable parts.

Main Results:

Validated on simulated and real PET systems, achieving alignment accuracy of ~50 µm and 0.15° for point sources and ~120 µm and 0.2° for a tube phantom.
Recovered configurations on real scanners yielded image resolutions of 1.0–1.2 mm, matching design specifications.
Demonstrated accurate alignment with modest data (~5 million coincidences) and robustness to model-sharing constraints.

Conclusions:

The framework bridges hardware imperfections and quantitative fidelity, enabling robust recalibration in mechanically unstable or mobile settings.
Compatible with diverse phantom types and scanner topologies, operating effectively even without precise scanner-phantom alignment.
Offers a practical and scalable solution for modern PET calibration by reducing dependency on strict manufacturing tolerances while preserving accurate line of response positioning.