Automated Calibration and Quality Scoring for Floodmap-Based Crystal Identification in PET Detectors

Positron emission tomography (PET) is a medical imaging technique in which gamma rays from electron-positron annihilation are detected by scintillator crystals coupled to photodetectors in a grid packing. To reconstruct these gamma ray events, the system must accurately identify which crystal detected each gamma ray by analyzing the flood map—a 2D histogram of energy centroids derived from raw SiPM signals. The algorithms for floodmap generation and crystal ID have several tunable parameters and rely on manual inspection, limiting calibration efficiency and scalability.

In this work, we present a method for automating parameter tuning and quality inspection. We developed an automated procedure to score floodmap quality based on whether the expected number of crystals are arranged on a grid and sufficiently distinct. We then apply Bayesian Optimization to find the set of parameters that best leads to crystal ID. We also calibrated a numerical threshold to automatically flag whether a floodmap successfully achieves accurate ID for all crystals or requires manual correction.

The method was evaluated for 20 PET detectors, optimizing for the best Crystal ID under a Ge68 point source irradiation. The optimized parameters correctly identified all crystals in over 90% of cases (up from 40% using fixed settings), with 100% sensitivity and 95% specificity for the automatic success flag. This approach greatly reduces the need for manual correction and enables scalable, high-throughput calibration of PET systems.