Probabilistic layer identification in a multi-layer fast timing detector for time-of-flight PET using machine learning

This work presents an effective algorithm to identify crystal element positions for the design and operation of practical PET imaging detectors capable of achieving both excellent time resolution required for time-of-flight (ToF) and depth-of-Interaction (DoI) information. The detector unit consists of a dual layer (LYSO:Ce and LSO:Ce,Ca(0.4%)) stack of two 3×3×10 mm³ crystals, 1:1 coupled to SiPM arrays. Features of energy, crossover time metric, and a probability density estimation based unsupervised machine learning approach have been used for identification of in which layer a 511 KeV photon interacts. A global coincidence time resolution of 224 ps and a 91% layer identification accuracy has been achieved.