Machine Learning-Based Events Classification in Heterostructured Scintillators

Time-of-flight positron emission tomography (TOF PET) faces the challenge of improving time resolution while maintaining high detector efficiency. Heterostructured scintillators, which consist of multiple layers of different scintillating materials with complementary properties, have emerged as a promising solution. These scintillators rely on the energy sharing mechanism, where the incoming 511 keV γ-ray can be fully stopped in the heavy material while the recoil photoelectron can escape and deposit some of its energy in the fast emitter (shared events) [1], [2]. Accurate event classification is crucial to fully leverage this concept. A well-established method is to correlate the amplitude and integrated charge of the pulse and apply a coordinate transformation, but it is not easily scalable to more complex systems [1]. In this study, we use the hierarchical clustering method to distinguish events based on the deposited energy in a BGO & plastic heterostructured scintillator. We compared the two methods by evaluating the CTR of the events coming from the corresponding classes of events. The results were found to be compatible, indicating the effectiveness of the proposed clustering method for event classification in heterostructured scintillators.