This study aims to enhance positron emission tomography (PET) imaging systems by developing a continuous depth-of-interaction (DOI) measurement technique using a single-ended readout. Our primary focus is on reducing the number of readout channels in the scintillation detectors while maintaining accurate DOI estimations, using a high-pass filter-based signal multiplexing technique combined with artificial neural networks (ANNs).
Approach: Instead of reading out all 64 signals from an 8×8 silicon photomultiplier array for DOI estimation, the proposed method technique reduces the signals into just four channels by applying high-pass filters with different time constants. To recover the original signal amplitudes, an ANN is used to demultiplex the multiplexed signals. Specifically, the ANN processes the sampled waveforms of these four multiplexed signals and estimates the energy information of the original 8×8 SiPM channels. In this study, two DOI estimation strategies were explored for a continuous DOI PET detector utilizing triangular teeth-shaped reflectors: a 'single-step estimation' method directly estimating DOI from multiplexed signals, and a 'two-stage cascade estimation' method that first demultiplexes the signals and then estimates DOI. The performances of proposed strategies were validated using data irradiated at five steps (2 mm, 6 mm, 10 mm, 14 mm, and 18 mm).
Results: The signal amplitude of row/column summed signals, which were recovered using the proposed ANN-based demultiplexing, showed strong correlation with ground truth (e.g., R2=0.98 for 125 MHz digitizer sampling rate). Moreover, both the single-step and two-stage estimation methods achieved high accuracy in DOI estimation, with an average DOI estimation accuracy of 72.9% and 74.0% at 125 MHz sampling rate when considering an error range of ± 1 DOI position. 
Significance: This novel signal multiplexing technique significantly reduces the number of required readout channels, making continuous DOI PET more cost-effective. 
.
Keywords: Positron emission tomography; depth-of-interaction; machine learning; scintillation detector; signal multiplexing.
© 2025 Institute of Physics and Engineering in Medicine. All rights, including for text and data mining, AI training, and similar technologies, are reserved.