Decision Tree-Based Demultiplexing for Prism-PET.

Area of Science:

• Medical Imaging
• Nuclear Physics
• Instrumentation

Background:

• Signal multiplexing in Positron Emission Tomography (PET) is crucial for reducing readout channels, minimizing cost, and lowering power consumption.
• Conventional methods like weighted average energy are insufficient for localizing multiplexed events, necessitating advanced demultiplexing techniques.

Purpose of the Study:

• To propose and evaluate a non-parametric decision tree model for accurate signal demultiplexing in prismatoid PET (Prism-PET) detector modules.
• To assess the impact of the proposed demultiplexing method on spatial and depth-of-interaction (DOI) resolution performance.

Main Methods:

• Developed a decision tree model to demultiplex signals from a Prism-PET detector module featuring a 16x16 lutetium yttrium oxyorthosilicate (LYSO) crystal array and 8x8 silicon photomultiplier (SiPM) pixels with 64:16 multiplexed readout.
• Trained 64 individual regression trees for demultiplexing encoded readouts per SiPM pixel.
• Employed Center of Gravity (CoG) and Truncated Center of Gravity (TCoG) methods for crystal identification using demultiplexed data.

Main Results:

• The decision tree model demonstrated accurate signal demultiplexing capabilities.
• Performance metrics including flood histogram, energy resolution, and DOI resolution were evaluated comparing multiplexed and non-multiplexed readouts.
• The proposed method successfully maintained the high spatial and DOI resolution of the Prism-PET detector module.

Conclusions:

• The non-parametric decision tree model provides an effective solution for demultiplexing signals in multiplexed PET detector modules.
• This approach enables the benefits of signal multiplexing without compromising critical detector performance characteristics like spatial and DOI resolution.