Improving timing resolution of BGO for TOF-PET: a comparative analysis with and without deep learning

Area of Science:

• Nuclear Instrumentation
• Medical Imaging Physics

Background:

• Bismuth Germanate (BGO) scintillators are gaining interest for Time-of-Flight Positron Emission Tomography (TOF-PET).
• Slower BGO scintillation light degrades Coincidence Time Resolution (CTR) with standard Leading Edge Discrimination (LED).
• Time Walk Correction (TWC) and Convolutional Neural Networks (CNNs) are explored to improve CTR.

Purpose of the Study:

• To compare the CTR performance of LED, TWC, and CNN-based timing estimation methods for BGO scintillators.
• To evaluate the effectiveness of different waveform analysis techniques in optimizing TOF-PET detector signals.

Main Methods:

• Experimental comparison of LED, TWC, and CNN methods using BGO crystals and NUV-HD-MT SiPMs.
• Digitized waveform analysis with high-frequency electronics.
• Quantitative measurement of Coincidence Time Resolution (CTR) in Full Width at Half Maximum (FWHM).

Main Results:

• CNNs achieved 115 ± 2 ps FWHM CTR for smaller crystals, a 26% improvement over LED.
• TWC achieved 129 ± 2 ps FWHM CTR, an 18% improvement over LED for smaller crystals.
• For larger crystals, both TWC and CNNs provided ~15% gain over LED, with CNNs showing better tail suppression.

Conclusions:

• CNNs offer superior CTR performance for BGO scintillators in TOF-PET compared to TWC and LED.
• Simpler methods like TWC may capture essential timing information for longer BGO crystals.
• Future deep learning advancements hold promise for further enhancing timing features in TOF-PET detectors.