Optimizing LM-DRAMA parameters and non-local means filtering to improve small-lesion detectability in SiPM-based TOF breast PET.

This study aimed to optimize image reconstruction parameters for a dedicated time-of-flight (TOF) breast positron emission tomography (PET) system equipped with silicon photomultipliers (SiPMs) that maximize lesion detectability while minimizing image noise. A cylindrical phantom containing four hot spheres (3-10 mm diameter) was scanned at sphere-to-background ratios of 4:1, 6:1, and 8:1. All data were reconstructed using a 3D list-mode dynamic row-action maximum likelihood algorithm with β values of 10-200, followed by non-local means (NLM) filtering at intensities of 0.5-2.0 or no filtering. Image quality was evaluated using background coefficient of variation (COV), contrast recovery coefficient (CRC), and detectability index (DI) for the 3 mm sphere. As β increased, CRC and DI improved, particularly for smaller spheres and higher SBRs; however, background noise also rose. Applying the NLM filter reduced COV, especially when increasing the filter intensity from 0.5 to 1.0, although noise reduction gains plateaued at intensities above 1.0. Optimal trade-offs in lesion detectability and noise were observed at moderate β (50-100) with NLM intensities of 1.0-1.5, yielding higher CRC and DI without excessive background noise or blurring effects. A balanced approach to β and NLM filtering substantially enhances small-lesion visibility in SiPM-based TOF-dedicated breast PET imaging. These findings offer a practical framework for parameter selection, supporting better lesion detectability and advancing breast cancer diagnostics through more sensitive PET protocols.