Optimization of the Non-Local Means Algorithm for Breast Diffusion-Weighted Magnetic Resonance Imaging Using a 3D-Printed Breast-Mimicking Phantom

Diffusion-weighted magnetic resonance (DWMR) images were acquired using a custom-designed, 3D-printed breast-mimicking phantom. The smoothing factor of the non-local means (NLM) algorithm was then optimized for noise reduction. Phantoms were fabricated using polylactic acid, polyethylene terephthalate, and various concentrations of polyvinylpyrrolidone. DWMR images were obtained across b-values ranging from zero to 2000 s/mm2. Based on image contrast, the NLM algorithm was applied to the b = 1000 s/mm2 image, testing smoothing factors from 0.001 to 0.150. The NLM algorithm’s performance was quantitatively evaluated using a single DWMR image acquired from this custom phantom. At the optimized smoothing factor, the signal-to-noise ratio (SNR) improved from 96.87 ± 3.42 to 215.81 ± 4.18, and the contrast-to-noise ratio (CNR) from 43.63 ± 2.97 to 131.98 ± 3.56, representing 2.22-fold and 3.02-fold enhancements, respectively. No formal statistical tests were conducted as the analysis was based on a single acquisition. The optimized NLM algorithm also outperformed conventional denoising methods—median, Wiener, and total variation—in both noise suppression and contrast preservation. These findings suggest that the NLM algorithm with optimized parameters is likely to be more effective than existing approaches for enhancing breast DWMR image quality. However, further validation using in vivo patient datasets is essential to confirm its diagnostic utility and clinical generalizability because of the absence of tissue heterogeneity, motion, and physiological noise in the phantom environment.