Noise level and similarity evaluations of non-local means algorithm in chest digital tomosynthesis X-ray imaging system: An experimental study

Abstract

As a method for diagnosing chest lesions, a chest digital tomosynthesis (CDT) X-ray imaging system based on projection data using a limited angle has been widely used in the medical field since its development. In these CDT X-ray images, noise reduction methods using software that reduce the accuracy of lesion detection are essential. The aim of this study is to model the non-local means (NLM)-based algorithm, which is known to be very effective in removing noise from X-ray images, and to confirm its applicability in the CDT system. CDT X-ray images using a human phantom were generated employing the filtered back-projection reconstruction method with a projection angle of ±36.15°. For quantitative evaluation, a reference image was obtained using the full radiation dose, and a noisy image was obtained using the minimum radiation dose to increase the amount of noise. The NLM noise reduction algorithm was modeled as a method for obtaining weights based on the similarity of the neighboring set of pixels. Conventional filtering methods were used in the comparison group to analyze the efficiency of the NLM algorithm. As a result, we confirmed that the coefficient of variation of the noise level was improved about 9.74 times compared with the noisy image when the NLM noise reduction algorithm was applied to the CDT X-ray image. When the similarity evaluation parameters were measured, we proved that 11%–53% better values were derived compared with the noisy image in the NLM algorithm. In particular, the performance of the NLM algorithm showed superior results to those obtained with the conventional filtering methods. In conclusion, the applicability of the NLM noise reduction algorithm to CDT X-ray images using limited projection data was demonstrated.