A comparative evaluation of manually optimized and adaptive non-local means approaches in abdominal low-dose CT images

초록

In low-dose computed tomography (LDCT), noise from insufficient photons degrades image quality. This study compares the traditional non-local means (NLM) algorithm, requiring manual optimization, with the automated adaptive non-local means (ANLM) algorithm for LDCT noise reduction. Gaussian noise was added to clinical abdominal CT images to simulate LDCT, and the search window size of the NLM algorithm was determined through manual optimization. To evaluate the optimization and comparative performance, noise reduction and similarity evaluations were conducted. The optimal search window size for the traditional NLM algorithm was 7 & times; 7, derived from the most significant slope change in quantitative evaluation. A comparative evaluation revealed that the ANLM algorithm showed superior noise reduction performance, whereas the traditional NLM algorithm demonstrated the most improvement in similarity metrics. Visual evaluation showed that the ANLM algorithm effectively reduced noise but lost details around edges. The traditional NLM algorithm remains relatively noisy, whereas the signal around the edges is better preserved. Based on this analysis, the ANLM algorithm effectively performs automatic denoising, while the NLM algorithm better preserves edge information through manual optimization. Therefore, each algorithm offers distinct advantages and should be selected based on clinical situations, as this study provides a preliminary quantitative comparison.

키워드