A Gaussian filter influenced the texture features on PET images

Purpose: Generally, the standardized uptake value is primarily used for the analyses in FDG PET. Recently, an increasing number of publications have described the application of textural features in PET image analyses. However, the influence of reconstruction parameters on the evaluation of the heterogeneous uptake has not yet been sufficiently elucidated. The purpose of this study was to examine the impact of a Gaussian filter on evaluating the heterogeneous uptake in PET images as assessed by texture analysis using numeric phantoms. Materials and Methods: The numeric phantoms represented a high activity homogeneous model, three heterogeneous models with high, medium and low activity and four heterogeneous models with high and medium activity. The images of the numeric phantoms were 15×15 pixels and filtered by the Gaussian filter using the Prominence Processor (Version 3.1). The full width at half maximum of the Gaussian filter (GF-FWHM) was 6 mm. The heterogeneity was evaluated based on 13 texture features on a co-occurrence matrix and 11 texture features on a size zone matrix. Results: In images without a Gaussian filter, all features on the co-occurrence matrix and six features on the size zone matrix differed between the homogeneous and heterogeneous models. In images with a GF-FWHM of 6 mm, 10 features on the co-occurrence matrix and 5 features on the size zone matrix differed between the homogeneous and heterogeneous models. Three features that reflected the small zones on the size zone matrix were influenced by the Gaussian filter. The influence of the Gaussian filter on the co-occurrence matrix was small. Thus, a Gaussian filter is considered to affect the evaluation of the heterogeneity by the texture analysis. Furthermore, the Gaussian filter had greater effects on the size zone matrix than on the co-occurrence matrix. Conclusions: The results of this study suggested that the evaluation of heterogeneity as assessed by texture analysis was affected by the Gaussian filter, which had a greater influence on the size zone matrix than on the co-occurrence matrix.