摘要： Post-filtering with a Gaussian filter is commonly used to reduce noise in positron emission tomography (PET) images. However, its non-selective smoothing obscures the edges of lesions or organs. We compared the performance of a newly developed anisotropic diffusion filter called “Statistical Transfer with Optimizing Noise and Edge Sensing” (STONES) with that of the Gaussian filter for small lesions on PET images. We selected seven PET/computed tomography (CT) image slices of the lungs from three patients with multiple lung metastases. For each slice, the lesion detection rates by two physicians (A and B) were compared for Gaussian- and STONES-filtered PET images. The maximum standardized uptake (SUVmax) values of the detected lesions were also compared for non-, Gaussian-, and STONES-filtered images. Physician A detected 19 lesions in the Gaussian-filtered images and 23 lesions in the STONES-filtered images, while Physician B detected 14 lesions in the Gaussian-filtered images and 19 lesions in the STONES-filtered images. SUVmax for the STONES-filtered images was significantly higher and closer to that of the non-filtered images compared to those for the Gaussian-filtered images. STONES improved the detection rate and increased SUVmax in comparison with Gaussian filter. Thus, it should be more advantageous for the detection of small lesions with PET.