Using software developed in-house that runs on MATLAB (MathWorks, MA, USA), regions of interest (ROI) were automatically drawn to include the spheres and background of the phantom image according to NEMA NU2-2007 criteria. To optimize the total-body 18F-FDG PET/CT scanning protocol, the background variability (BV) and CR were considered the measures of image noise and quantitation accuracy, respectively. BV was defined by the standard deviation (SD) of 12 ROIs located at five slices (a total of 60 background ROIs of each size)—one central slice crossing the sphere center and four close slices at ± 1 cm and ± 2 cm—divided by the mean activity of these 60 background ROIs. CR was defined as follows [39 ]: CR=MH/MB-1CH/CB-1×100%, where M is the count from the PET image and C is the activity in hot spheres and background ROIs (subscripts H and B, respectively). Notably, the contrast-to-noise ratio (CNR), which is considered a key measure of PET imaging performance, is defined as CR divided by BV. First, we regrouped the subsets based on the BV value, where the upper threshold of BV was 15% [40 ]. We then calculated the CR in each filtered condition and produced the CNR that provides a quantitative measure of balance for both BV and RC with the same weight ratio [41 (link), 42 ]. A subjective method of assessment—namely the 5-point Likert scale—was also used to evaluate overall impression of image quality and image noise. All of the center slice images were independently rated by two nuclear radiologists (a senior radiologist with > 5-year experience and a junior radiologist with 1 year of experience reading PET/CT scans) in blinded. The 5-point Likert scale of overall image quality comprises five categories: (1) poor, (2) barely diagnostic, (3) clinically acceptable, (4) superior to the regular quality of daily practice, and (5) excellent. The relationship plot between the CNR value and 5-point Likert scale combines objective and subjective assessment for optimal protocol recommendation. Specifically, the protocol that generates images with high quality (e.g., a score greater than 4- on the 5-point Likert scale) and simpler acquisition settings (e.g., a shorter scan duration and lesser iterations) was preferred in this study.
To facilitate the localization of small lesions, our results—such as BV, CR, and CNR—are mainly demonstrated for hot-spheres with a diameter of 10 mm.
Free full text: Click here