*2.2. Image Analysis*

The images were analysed at a single site using commercial software (Hybrid Viewer v5.1, Hermes Medical Solutions AB, Stockholm, Sweden). The quantitative accuracy of the data was first verified for the stationary phantom acquisitions. For each image, 7 spherical volumes of interest (VOI), each 2 cm in diameter, were placed within the (background) liver compartment. These regions were placed at least 1 cm from the walls of the compartment and the inserts and were expected to be free of partial volume effects. We hence expected the mean radioactivity concentrations to closely match the known activity concentration, as determined from radionuclide dose calibrator measurements during preparation of the phantom.

The effects of respiratory motion on dosimetric measures were assessed by investigating the accuracy of activity quantification for the spherical inserts. Spherical VOIs were placed on each of the inserts, with the VOI manually centred on the given sphere as observed in the PET image. The VOI diameters matched the inner diameters of the spheres. In the case of the sphere with a solid core, two VOIs were analysed: one pertaining to the hot outer shell (a 40 mm spherical VOI but with a central 25-mm diameter sphere excluded), and one 25 mm spherical VOI centred on the cold core. These VOIs were repeated with manual alignment on each image from the study, i.e., for the phantom stationary or moving, with or without motion compensation via respiratory gating. From the hot spheres, an activity concentration volume histogram (ACVH) was extracted alongside the VOI mean

activity concentration (ACmean). From these ACVHs, the dosimetric measure AC80 was extracted for summary analysis. AC80 is defined as the activity concentration threshold that incorporates 80% of voxels within the VOI. Although other thresholds can be used (e.g., AC50 as the median), we narrowed the current investigation to AC80 on the basis that we expect the mean or median dose to a tumour volume to have a weaker correlation with tumour response compared to AC80 [27], but higher thresholds such as AC90 to be more greatly affected by image noise. Results were converted to recovery percentages (i.e., the measured activity concentration as a percentage of the true activity concentration) to allow comparative analysis across the range of activity concentrations studied. The contrast recovery for the cold core was calculated as QC = (1 − Ccore/Cshell) × 100%, where Ccore is the VOI mean value of the cold core, and Cshell the VOI mean of the hot shell. Comparison of the results for the phantom in motion to those found with the phantom stationary, for the different sphere sizes, allowed the impact of respiratory motion and motion correction to be placed into perspective alongside the partial volume effect. Hence, the relative benefit of respiratory motion correction, in terms of impact on dosimetric measures, was assessed.
