*2.1. Phantom Experiments*

Experiments were performed on a Discovery D710 PET-CT scanner (GE; Milwaukee, WI, USA) and on a Biograph mCT (Siemens; Knoxville, TN, USA). The Abdo-ManTM Phantom [21] was used as a test object throughout. This is a 3D-printed phantom of the abdomen, with a fillable liver compartment and fillable spherical inserts. The phantom was specifically designed to allow a SIRT-like radioactivity distribution to be generated within an object that mimics the lower-torso in terms of imaging characteristics (i.e., shape and density). Five inserts were placed within the liver region of the phantom, four of which were hollow fillable spheres of different diameter (10, 20, 30, 40 mm). The fifth insert was a 40 mm diameter sphere containing a 25 mm diameter solid inner sphere; this insert mimics the distribution of microspheres in the neovascular rim of a tumour that has a necrotic or poorly perfused core. Two of the spheres (the 40 mm with solid core, and the 20 mm diameter) were placed close to the superior end of the liver compartment. The 30 and 40 mm diameter spheres were close to the centre (in the superior-inferior direction) and the 10 mm diameter sphere was located near the inferior end of the compartment. All spheres were at least 10 mm from the boundaries of the liver compartment.

The phantom was filled with 90Y on four different occasions for each scanner. Four different total activities were used for each scanner, in the range 0.7–3.2 GBq. The spheres were filled with radioactivity concentrations that were greater than the surrounding liver compartment. Two sphere-to-liver concentration ratios were tested, 4:1 and 8:1. The total activity, activity concentrations and concentration ratios were accurately measured during phantom filling, through use of accurate scales and dose calibrators with calibration factors for 90Y that are traceable to the national standard.

The phantom was placed on the QUASARTM respiratory motion platform (Modus QA; London, ON, Canada), which was set to be either stationary or to move according to a typical respiratory waveform. The platform translated the phantom axially to simulate the cranio-caudal motion of the liver during respiration. The maximum displacement from the central position was set to be ±10 or ±15 mm. Data were hence acquired with

maximum displacements between inhalation and exhalation of 0, 20 and 30 mm. The respiratory motion system includes a platform that moves vertically in-time with the axial displacement of the phantom. This platform allows attachment of external devices used for monitoring respiratory motion, from which respiratory gating signals are obtained. For data acquired on the Discovery D710, the Real-time Position Management™ (RPM) Respiratory Gating system (Varian Medical Systems; Palo Alto, CA, USA) was used to track the position of a marker that was placed on the chest-wall platform. The system uses an infrared video camera for this purpose. For acquisitions on the Biograph mCT, gating signals were provided by a pressure belt on the same platform (Anzai Medical Corp.; Tokyo, Japan).

Following a helical CT scan, PET data were acquired for 15 min at a single bed position which included all spherical inserts. This acquisition protocol matches our local acquisition protocol for post-SIRT verification imaging [25]. Additional PET datasets were acquired with the phantom in motion, with motion amplitudes of 10 mm and 15 mm. For acquisitions with 10 mm motion amplitude the duration was increased, when possible, to 45 min and the data processed into 3 × 15 min images. This provided additional data to increase the power of the study. The process was repeated at four phantom activities on both PET-CT scanners. The total number of 15 min duration sets of raw data was 32. Data were reconstructed to provide attenuation and scatter corrected PET images. Different image reconstruction algorithms were tested for the Discovery 710, one based on suggestions from the QUEST study [17] as optimal for quantitative 90Y imaging, alongside a Bayesian penalized likelihood reconstruction. For the Discovery 710, the QUEST reconstruction was the manufacturer's 3D OSEM reconstruction (VPFX) including resolution recovery (SharpIR) and TOF data, with two iterations of 24 subsets, without any z-axis filtration and using a 256 × 256 matrix size (giving voxels of 2.7 × 2.7 × 3.3 mm3). For the BPL (Q.Clear) reconstruction, a beta value of 1000 was used [25]. For the Biograph mCT, a local protocol was followed consisting of the manufacturer's 3D OSEM reconstruction (1 iteration, 21 subsets) including resolution recovery and time of flight (without post filtering). For the cases where motion was present, images were reconstructed with and without quiescent period respiratory gating (QPG) as implemented by the manufacturers, retaining approximately 50% of the acquired data within the quiescent phase [26]. For the Discovery 710, the gating is phase-based (Q.Static™), while for the Biograph mCT it is amplitude-based (HD•Chest™).
