Patients with unresectable HCC treated with [90Y] resin microspheres SIRT in our institution from May 2018 to November 2020 were considered for this retrospective study. Inclusion criteria consisted of contrast-enhanced CT or MRI which was performed 12 weeks prior to SIRT, a targeted lesion long-axis diameter of at least 2 cm, and a follow-up MRI at four months. Only patients receiving Sirtex 90Y-resin microspheres were included, as resin and glass microspheres differ in general kinetics and dose calculation. Individual informed consent was not required, because studies involving a retrospective review, collection, and analysis of patient records do not fall under the scope of the Dutch Act on Medical Scientific Research involving Human Beings (WMO). For privacy, data were stored and analysed anonymously. Patient characteristics, such as age, sex, comorbidities, other risk factors and outcomes, were extracted from the electronic medical records. Overall survival (OS) and progression-free survival (PFS) were noted. Relevant follow-up therapy and staging of Barcelona-Clinic Liver Cancer (BCLC) and Child–Pugh (CP) were noted. Evaluation of treatment response to SIRT was done according to the modified response evaluation criteria (mRECIST) at 4-month MRI [15, 16].
Planning angiography and 99mTc‑MAA SPECT/CT
All patients were subjected to angiography of the upper abdominal vessels to define vascular anatomy and to assess optimal catheter-tip placement [4]. Following angiography, 150 MBq (4 mCi) of [99mTc] MAA (Pulmocis, Curium Pharma, Petten, the Netherlands) was administered. One hour after injection of [99mTc] MAA, lung and liver planar scan and low dose, no contrast-enhanced SPECT/CT acquisitions were performed using a hybrid scanner combining a dual-head gamma camera and a 2-slice SPECT/CT scanner (Symbia T2, Siemens Healthcare, Germany). Images were then reconstructed on a Siemens workstation (SyngoVia VB30, Siemens Healthcare, Germany). The amount of 90Y-microsphere activity needed during treatment phase was determined by the partition model, provided and detailed by the manufacturer (SIR-Sphere®, Sirtex Medical Limited Australia, Sydney, Australia) [4, 15].
SIRT and [90Y] PET/CT
SIRT was performed within two weeks after planning angiography. The planned activity of 90Y-loaded microspheres was injected through a microcatheter at the same position as determined during planning angiography. Within one day after SIRT, patients underwent [90Y] PET/CT scan (Biograph mCT PET/CT, Siemens Healthcare, Erlangen, Germany) with a maximum of two bed positions, and 15-min acquisition per bed position, for treatment verification and post-treatment dosimetry. PET data were reconstructed with Siemens Ultra HD (TrueX and time of flight), using three iterations and 21 subsets with a 400-matrix size and a 9-mm Gaussian (isotropic) filter. Attenuation and scatter correction of PET emission data were achieved by a low-dose CT scan with 120 kV and 35 mAs.
Dosimetry
For pre-treatment planning of injected 90Y-activity, liver and tumour contours were manually delineated on CT images, acquired during planning angiography to be used in the partition model. Pre-treatment contrast-enhanced CT (Siemens SOMATOM Force CT) or gadolinium-enhanced fat-saturated T1-weighted MRI (Siemens Magnetom Skyra MRI) were used for 3D delineation of liver and tumour contours for post-treatment dosimetry. Post-treatment dosimetry contouring was performed in MIM SurePlan (v7.0.4, MIM software, Cleveland, USA). In all three planes and for every three slides, the researcher manually delineated vital liver tissue and tumours. The software then interpolated all contours to create a 3D representation of all contours. These contours were then transformed to contours on post-therapy PET/CT by a MIM SurePlan clinical workflow (“90Y Dose Calculation”) using deformable registration algorithms. The computed contours were then, in some cases, manually translated or rotated to achieve optimum visual fit.
90Y-dose and DVH for each tumour were calculated with the local deposition method (LDM), as previously described [16]. The mean tumour-absorbed dose (in Gy) were extracted from DVH, where area under the DVH (AUDVH) equals tumour-absorbed dose [17]. V100, V150, V200, and V250 were calculated from the DVH, representing the percentage of the tumour volume receiving indicated value of radiation (in Gy). D30, D50, D70, and D90 were computed showing the minimum absorbed dose delivered to those tumour volume percentages.
Excluding small tumours reduced the chance of partial volume effects of dosimetry data in relation to the PET/CT, as a sphere diameter of at least 2 cm with no filtering should give a better reading of activity according to the literature [18]. Depending on the mean tumour-absorbed dose, treated lesions were assigned to a < 120 Gy or ≥ 120 Gy group. Patients who had both a < 120 Gy and a ≥ 120 Gy lesion were added to both groups. For computing OS, time between first (or only) treatment and death was calculated and, patients were not added twice in case of group comparisons. Complete response (CR), partial response (PR), and stable disease (SD) mRECIST results were combined into a disease control (DC) group to be compared to progressive disease (PD).
Statistics
All descriptive statistics are given by numbers with percentiles or the median with its interquartile ranges, unless stated otherwise. Comparisons of tumour-absorbed dose between DC and PD are performed by an unpaired t test with Welch’s correction. Comparisons of mRECIST with mean tumour dose and D- and V-values were compared by Kruskal–Wallis (with Dunn’s multiple comparisons test) or two-way ANOVA. A nonlinear second-order polynomial (quadratic) least squares fit was performed on the DVH of DC and PD groups. Receiver operating characteristic (ROC) analysis was performed to identify the optimal cut-off (defined by the Youden index) of tumour-absorbed dose to predict DC. By averaging the chance of all patients to have DC, binned by intervals of 20 Gy, the tumour control probability (TCP) was computed and related to tumour dose using a linear quadratic model. OS and PFS between-group comparisons were determined with Kaplan–Meier Chi-square log rank Mantel–Cox. Statistical analysis was performed using SPSS version 23.0 software (SPSS Inc., Chicago, IL). p values lower than 0.05 were considered to be significant.