The records of patients who were treated with ⁹⁰Y-TARE for HCC at our institute from April 2011 until March 2016 were reviewed. Ninety-three consecutive patients were identified. ⁹⁰Y-TARE was performed in patients with HCC unsuited to curative treatments, e.g., surgical resection, local ablative therapy (radiofrequency ablation or ethanol injection), and liver transplantation. Inclusion criteria included: patient age of ≥ 20 years, surgically unresectable HCC, single or multifocal measurable HCC without extra-hepatic metastasis, Eastern Cooperative Oncology Group performance status of 0 or 1, Child-Pugh class A or B, and adequate hematological (white blood cell count ≥ 2000/µL, hemoglobin level ≥ 10 g/dL, and platelet count ≥ 10⁵/µL) and renal function (serum creatinine ≤ 1.5 mg/dL). The bilirubin cutoff was less than 2.0 mg/dL, but 3 patients with a bilirubin level exceeding the cutoff value were included, because the tumors were localized, minimizing normal hepatic parenchymal damage.

Patients were excluded from the study if they 1) had insufficient CT data or a period between imaging and the procedure exceeding 8 weeks (n = 22), 2) did not undergo a second follow-up imaging examination as they were lost to follow-up or had expired (n = 9), 3) were not treated with resin-based microspheres (n = 15), or 4) had simultaneously undergone other treatments (n = 3) (Fig. 1).

The study population comprised 44 patients (34 men and 10 women; mean age: 60 ± 12 years; age range: 33–91 years) who had undergone dynamic liver CT before and 4 weeks post-TARE (Fig. 1).

All patients underwent diagnostic angiography for evaluating the suitability of the treatment and the dose calculation. Celiac, common hepatic arteriography, and cone-beam CT arteriography were performed to evaluate the tumor feeding artery distribution and to determine the target artery of radioembolization. Twelve patients underwent coil embolization of the cystic artery (n = 10), the accessory left gastric artery (n = 1), and both the right gastric and cystic arteries (n = 1), to avoid non-target embolization of ⁹⁰Y. Depending upon the extent of tumor burden, patients were treated with either a segmental (n = 12), lobar (n = 20), or lobar and segmental (n = 12) approach. Hepato-pulmonary shunt was evaluated by injection of technetium-99m-labeled macroaggregated albumin at the target artery. We measured the liver and the tumor volume of each patient and the radiation dose was then determined based on the planning angiogram and hepato-pulmonary shuntogram, using a partition model (10), after calculating the proper tumor dose and the safety dose to the lung and normal liver. The target-absorbed radiation dose of the normal liver and lung was not to exceed 70 Gy and 25 Gy, respectively. ⁹⁰Y was prepared in the Nuclear Medicine Department of our institution, according to the manufacturer's preparation guide. ⁹⁰Y-resin microspheres (SIR-Spheres; Sirtex Medical, North Sydney, Australia) were infused at the feeding artery of the tumor as selectively as possible within 2 weeks after planning angiography by using a 3-French microcatheter (Terumo, Tokyo, Japan).

Four-phase dynamic CT images were obtained before and 4 weeks post-TARE using the same imaging protocol with 16-, 64-, or 128-channel multidetector CT scanners. The CT scanners used pre- and post-procedure were not necessarily the same. First, a pre-contrast CT scan was performed. CT was performed after intravenous administration of 1.5–2.0 mL/kg of iodinated contrast media (300 mg I/mL) for a fixed injection duration of 30 seconds (rate, 2–4 mL/s), followed by a 20-mL saline chaser bolus injection. Injection velocity and duration were adjusted for the individual's vessel condition. Using a bolus-tracking technique, the late arterial phase was performed 18 seconds after the attenuation value reached 100 HU at the abdominal aorta. The portal venous phase and delayed phase were obtained with a scan delay of 30 seconds and 150 seconds after the end of the previous phase, respectively (Table 1).

After radioembolization, tumor progression was defined as the emergence of enhanced arterial phase areas, followed by washout in the portal phase, or when nodules had increased in size during the follow-up. When such patterns were observed next to the post-treatment zone, they were regarded as local tumor progression (LTP). LTP was reported for the largest tumor in each patient. LTP was evaluated at each follow-up CT scan, but early treatment response was evaluated 4 weeks post-TARE, based on the RECIST, mRECIST, and newly proposed quantitative criteria.

Two board-certified abdominal radiologists (with 20 and 5 years of experience in abdominal CT, respectively) retrospectively and qualitatively analyzed the pre-TARE CT images and achieved consensus on the following: the largest diameter of the tumor, the largest diameter of the arterial enhancing portion (DAE), number of tumors, bilobar involvement, portal vein invasion (PVI), and hepatic vein invasion.

Quantitative analysis was performed using CT images before and 4 weeks post-TARE. Contrast-enhanced dynamic CT studies were retrieved from the institutional archive and loaded onto a standard workstation in digital imaging and communications in medicine format. We subtracted the pre-contrast images from the delayed-phase images to create the delayed-phase enhancement map (11). The delayed-enhancing area was defined as an area with an enhancement greater than the mean value + 2 times the standard deviation (SD) of the normal liver parenchymal enhancement. The necrotic area was defined as an area with an enhancement of 20 HU or less in the delayed phase. The area below the threshold value of the delayed enhancement was removed from the enhancement map. A freehand-drawn region of interest (ROI) was placed along the boundary of the tumor-containing hepatic segment in the cross-section showing the largest diameter of the largest tumor. The ROI was drawn for the whole tumor-containing segment, not for the tumor boundary, to reflect the peritumoral change after treatment and to increase reproducibility. The DN in the ROI was measured automatedly (Fig. 2) on images before and post-TARE, and differences were calculated. Specifically, the increase in the delayed-enhancing area (DI), the increase in the necrotic area (NI), and the increase in the DN (DNI) in the tumor-containing segment were calculated. The percentage increase in these areas (percentage DI [pDI], percentage NI [pNI], percentage DNI [pD + N]) was calculated, as these are more generalizable to clinical practice: percentage increase value = (post-TARE value - pre-TARE value) / pre-TARE value. All processes were performed by an abdominal radiologist with 5-years' experience, using in-house software written in Matlab (version 2017a, MathWorks Inc., Natick, MA, USA) (12). Quantitative analysis is described in full in the Supplementary Materials 1 (in the online-only Data Supplement). One additional radiologist performed measurement of the pD + N, pDI, and pNI to test interreader reproducibility.

The descriptive variables are presented as mean (± SD) values or the median and interquartile range (IQR) values for non-normally distributed data. Local progression-free survival (PFS) was defined as the interval between TARE and death, last follow-up imaging, the most recent follow-up visit, or the date of LTP at post-therapeutic imaging. Patients who had undergone liver transplantation were censored at the date of transplantation. Patients with a lesion that had been additionally treated during the follow-up period, based on clinical considerations other than LTP imaging evidence, were censored. Differences in PFS between patient groups, based on Kaplan-Meier survival curves, were assessed with the log-rank test. Each continuous variable was dichotomized to determine the optimal cutoff value. This procedure considered all possible continuous variables values as potential cutoff points. The cutoff corresponding to the largest difference between the 2 groups was determined based on the log-rank statistic (1314). Univariate and multivariate Cox regression analyses were used for identifying PFS prognostic factors. Factors were entered into the multivariate Cox regression model using backward stepwise selection with the Akaike information criterion, and hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated. The interreader reproducibility was evaluated in terms of the intraclass correlation coefficient (ICC). An ICC value higher than 0.75 indicates good reproducibility (1516). For all tests, statistical significance was set at p < 0.05. R (version 3.3.1, R Foundation for Statistical Computing, Vienna, Austria) was used for all statistical analysis.

Patient characteristics are described in Table 2.

The first follow-up CT imaging was performed a median 23 days (IQR, 21–28 days) post-TARE. The mean diameters of the tumors were 8.36 ± 3.07 cm and 7.96 ± 3.37 cm before and at the first post-TARE follow-up, respectively, and the median diameter decrease was 2.8% (IQR −3.34% to 15.7%). The mean DAE in tumors were 7.47 ± 3.26 and 6.58 ± 3.50 cm before and at the first follow-up, respectively, and the median decrease in DAE was 10.2% (IQR 0.0–26.6%). The median pD + N was 17.9% (IQR −8.9% to 73.3%), and the median pDI and pNI were 28.7% (IQR −25.4% to 106.9%) and 12.8% (IQR −39.3% to 105.1%), respectively. At first follow-up, 3 (6.8%) and 10 patients (22.7%) showed partial response in RECIST and mRECIST, respectively, and all the other patients showed stable disease.

Of the 44 treated HCCs, estimated probabilities of PFSs were 70.6% and 33.5% for 1-year and 2-year post-TARE, respectively. Nine patients (20.5%) underwent hepatic resection and 1 (2.3%) had a liver transplantation post-TARE, with curative intent.

In univariate Cox regression analysis that included DNI, NI, DI, pD + N, pNI, pDI, the tumor diameter decrease (%), and the DAE decrease (%) as continuous variables, DNI and pD + N showed a statistically significant correlation with PFS (DNI, HR 0.999, p = 0.033; pD + N, HR 0.991, p = 0.043); other variables showed no statistical significance (Table 3).

The optimal cutoff points for predicting PFS were 35.5%, −7.1%, 1.4%, and 17.3% for pD + N, pNI, diameter decrease, and DAE decrease, respectively, and each variable was dichotomized into 2 groups: those with values ≥ the optimal cutoff point, and those with values < the optimal cutoff point. In Kaplan-Meier curves, estimated probabilities of PFS were 93.0% at 1-year, and 50.0% at 2-years post-TARE for the higher pD + N group, and 48.9% at 1-year, and 0% at 2-years post-TARE for the lower pD + N group (median PFS, 21.2 months vs. 6.1 months) (Fig. 3). Estimated probabilities of PFS were 81.2% at 1-year, and 54.1% at 2-year post-TARE for the higher pNI group, and 54.2% at 1-year, and 0% at 2-year post-TARE for the lower pNI group (medial PFS, 29.6 months vs. 16.9 months). The PFS of the higher pD + N, higher pNI, and higher DAE decrease groups was significantly greater than that of the lower groups, respectively (pD + N, p = 0.001; pNI, p = 0.016; DAE, p = 0.038; log-rank test).

In univariate Cox regression analysis, pD + N (lower vs. higher, HR 14.8, p = 0.011), pNI (lower vs. higher, HR 4.0, p = 0.025) and pre-TARE tumor size (HR 1.2, p = 0.018) significantly predicted PFS (Table 4, Fig. 4). In multivariate analysis, pD + N (lower vs. higher, HR 17.3, p = 0.009) and pre-TARE tumor size (HR 1.3, p = 0.016) were identified as independent PFS predictors.

Measurement on the pD + N, pDI, and pNI showed good interreader reproducibility (ICC; 0.77, 0.86, and 0.84, respectively).