Prognostic Significance of Apparent Diffusion Coefficient in Hepatocellular Carcinoma Patients treated with Stereotactic Ablative Radiotherapy

The role of diffusion-weighted magnetic resonance imaging (DW MRI) in assessing durable tumor control for patients with hepatocellular carcinoma (HCC) treated with stereotactic ablative radiotherapy (SABR) was not defined. This retrospective study included 34 HCC patients with 45 lesions who had DW MRI data at baseline and within 6 months post-SABR. On the first post-SABR MRI, 13 lesions (28.9%) had a complete response (CR), 12 (26.7%) had a partial response (PR), 17 (37.8%) had stable disease, and 3 (6.7%) had progressive disease by modified Response Evaluation Criteria in Solid Tumors (mRECIST). On subsequent imaging, the response rate improved from 55.6% to 75.6%. The apparent diffusion coefficients (ADCs) (mean ± standard deviation) pre- and post-SABR were 1.43 ± 0.28 and 1.72 ± 0.34 (×10−3 mm2/s), respectively (p < 0.001). An ADC change ≥25% (DW[+]) was identified as a predictor of favorable in-field control (IFC) (1-year IFC, 93.3% vs. 50.0% for DW[−], p = 0.004), but an mRECIST-based positive response (CR and PR) at the first MRI was not (p = 0.130). In conclusion, ADC change on early MRI is closely related to IFC in HCCs treated with SABR. Standardization of the DW MRI protocol, as well as prospective validation studies, are warranted.

The intraclass correlation coefficient for ADC values pre-and post-SABR were 0.94 and 0.91 respectively, indicating good interobserver agreement; the average ADC value obtained by the observers was used for subsequent analyses. ADC values were measured in all 45 tumors before SABR and in 42 tumors afterwards (the values of 3 lesions that completely resolved on the first post-SABR DW MRI were not measured). ADC values increased significantly after SABR (mean ± standard deviation, pre-SABR: 1.43 ± 0.28 × 10 −3 mm 2 /s vs. post-SABR:  On ROC analysis, the optimal cut-off value of the ADC increment percentage for predicting freedom from in-field failure was 25% (sensitivity = 57.6% and specificity = 88.9%). Accordingly, DW response was defined as an ADC change ≥25% post-SABR; the 3 completely resolved lesions were assumed to have DW responses on IFC analysis.
In-field control and prognostic factors. The median time to in-field failure for all lesions was not reached; the 1-year IFC rate was 72.9%. Factors associated with improved IFC are shown in Table 2; only tumor size and DW response were identified as predictive of IFC on multivariable analysis (p = 0.006 and 0.037, respectively). The median times to in-field failure were not reached for DW responding lesions and 9.6 months for DW non-responding lesions (1-year IFC, 93.3% vs. 50.0%) (p = 0.004) (Fig. 3A). The first MRI response did not significantly correlate with IFC; the 1-year IFCs were 78.0% and 71.4% for mRECIST responding and non-responding lesions, respectively (p = 0.130) (Fig. 3B).
On first MRI of SD lesions, 9 had a DW response (DW[+] SD) and 8 did not (DW[−] SD); the respective 1-year IFCs were 100% and 65.6% (p = 0.127). The IFCs of DW(+) SD lesions were not significantly different from those of mRECIST responding lesions on first MRI (p = 0.327), whereas the IFC of DW(−) SD lesions tended to be inferior to that of mRECIST responding lesions (p = 0.060).
Explant pathology and ADC correlation. Five patients with 5 lesions were censored owing to local intervention without PD (transarterial chemoembolization [TACE], n = 1; radiofrequency ablation, n = 1; and liver transplantation, n = 3). The radiologic, ADC, and pathologic responses of the liver explants are detailed on Table 3. In contrast to 1 lesion with no obvious ADC change and poor pathologic response, favorable outcomes (near-total necrosis) were achieved in 2 lesions with high ADC changes (35% and 70%) post-SABR.

Figure 1.
Early ADC change before HCC lesion reduction obtained at baseline and at different times post-SABR. A 71-year-old man with hepatitis B virus-related liver cirrhosis and a 2.8 cm HCC in segment 7. At baseline MRI, the lesion shows strong enhancement in the arterial phase (A-phase) and washout in the portal venous phase (PV-phase), with moderate hyperintensity on DWI (b = 500) and an ADC of 1.37 × 10 −3 mm 2 /s. At the 2-month follow-up, the lesion was stable in size by mRECIST. Lower hyperintensity at DWI was noted with a corresponding ADC of 2.00 × 10 −3 mm 2 /s (a 46% increase). Geographic parenchymal hyperenhancement was noted in both the A-and PV-phases, consistent with focal liver reaction to SABR. At the 5-month follow-up, the lesion had a lower enhancement size (1.9 cm), indicative of partial response, with resolution of geographic parenchyma hyperenhancement. At the 8-month follow-up, no enhancement was observed, indicating complete response. Continuous volume loss of the overlying liver parenchyma was observed at 5 and 8 months. (2019) 9:14157 | https://doi.org/10.1038/s41598-019-50503-7 www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
Identifying early response to therapy allows for individualized treatment plans and potentially improves overall prognosis. Our results suggest that ADC change on early MRI is closely related to durable tumor control in HCCs treated with SABR.
Current enhancement-based criteria are highly predictive of HCC patient outcomes after ablation or embolization 7 . Nevertheless, the predictive values of these criteria have not been validated in the context of radiotherapy. In fact, the unique post-SABR HCC imaging features may preclude the interpretation of response based on current criteria 15,16 . A recent study by Mendiratta-Lala et al. focusing on MRI followed 67 HCCs following SABR; 58% had persistent arterial hyperenhancement and 54% had a washout appearance by 3-6 months 15 . These features disappeared over time without detectable tumor progression, resulting in alternating response rates (3-6 months: 25%; and 12 months: 70%). They concluded persistent arterial hyperenhancement is common and does not necessarily indicate viable tumor. Other SABR studies produced similar findings 16,17 . A retrospective study by Mannina et al. demonstrated poor concordance between pathologic response with available radiologic grading criteria 18 . Assessment based on mRECIST had a poor kappa coefficient of 0.224. In our study, the response rate on first MRI was 55.6%, which improved to 75.6% at the most recent follow-up. More than half of the lesions initially deemed to be SD had absent or declining arterial enhancement over time. Initial treatment response by mRECIST was not significantly correlated with long-term tumor control (p = 0.130). Collectively, these findings  www.nature.com/scientificreports www.nature.com/scientificreports/ stressed that traditional response assessment criteria designed for HCC after locoregional therapy may not be useful post-SABR, particularly in early phases. The underestimation of the true effect of SABR may result in improper salvage treatment in clinical practice DW MRI is a promising technique for early assessment of treatment response. The vast majority of HCC data pertain to TACE 19,20 or transarterial radioembolization [21][22][23] . While anatomic changes post-locoregional therapies are noted within 1-3 months, a significant increase in ADC can be detected within several hours to 2 weeks post-therapy [19][20][21][22][23] . A pilot study by Eccles et al. evaluated ADC values in 4 HCCs, 2 cholangiocarcinomas, and 5 liver metastases post-SABR 14 and found significant changes as early as 1 week into radiotherapy. Early ADC response was correlated with a higher dose and sustained tumor response, whereas no significant change in tumor size as assessed on T2 MRI at the same ADC time points were observed. Furthermore, Yu et al. demonstrated the added value of DW MRI for radiotherapy response prediction in 48 HCC patients treated with either hypofractionated radiotherapy (the majority) or SABR who had follow-up MRI at 3-5 months 13 . They found that mRECIST was a significantly better predictor of local progression-free survival (LPFS) than RECIST and an ADC change ≥20%. The latter 2 parameters combined predicted LPFS with comparable performance to the mRECIST. Another recent study that investigated imaging changes observed on MRI before and within 3 months of radiotherapy (3.5-5 Gy for 10 fractions) demonstrated that the ADC value increased in both responding and non-responding lesions with no significant difference (46.7% vs. 21.9%, p = 0.220) 24 .
In our study, we found that an ADC change of ≥25% within 6 months post-SABR was an independent predictor of sustained HCC tumor control, while the mRECIST was not. Our findings suggest a promising role for ADC in predicting early response in HCC patients treated with SABR. Different evaluation time points and methodologies may account, at least partly, for the divergence of our results from those of others.
In contrast to other publications on liver cancer radiotherapy response 13,14 , a relatively low b value of 500 s/ mm 2 was adopted in this study. Of note, the diffusion contrast and signal-to-noise ratio (SNR) should be simultaneously considered in DW MRI with different b values 25,26 . The better diffusion contrast in higher b-value images, and the more substantially reduced SNR are noted due to the increased MR signal loss. Several studies [26][27][28] have reported that high b values (>500 s/mm 2 ) are preferable in order to minimize perfusion effects, and a b value ranging between 500 and 800 s/mm 2 is recommended for assessment of early response of HCC to nonsurgical local treatment. Because of the retrospective nature of the study, we did not have ADC data using a variety of b values. It was believed that the different b values may yield different ADC values, but the tendency of an ADC change ≥25% to indicate SABR in-field tumor control would not be changed.
Given the lack of universal standards, response assessment based solely on ADC change is evolving and requires further validation. DW MRI should be considered a routine adjunct modality to assess HCC after SABR with current enhancement-based criteria 29 . In our study, SD lesions were subdivided into DW(+) SD or DW(−) SD groups based on ADC response; with similar IFC rates following SABR, DW(+) SD lesions could receive   www.nature.com/scientificreports www.nature.com/scientificreports/ sustained follow-up in line with mRECIST responding lesions. Early salvage treatment may be considered for DW(−) SD lesions because of their inferior IFC, as revealed in our study.
Our study had several limitations. First, the single-center, retrospective design and small sample size may have introduced selection bias. Second, the ADC values were not acquired at the same time point post-SABR; it may introduce bias in the judgement of ADC changes. Serial changes in ADC pre-and post-SABR should be investigated to determine the optimal protocol of DW MRI and accurately assess therapy response. Given most ADC values were obtained within 1-3 months post-SABR in this study, we believe the impact of different time intervals is relatively small. Third, there was no pathologic correlation between all treated lesions and their corresponding ADC change, although such validation by biopsy or surgery is not feasible for obvious ethical or technical reasons. Fourth, inter-observer difference for ADC measurement is an inevitable issue in clinical practice. Radiologists may subjectively choose different levels of slice and delineate diverse areas of ROI, even though the ROI measurement has been defined in this study. In order to minimize the variances, the final ADC value was the average of 3 ROIs obtained from 3 different slices, making our ADC change threshold more reliable.
In conclusion, we demonstrated that an ADC change ≥25% within 6 months is an independent predictor of sustained tumor control in HCC patients treated with SABR. Assessments based on mRECIST should be interpreted with caution, especially in the early phase post-SABR. Standardization of the appropriate DW MRI protocol and ADC acquisition parameters, as well as validation in prospective, large, multicenter trials, are warranted.

Methods
Data source. All HCC patients treated with SABR at our hospital between December 2007 and March 2018 were reviewed. The study eligibility criteria were (1) patients with available DW MRI sequences at baseline, and (2) patients with at least 1 DW MRI performed within 6 months post-SABR completion. HCC diagnosis was histologic or based on radiologic findings. Our institutional review board approved the study and waived the informed consent requirement owing to the investigation's retrospective nature.

MRI protocols and ADC measurement. MRI was performed on a 1.5-T MR system (Signa HDx, GE
Healthcare) with an 8-channel body phased-array coil. The protocol of non-contrast and dynamic MR sequences have been detailed previously 30 . MRI protocol for liver tumors included non-fat suppressed axial and coronal single-shot fast spin-echo T2-weighted imaging (WI), axial fat suppressed fast spin echo T2WI, in-and out-of-phase, diffusion-weighted imaging and dynamic contrast enhanced T1WI with fat subtraction. A real time bolus-triggered technique using gadopentetate dimeglumine (Magnevist, Bayer HealthCare) was used to acquire the arterial (20-35 s), portal venous (70 s), and equilibrium phases (3 min).
DW MRI was performed in the axial plane with simultaneous use of respiratory triggering technique. DW images were obtained during dynamic MRI scanning; the TR was matched in each patient to the length of the respiratory cycle before gadoxetic acid enhancement. The scanning parameters were b values of 0 and 500 s/mm 2 ; matrix size, 128 × 128; acceleration factor of SENSE, 4.0; field of view, 42 × 42 cm; number of excitations, 8; slice thickness, 6 mm; slice gap, 2 mm; and axial slices, 25. An ADC map using a monoexponential diffusion model was automatically generated.
Two radiologists (Wei-Chou Chang and Chih-Weim Hsiang) who were aware of the treated lesion but blinded to outcomes measured the ADC values. The DW MRI data were transferred to a separate computer-based workstation (GE Healthcare Systems) specifically designed for post-processing work, and the ADC measurement was calculated using a dedicated software tool (FuncTool, Advantage Workstation 4.3_07, GE Healthcare Medical Systems). Region of interests (ROIs) were delineated over the entire area of the treated lesion on the ADC image and positioned at identical or comparable slice positions on a T2-weighted image (the reference sequence). Each ADC measurement included 3 ROIs: the slice with the largest tumor diameter (the reference slice) and 2 other central slices between both ends and the reference slice. The final ADC value of each treated lesion was the average of those obtained from the 3 ROIs. If the inter-observer difference was larger than 15%, a third ADC measurement was conducted and the result was determined by consensus. SABR and follow-up. Treatment planning, dose prescription and radiation delivery methods have been detailed previously 31 . The SABR protocol is for both recurrent and newly-diagnosed HCC. In brief, patients were immobilized with a customized vacuum cushion during simulation and treatment. For most patients, fiducial markers in conjunction with a Synchrony respiratory motion-tracking system (Accuray Inc.) were employed to manage respiratory motion on a CyberKnife (Accuray Inc.). After June 2017, patients were treated with Versa HD (Elekta AB), using a controlled breath-hold technique or 4-dimensional computed tomography (CT) with forced abdominal compression to manage respiratory motion. The prescribed dose depended on the dose volume constraints of critical organs used in treatment planning 31 . The median prescribed does was 45 Gy (range, 30-60 Gy) in 4-6 fractions in 4-6 consecutive working days. Because of non-uniform fractionation, the dose regimens were converted to equivalent doses of 2 Gy per fraction with the assumption that the tumor α/β value was 10 Gy 32 .
Patients were assessed by clinical examination, blood work, and liver triphasic CT and/or MRI 1-3 months post-SABR completion and every 3-4 months thereafter. SABR response was assessed per tumor according to the mRECIST 5 . In-field failure was defined as presence of progressive disease (PD) or new enhancement within the planning target volume (PTV) or at its margins, which was defined as 1.5 cm from PTV, an area usually received high radiation dose. Statistical analysis. All analyses were performed on SPSS version 17 software (SPSS Inc.). Interobserver ADC value variability was evaluated using the intraclass correlation coefficient. The difference in ADC values pre-and post-SABR was evaluated using the Wilcoxon signed-rank test. The Mann-Whitney U-test was used to