Diffusion-weighted and dynamic contrast-enhanced magnetic resonance imaging after radiation therapy for bone metastases in patients with hepatocellular carcinoma

The objectives of this study were to assess changes in apparent diffusion coefficient (ADC) and dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) parameters after radiation therapy (RT) for bone metastases from hepatocellular carcinoma (HCC) and to evaluate their prognostic value. This prospective study was approved by the Institutional Review Board. Fourteen patients with HCC underwent RT (30 Gy in 10 fractions once daily) for bone metastases. The ADC and DCE-MRI parameters and the volume of the target lesions were measured before (baseline) and one month after RT (post-RT). The Wilcoxon signed-rank test was used to compare the parameters between the baseline and post-RT MRI. The parameters were compared between patients with or without disease progression in RT fields using the Mann–Whitney test. Intraclass correlation coefficients were used to evaluate the interobserver agreement. The medians of the ADC, rate constant [kep], and volume fraction of the extravascular extracellular matrix [ve] in the baseline and post-RT MRI were 0.67 (range 0.61–0.72) and 0.75 (range 0.63–1.43) (× 10–3 mm2/s) (P = 0.027), 836.33 (range 301.41–1082.32) and 335.80 (range 21.86–741.87) (× 10–3/min) (P = 0.002), and 161.54 (range 128.38–410.13) and 273.99 (range 181.39–1216.95) (× 10–3) (P = 0.027), respectively. The medians of the percent change in the ADC of post-RT MRI in patients with progressive disease and patients without progressive disease were − 1.35 (range − 6.16 to 6.79) and + 46.71 (range 7.71–112.81) (%) (P = 0.011), respectively. The interobserver agreements for all MRI parameters were excellent (intraclass correlation coefficients > 0.8). In conclusion, the ADC, kep, and ve of bone metastases changed significantly after RT. The percentage change in the ADC was closely related to local tumor progression.


Materials and methods
This prospective study was approved by the Institutional Review Board (Samsung Medical Center, IRB File No. 2018-07-159), registered at cris.nih.go.kr (KCT0004861), and was conducted from February 2019 to July 2020. Written informed consent was obtained from all patients and the study was conducted in accordance with the declaration of Helsinki.
Patients. According to a previous study 27 , the volume transfer constant (K trans ) after RT was expected to decrease by 33.5% (standard deviation 28.9%). Using paired t-tests, the required sample size was calculated to be 10 in order to have a 90% chance of finding an average of 33.5% difference at a significance level of 5% (α = 0.05, β = 0.10) (MedCalc Statistical Software version 19.1.5; MedCalc Software Ltd.). Therefore, assuming a dropout rate of 25%, a total of 14 patients were required.
Patients with HCC who were scheduled to undergo RT for bone metastases were included if they (a) were older than 18 years, (b) had a histopathological or imaging diagnosis of HCC, and (c) had metastases in the thoracic or lumbosacral spine or pelvic bone; patients with metastatic lesions in other locations were not included to minimize unwanted contributions from any potential region-dependent biases. The criteria for considering bone metastases from HCC included newly developed or progressed bone lesions noted during computed tomography (CT) surveillance or histopathological confirmations by bone biopsies, where available. The exclusion criteria were as follows: (a) another primary malignancy, (b) history of surgery, RT, or metallic instrumentation at the above metastatic sites; (c) contraindication to gadolinium-based contrast agents or MRI examinations, or (d) pregnant or nursing female patients. We collected clinical information, including sex, age, etiology of HCC, Child-Pugh score, serum α-fetoprotein level, status of liver cirrhosis, concurrent systemic therapy, and survival. MRI examinations. All examinations were performed using a 3.0-T MRI scanner (Ingenia; Philips Medical Systems). MRI was performed before (baseline) and one month after completing RT (first post-RT; range 15-45 days). Conventional MRI sequences consisted of turbo spin-echo axial T1-weighted imaging (T1WI), T2-weighted image (T2WI), sagittal T2WI, and sagittal (spine) or coronal (pelvic bone) T1WI. Axial plane DW-MRI was obtained using a single-shot echoplanar sequence as follows: fat suppression method, chemical shift selective saturation; phase encoding direction, anteroposterior; number of averages, 3; parallel imaging, SENSE with a reduction factor of 2; water-fat shift, 10.84 pixels; interpolated voxel size, 1.367 mm; slice gap, 0 mm; breath-hold or triggering, none; diffusion time (Δ), 26.5 ms; length of the gradient pulse (δ), 16.06 ms. Sensitizing diffusion gradients were applied sequentially in the x, y, and z directions using b values of 0, 400, and 1400 28,29 . ADC maps were generated automatically on the main MRI console with a mono-exponential fitting of the three selected b values. DCE-MRI was performed using a three-dimensional fast field-echo sequence in the axial plane. Before injecting the contrast material, the pre-contrast T1-weighted fast field-echo sequences (flip angles, 5° and 10°) were applied according to the same geometry to calculate the baseline T1 maps with the same axial three-dimensional fast field-echo sequence. DCE-MRI was performed immediately after injecting a bolus of gadoterate meglumine (Dotarem ® ; Guerbet) at a rate of 3 mL/s and a dose of 0.1 mmol/kg, followed by a 15-mL flush of normal saline. DCE-MRI included 1050 dynamic images with a temporal resolution of 4.3 s obtained over 5 min (flip angle, 15°; parallel imaging, SENSE with a reduction factor of 2; breath-hold or triggering, none; fat suppression, none). Contrast-enhanced T1WIs were collected after DCE imaging in the axial and sagittal planes of the thoracic or lumbosacral spine and in the axial and coronal planes of the pelvic bone (Table 1). Three months after completing RT, follow-up MRI was performed using conventional sequences (second post-RT; range 70-100 days). Image analysis. DCE-MRI maps were generated using image-processing software (IntelliSpace Portal version 10.0; Philips). The signal intensity on MRI was converted into an equivalent concentration of contrast material using the variable flip angle method 30 . DCE parameters (K trans , rate constant [k ep ], volume fraction of the extravascular extracellular matrix [v e ], and blood plasma volume [v p ]) were estimated using the extended Tofts model 31 with the population-averaged arterial input function (AIF) 32 . Two independent radiologists (readers I and II with 15 and 5 years of experience in musculoskeletal MRI, respectively) who were blinded to clinical information performed the tumor segmentation on anatomic reference images; the axial T1WI, T2WI, and postcontrast T1WI, in which tumor margins were most clearly delineated, were selected. The reference image, ADC, and DCE parameter maps were loaded into a multimodality tumor tracking application (IntelliSpace Portal version 10.0; Philips). After selecting the regions of interest (ROI) using the "smart ROI" tool with edge detection, the ROIs were automatically propagated in the craniocaudal direction ( Fig. 1). Manual adjustments were performed to ensure accuracy in encompassing the whole tumor volume, including both intraosseous and extraosseous components. Adjacent vertebral endplates or intervertebral discs were carefully avoided. ROIs drawn on anatomic reference images were simultaneously and automatically drawn on the corresponding location on ADC and DCE parameter maps. The mean values of ADC, K trans , k ep , v e , v p , and volume from the volumetric ROI were recorded.
Evaluation of treatment response. The  www.nature.com/scientificreports/ www.nature.com/scientificreports/ 33 . The patients were categorized into CR, PR, PD, or SD as follows: CR, normalization of signal intensity; PR, ≥ 50% decrease in size; PD, ≥ 25% increase in size; and SD, < 25% increase or < 50% decrease in size. The measurements were based on the sum of the perpendicular bi-dimensional measurements of the greatest diameters of each lesion in the baseline and second post-RT MRIs analyzed by reader II. Patients with CR, PR, and SD were regarded as the non-PD group, and those with PD were regarded as the PD group. Pain status was assessed using the numeric rating scale (NRS) score three days before initiating RT, during the course of RT, and one and three months after completing RT. To evaluate the pain response after RT, the categories of the International Bone Metastases Consensus Group were used to adjust the confounding effects of analgesics 34 . To apply these categories, we calculated the oral morphine equivalent dose (OMED) of all analgesics administered to patients before and after RT. Neurological symptoms were graded according to the neurological grading system for spinal cord compression by metastatic tumor 35 . Toxicities related to the treatment were evaluated according to the Common Terminology Criteria for Adverse Events version 5.0. . Patient characteristics were compared between the non-PD and PD groups; the continuous and categorical variables were analyzed using the Mann-Whitney test and Fisher's exact test, respectively. For continuous variables with P values < 0.20, a receiver operating characteristic (ROC) curve was constructed, and the area under the curve (AUC) was calculated. The optimal cutoff points were based on the maximum Youden index. The interobserver agreement between readers I and II was assessed using the intraclass correlation coefficient (ICC). An ICC of 1.0 was considered to represent perfect agreement; 0.81-0.99, almost perfect agreement; 0.61-0.80, substantial agreement; 0.41-0.60, moderate agreement; 0.21-0.40, fair agreement; and 0.20 or less, slight agreement 36 .
All statistical analyses were performed using MedCalc Statistical Software version 19.4.0, and P values < 0.05 were considered statistically significant.

Results
Among 14 patients, four were excluded for the following reasons: withdrawal of consent (n = 2), inability to undergo MRI examination owing to a deterioration in his/her general condition (n = 1), and inappropriate MRI acquisition (n = 1). Ten patients were finally included. Proton beam therapy was performed in only one patient among the ten patients. The median age and follow-up duration were 63 years (range 43-73 years) and 6 months (range 3-7 months), respectively. The median time interval between completing RT and the first post-RT MRI was 30 days (range 23-34 days). Four patients experienced PD of the target lesions in the second post-RT MRI and two died of disease progression. There was no significant difference in clinical variables between the PD and non-PD groups ( Table 2).
Only three (30%) patients complained of pain, with NRS scores of four (n = 1) and three (n = 2), while the other seven patients did not have any pain relevant to the target lesions and were not administered any analgesics.  www.nature.com/scientificreports/ patient had neurological symptoms graded as b, showing radiculopathy, which was relevant to the target lesion. The neurological symptoms were relieved one month after RT; however, relapse was observed owing to PD of the target lesion. No grade 3 or 4 toxicities were observed during the follow-up (Table 5).

Discussion
We evaluated changes in DW-and DCE-MRI parameters of bone metastases from HCC after RT and assessed their prognostic significance. Significant post-RT changes were noted in ADC, k ep , and v e . In addition, the percent change in ADC one month after RT was significantly different between the PD and non-PD groups, suggesting that it may help predict treatment response, which is considered to be unique to our study.  Figure 6. Boxplots for (a) ΔADC%, (b) baseline ADC, and (c) baseline volume in the non-PD and PD groups. The top and bottom of the box denote the 25th and 75th percentiles, respectively. The mid lines and bars indicate the medians and 5th-95th percentiles, respectively. ADC apparent diffusion coefficient, PD progressive disease. Table 5. Toxicity profiles related to radiation therapy. www.nature.com/scientificreports/ Several studies have suggested that pre-and posttreatment ADC could serve as a prognostic factor in various malignant tumors 20,37-39 , including HCC 40,41 . Our results were comparable to those of previous studies 20,25,37-41 , showing lower ΔADC% in the PD group. Furthermore, the ΔADC% could help differentiate between the PD and non-PD groups with 100% sensitivity and specificity using a cutoff of 6.79%, suggesting its potential as a predictor for early local tumor recurrence. Indeed, we acknowledge that validation of this cutoff value should be mandatory in future investigations, considering the repeatability of ADC measurements 18 and intervendor differences 42 , which could be regarded as a limitation of DWI, and the small sample size of the present study; whether MRI can predict treatment response even earlier (e.g., within one month post-RT or during RT) or whether artificial intelligence and machine learning can predict treatment response are topics for future research. Regarding RT, there have been controversies regarding the optimal RT regimen for HCC bone metastasis 9,43 and the dose-response relationship in HCC 44 . However, the high rates of up to 50% of retreatment following the use of conventional doses of RT 7 have suggested the need for high-dose irradiation 9,10 . In this study, the crude rate of early local tumor progression 3 months after conventional RT was 40%. A subsequent boost with RT or early surgical interventions in patients showing a low ΔADC% at 1 month after the initial RT may improve local tumor control, and further studies are necessary to define optimal patient selection.

Toxicities Grade 1 (%) Grade 2 (%) Total (%)
Similar to a previous study 45 , the baseline ADC of HCC bone metastases was relatively low in both the PD and non-PD groups, considering that the ADC of various pathologic bone marrow lesions generally ranged between 0.7 and 1.0 (× 10 -3 mm 2 /s) [46][47][48] . With HCC being a hypervascular tumor, we considered that intratumoral hemorrhages within metastatic bone lesions may have contributed to the low ADC 12 . Unexpectedly, baseline ADC tended to be higher in the PD group, in contrast to previous studies that reported lower baseline ADC to be a risk factor for early recurrences or incomplete responses 49,50 . However, studies with contrasting results have also been reported, with higher baseline ADC values showing poor responses to chemotherapy or RT 38,51,52 . As necrotic tumors are less sensitive to chemotherapy or RT 52 , poor responses with higher baseline ADCs are likely to result from tumor necrosis. Although pseudo-diffusion could be another possible explanation 12 , its contribution is unlikely, as no significant differences were noted between DCE-MRI parameters of the two groups 53 .
It has been suggested that DCE-MRI parameters have potential as biomarkers for predicting prognoses and detecting treatment responses [54][55][56][57] . Regarding bone lesions, the v p and K trans decreased after RT, with v p being the most strong predictor of treatment responses 27,58,59 . In contrast, we observed a significant decrease in k ep and an increase in v e after treatment; K trans showed no significant change, possibly because k ep and v e changed in opposite directions. Furthermore, none of the DCE-MRI parameters could differentiate between the PD and non-PD groups, contrary to our expectation that they may also serve as prognostic factors for metastatic bone lesions from HCC. Although irrelevant to clinical outcomes, their significant changes implied that they can reflect pathophysiological changes after RT. As tumor cellularity and volume of extravascular extracellular space are inversely correlated 60,61 , it was reasonable that v e decreased and ADC increased after RT. Meanwhile, the discrepancy observed between ADC and v e in terms of their predictive values may be explained by the different extravascular extracellular space-related tumor environments 23,62 . In addition, we speculated that the method of ROI placement in our study could be one of the contributing factors for the negative results regarding DCE-MRI parameters, considering that previous studies placed ROIs mostly around hot spots representing a higher overall perfusion 58,59 . While the desirable placement of a ROI for tumor analysis remains debatable, tumor vascularity may have been underestimated in our study by the whole tumor assessment that did not exclude non-enhanced necrotic areas 63 . Nonetheless, we believe that our method using a multimodal tumor tracking application is one of the strengths of this study as it is less biased by the ROI choice and ensures the same ROI placement among different MRI sequences. Scanner, software, or operator-dependent variabilities, which are limitations in DCE-MRI 64 , or inhomogeneous responses between the intraosseous and extraosseous components 65 can also be potential factors for the negative results that are contradictory to those of previous studies 27,58,59 .
Although there was no significant difference in volume when the whole study sample was assessed, some tumors showed an apparent increase in volume in the first post-RT MRI. Among the five patients who showed an increased post-RT tumor volume, only two were categorized into the PD group, which is partially comparable to the phenomenon termed "pseudo-progression" 66 . Pseudo-progression, first described in brain gliomas after RT and chemotherapy, is defined as treatment-related transient tumor growth 67 . Although there have been reports regarding pseudo-progression of bone lesions after high-dose stereotactic radiosurgery 66 , our results may imply that pseudo-progression can occur even after conventional dose regimens for bone metastasis from HCC. Further large-scale studies are necessary to validate these results.
Our study had several limitations. First, the sample size was limited to only ten patients, which may have influenced the reliability of the results, and the lack of multivariable analyses owing to the small sample size prohibited the determination of whether the predictive value of ΔADC% was independent of other MRI and clinical variables. Second, we used average DW-and DCE-MRI parameters calculated by two readers. However, owing to the high interobserver agreement, there was partial justification for the adoption of this method. Third, the physiology of individual patients may not have been appropriately reflected in the DCE-MRI parameters that were calculated based on a population-averaged AIF; although this method may have been advantageous in terms of reproducibility 32 . Fourth, the use of 0 s/mm 2 as the first b value instead of 50 s/mm 2 may have led to perfusion-related contributions to the ADC measurement 12 . Fifth, there may have been a mismatch of ROI between sequences. In particular, different slice thicknesses may have potentially resulted in discrepancies at the periphery of the lesions. Finally, the inclusion of both the enhanced and non-enhanced areas may have influenced the study results.
In conclusion, ADC and quantitative DCE-MRI parameters of metastatic bone lesions from HCC changed significantly in post-RT MRI. The percent change in ADC in early post-RT MRI can be used to evaluate treatment responses and may also predict local tumor progression. Future studies with larger patient populations and long-term clinical outcome evaluations are necessary to validate these findings.