A Study of Dose Painting with IMRT Guided by Perfusion-weighted Magnetic Resonance Imaging for Brain Metastases

Purpose: To investigate dose escalation for brain metastases (BM) subvolumes with low cerebral blood ow (CBF) using dose-painting technology based on MR 3D arterial spin labeling (3D-ASL) images. Methods: A total of 50 patients with single BM were selected for this study. CT and MR simulation images, including contrast-enhanced T1-weighted (T1W) images and CBF images, were obtained. The gross tumor volume (GTV) was determined by fusion of contrast-enhanced CT and T1W images, and this volume was divided into high- and low-perfusion areas according to the difference in CBF. Sub-volume with less than 25% of the maximum CBF was dened as hypoxic region GTV H which was settled for dose escalation. The planning target volume (PTV) and PTV H were calculated from the GTV and the GTV H , respectively. The PTV N was calculated by subtracting the PTV H from the PTV. Plan 1 was dened as the conventional plan with a prescription dose of 60 Gy for the PTV. Plan 2 and Plan 3, respectively, escalated the prescription dose for the PTV H to 72 Gy with and without the maximum dose constrained based on Plan 1. Dosimetric indices were compared among the three plans. Results: On average, the GTV volume was 34.53 cm 3 , and the GTV H volume was 16.95 cm 3 , accounting for 49.09% of the GTV. Compared to Plan 1, the D 2% , D 98% and D mean values of the PTV H escalated by 20.50%, 19.32%, and 19.60%, respectively, in Plan 2 and by 24.05%, 6.77%, and 17.00%, respectively, in Plan 3 (all P<0.05). The doses administered in Plans 2 and 3 were 73.67±0.34 Gy and 72.37±0.61 Gy, respectively (both P<0.05). In Plans 2 and 3, respectively, the conformity index of PTV H was increased by 45.45% and 63.64%; meanwhile, the homogeneity index was sacriced, increasing from 0.04 to 0.05 and 0.11 (all P<0.05). Plan 2 achieved better values of D 98% and D mean , as well as a smaller increase in the homogeneity index, than Plan 3. The doses received by organs at risk (OARs) did not signicantly differ between the conventional plans and the boost plans (all P >0.05). Conclusions: Targeted dose escalation guided by the 3D-ASL-based CBF differences in BM effectively escalated the dose delivered to low-CBF subvolumes without increasing the dose to OARs. Computed GTV: Gross target volume; GTVH: Hypoxic gross tumor volume; GTVN: Normal gross tumor volume; PTVH: planning target volume; PTV N : Normal planning tumor volume; SIB-IMRT: Imultaneous integrated radiotherapy; OARs: Organs at risk; DVHs: Dose-volume histograms; D 2% : Radiation dose that covered 2% of PTV; D mean : Radiation dose that covered 50% of PTV; D 98% : Radiation dose that covered 98% of PTV; CI: Conformity index; HI: Homogeneity index; LSD: Least signicant difference; PLD: Post-labeling delay;


Introduction
Brain metastases (BM) are the most common intracranial malignant tumors, and approximately 8-10% of tumor patients will develop BM during their disease course [1] . At present, the main treatment for BM is radiotherapy (RT), primarily whole brain radiotherapy (WBRT) and stereotactic radiosurgery (SRS) [2,3] .
Although high-dose SRS plays an important role in the treatment of small-volume BM, its clinical application to large-volume BM is limited because of the risk of radiation-induced brain injury [4] . Largevolume BM are more heterogeneous than small ones because of the long growth cycle and complex blood supply of the former; chie y as a result of this heterogeneity, different regions of large-volume BM respond differentially to RT [5][6][7] . Previous investigation has demonstrated that large-volume tumors such as neck and head cancers have extensive hypoxic areas and that the volume of hypoxic tissue is closely related to the tumor volume [8] . It has been widely reported that tumor hypoxia is related to RT resistance and poor prognosis [5][6][7][8] .
Despite being a common strategy to recognize hypoxic regions, positron emission tomography-computed tomography(PET-CT) has shortcomings such as poor resolution and unclear presentation of tumor boundaries [9][10][11][12] . Functional magnetic resonance(MR) can address this problem by simultaneously providing anatomical and histobiological information on tumors; this technique is superior to conventional MR in evaluating the internal microenvironment of solid tumors. Dynamic contrast enhanced (DCE) can obtain hemodynamic parameters such as cerebral blood ow (CBF), cerebral blood volume (CBV) and time-to-peak (TTP), which is the main technique for clinical perfusion of MRI [13] . However, DCE has the disadvantages of inaccurately re ecting the damaged area without blood-brin brrier (BBB), long imaging time and contrast medium injection. Notably, three-dimensional arterial spin labeling (3D-ASL) perfusion imaging, which uses hydrogen protons in blood as a contrast medium, analyzes CBF parameters noninvasively and quantitatively, and tumor perfusion information is re ected indirectly [14,15] . Theoretically, a subvolume with low CBF is a hypoxic or potentially hypoxic area [5] .
Accurate identi cation of these areas, followed by targeted dose escalation for them, can solve the problem of local RT resistance and improve the curative effect of treatment [16] .
Safe targeted dose escalation for the low-CBF subvolumes of BM will help to improve local control and patient prognosis of patients on the basis of compared to population dose division. Therefore, this study examines the feasibility and dosimetric characteristics of subvolume segmentation and dose escalation for BM based on the intratumoral distribution of CBF.

Patients
From July 2018 to April 2020, 50 patients with single BM who received only RT were selected, including 29 males (age range: 33-74 years old; average age: 58.24 years old) and 21 females (age range: 34-72 years old; average age: 55.76 years old). There were 33 cases secondary to lung cancer, 9 cases secondary to breast cancer, 4 cases secondary to renal carcinoma, 2 cases secondary to esophageal cancer and 2 cases secondary to colon cancer. The patient eligibility criteria were as follows: an imagingbased diagnosis of BM; no contraindications to CT or MR; clear images with no artifacts; and a maximum tumor cross-sectional diameter of more than 2 cm. This study was approved by the Medical Ethics Approval Committee at Shandong Cancer Hospital and Institute. The need for informed consent was waived by the Medical Ethics Committee because the study was an observational, retrospective study using a database from which the patients' identifying information had been removed.
Computed tomography (CT) and MR simulation CT simulation images with a slice thickness of 3 mm and an interslice gap width of 3 mm were obtained with a CT scanner (Brilliance Big Bore, Philips, Netherlands). MR imaging was performed on a 3.0 T superconducting MR scanner (Discovery 750W, GE Healthcare, USA) if the head position was the same as in the CT simulation. 3D-ASL and contrast-enhanced T1-weighted (T1W) image data were acquired using 3D volume scanning with a 26 cm eld of view (FOV), a matrix size of 256 × 256, and a slice thickness of 3 mm. For 3D-ASL, the special acquisition conditions were as follows: repetition time (TR), 5160 ms; echo time (TE), 11.5 ms; postlabel delay, 2025 ms. For contrast-enhanced T1W images, a TR of 8.5 ms and a TE of 3.2 ms were used. Gadopentetate dimeglumine was power-injected at doses standardized by patient body weight (0.2 mL/kg body weight) at 2 mL/s, and the scan was started after 3-5 min after injection.

Target volume de nition
Contrast-enhanced CT and T1W images were fused and registered in MIM Maestro software (6.8.8, USA), and the gross tumor volume (GTV) was de ned as the region with high signal and obvious enhancement on MR. On this basis, different targeted subvolumes within the BM were identi ed according to their different CBF values, as follows.
The hypoperfused subvolume GTV H was de ned as the region with less than 25% of the maximum CBF, serving as the target subvolume for dose escalation [17] . The high-perfusion subvolume was de ned as the region with at least 25% of the maximum CBF. The nonperfused subvolume was de ned as the region with low signal on contrast-enhanced T1W images and nonperfusion on CBF images. The high-perfusion and nonperfusion subvolumes together were considered the normal prescription dose region GTV N . The planning target volume (PTV) was designed to allow a 5 mm margin beyond the GTV, and a 3 mm margin was added to the GTV H to obtain the PTV H . And the PTV N was calculated by subtracting the PTV H from the PTV, and the small volume that could not be identi ed was classi ed as part of the PTV N . A schematic diagram is shown in Fig. 1.

Treatment planning
Conventional and simultaneous integrated boost (SIB) intensity-modulated radiotherapy (IMRT) plans were de ned using Eclipse (Version 15.6, Varian, USA) for the PTV and PTV H , respectively. The conventional plan, designated Plan 1, prescribed 60 Gy to the primary PTV with a maximum dose constraint of 66 Gy. Plan 2 was based on Plan 1 but included a boost of 12 Gy to the GTV H with a maximum dose constraint of 79 Gy. The settings of Plan 3 were the same as those of Plan 2 except that no maximum dose constraint was applied to the PTV H .
The optimized parameters and dose constraints for organs at risk (OARs) in the three plans were uni ed.
OARs were restricted to the following doses: maximum dose (D max ) < 50 Gy for each eye, D max <54 Gy for each optic nerve, D max < 8 Gy for each lens, and D max <54 Gy for the brainstem.
Dose calculation was performed in anisotropic analytical algorithm optimization mode (version 15.512) using 6 MV X-rays. The calculated grid was 2.5 mm × 2.5 mm, and the prescription covered 95% of the target volume.

Plan evaluation
In order to compare the three treatment plans, the dose distribution and dose-volume histograms (DVHs) of the PTV, PTV H , PTV N and OARs were calculated. Thus, the D 2% , D 98% , and D mean of the PTV were counted. The D max values of the bilateral eyeballs, bilateral optic nerves, bilateral lenses and brainstem were compared. Furthermore, the target coverage, conformity index (CI) and homogeneity index (HI) were calculated. A CI near 1 indicates that region receiving the reference dose closely matches the shape of the target region, and an HI close to 0 indicates good uniformity [18] . The formulas for calculating CI and HI are as follows: where V t represents the volume of the target region, V t,ref represents the volume of the target region covered by the reference dose, and V ref represents the volume of all region covered by the reference dose.
A owchart of this study is shown in Fig. 2.

Statistics
Statistical analysis was performed using IBM SPSS Statistics Version 22.0. Analysis of variance was used to evaluate the differences among the three plans. The least signi cant difference (LSD) was used in pairwise comparisons. All data are expressed as x̄ ± s (mean ± standard deviation), and P < 0.05 indicates a signi cant difference.

Target volume
On average, the GTV was 34.53 cm 3 , and the GTV H was 16.95 cm 3 , accounting for 49.09% of the GTV.
The average PTV, PTV H and PTV N were 71.97 cm 3 , 41.46 cm 3 and 30.51 cm 3 , respectively. The ratios of PTV H to PTV and PTV N to PTV were 57.61% and 42.39%, respectively, as shown in Table 1 and Fig. 3. There was no signi cant difference in the D 2% of the PTV N between Plans 2 and 3 (P > 0.05), but the differences in the other parameters were statistically signi cant (P < 0.05). Notably, the D 2% values of the PTV, PTV H and PTV N all showed increasing trends among the three plans, while the improvement rates of the D 98% and D mean were higher for Plan 2 than for Plan 3, as shown in Table 2 and Fig. 4. The D max received by OARs did not signi cantly differ between the conventional plan and the boost plans (P > 0.05). Nevertheless, the boost plans slightly increased the D max of all OARs compared with Plan 1, with the D max value of the right eye increased by 5.27% to only 0.47 Gy, while those of the other OARs increased by less than 4.10% (2.19%-4.08%), as shown in Table 3.

Discussion
This study demonstrates that, through increased doses to hypoxic or potentially hypoxic low-CBF subvolumes, dose painting with 3D-ASL-de ned subvolumes can feasibly be accomplished without increasing the dose delivered to the OARs. This technique holds the potential to achieve better control of BM than can be achieved with traditional RT.
BM is a signi cant health problem whose incidence is increasing, and the median survival time is only 1-2 months without treatment [1,19−21] . Numerous studies have indicated that increasing the dose of RT signi cantly improves the local control of BM, and some patients have achieved long-term survival [22,23] .
Therefore, RT is an irreplaceable treatment for BM [3,4,22,23] . However, an increased radiation dose, particularly to vascular endothelial cells and glial cells, is associated with increased toxicity and reduced tolerance to treatment [24] . Thus, dose escalation without guidance is associated with a high risk of radiation damage to brain tissue. Targeted dose escalation in speci c locations at risk for radiation resistance can effective improve on the safety of RT. BM is highly heterogeneous tumors whose radiationresistant regions consist mainly of hypoxic or potentially hypoxic tissue.
Some studies have shown that BM has highly heterogeneous biological characteristics according to their pathological sources or even their sites of pathological origin [25] . Hypoxia is one of the most important aspects of tumor heterogeneity.
The blood ow in hypoxic tissue is slower than that in normal tissue and cannot satisfy the oxygen requirement of the rapidly proliferating tumor cells because of highly irregular tumor vessels, arteriovenous shunts, blind ends, an incomplete basement membrane of vascular epithelial cells and other factors [26] . Hence, due to the uneven distribution of blood ow and cancer cells, BM tumors appear as RT-sensitive regions with high perfusion, hypoxia radiation-resistant regions with low perfusion, and necrotic regions. Furthermore, the isoeffective dose can be up to three times higher under hypoxic conditions than under normoxic conditions [27] . However, the group-based uniform dose given under a conventional plan cannot guarantee a su cient dose for radiation-resistant regions, which eventually leads to further tumor progression and recurrence; local control failures are not uncommon under conventional treatment. Accurate and comprehensive identi cation of hypoxic regions is crucial.
Perfusion-weighted MR can be used for quantitative analysis of blood ow parameters, which is useful for identifying hypoxic regions [14,15,17,28] . The noninvasive technology of 3D-ASL re ects angiogenesis and other functional features of the tumor microvascular system, rather than re ecting only morphology as CT and conventional MR; 3D-ASL provides quantitative information for the diagnosis and treatment of brain tumors from the perspectives of morphology and function [14] . Yukie et al. demonstrated that the area under the curve (AUC) value of ASL for the recognition of hypoxic areas reached 0.830 based on the hypoxia tracer 18 F-uoromisonidazole [10] . Our previous studies also demonstrated that the CBF variations of brain tissue and BM following radiation dose gradients could be quanti ed by 3D-ASL. With the development of IMRT, a dose escalation method called SIB can implement dose escalation for speci c tumor subvolumes [29] . Dose painting aims to apply relatively high radiation doses to hypoxic, radiationresistant regions to improve tumor control without damaging OARs, and both 3D-ASL and SIB-IMRT are feasible methods for dose painting [30] . Thus, in the present study, SIB-IMRT was used to achieve dose escalation in hypoxic subvolumes that were recognized and segmented by 3D-ASL.
Tumors tend to grow rapidly, showing a hypoxic state of low blood ow, if the growth rate of tumor cells exceeds the production rate of intratumoral blood vessels [25] . Interestingly, 49.09% of the GTV was within a region of low blood ow. Our results showed that an extensive hypoxic subvolume existed within BM and that a dose-painting approach using 3D-ASL is feasible for BM treatment.
Dose painting for hypoxic regions based on various PET tracers has been widely studied in head and neck tumors, but the application of PET is indisputably limited at present because of its invasiveness and high price [11,12] . In the present study, the results of dose escalation based on 3D-ASL demonstrated that the two boost plans signi cantly increased the dosimetric indices compared with the conventional plan of 60 Gy. However, there were two trends. On the one hand, the maximum doses of the three classes of PTVs all took on ascend trend among the conventional plan, the constrained boost plan and the unrestricted one. On the other hand, the constrained boost plan had better average and minimum doses than the other two plans, and this was the case for all three classes of PTVs. Thorwarth et al. noted that doses of up to 82 Gy may be applied to head and neck tumors without increasing toxicity, but constraints for normal tissue were not stated in their work [31] .
Our results indicated that the dose delivered to OARs was increased less than 4.10% (2.19%-4.08%) except for the right eye (5.27%; 0.47 Gy) when the prescription dose for hypoxic subvolumes was increased by 20%. Undoubtedly, the absolute dose was still far below the dose constraint. In addition, the constrained boost plan outperformed the unconstrained plan in terms of OAR protection.
The conventional plan and both boost plans achieved target coverage of more than 98%, which indicated that the boost plans met the clinical dose requirement. Likewise, the constrained boost plan had better target coverage than the unconstrained boost plan and the conventional plan. Reducing the uniformity of the radiation dose is bene cial for dose escalation and OAR protection [11] . The HI of the PTV H increased from 0.04 to 0.05 in Plan 2, suggesting that this boost plan achieved a well-targeted dose distribution.
Meanwhile, it is necessary to con rm CI because the shape of the hypoxic subvolume as de ned by blood ow is often irregular [30] . In this work, the CI of the hypoxic subvolume was effectively improved through the boost plans with a minor sacri ce in terms of the HI, which further demonstrated the feasibility of our experimental method.
This study demonstrated that hypoxic subvolume delineation and dose escalation based on low CBF are potentially feasible and useful. There are still some limitations in this study. First, post-labeling delay(PLD) is one of the important parameters for accurate evaluation of CBF. However, in practice, it is di cult to ensure that PLD is set according to the speci c conditions of patients to adapt to the arrival time of labeled arterial blood. Second, the effect of boost plans for BM still needs to be con rmed in clinical practice. Relevant research is under way.
In summary, hypoxia is one of the main reasons for the poor effect of RT on BM. It is feasible to delineate hypoxic subvolumes and escalate their radiation doses under the guidance of CBF distribution based on the quantitative method of 3D-ASL. An increased dose is targeted to the hypoxic subvolume without increasing the dose delivered to OARs. This approach provides an effective individualized dose-painting strategy for patients with BM.

Availability of data and materials
The data set used and analyzed during the current study are available from the corresponding author on reasonable request.

Ethical approval and consent to participate
This study was approved by the Medical Ethics Approval Committee at Shandong Cancer Hospital and Institute. The need for informed consent was waived by the Medical Ethics Committee because the study was an observational, retrospective study using a database from which the patients' identifying information had been removed.

Consent for publication
Not applicable.