A comparative study based on deformable image registration of the target volumes for external-beam partial breast irradiation defined using preoperative prone magnetic resonance imaging and postoperative prone computed tomography imaging

Background To explore the differences and correlations between the target volumes defined using preoperative prone diagnostic magnetic resonance imaging (MRI) and postoperative prone computed tomography (CT) simulation imaging based on deformable image registration (DIR) for external-beam partial breast irradiation (EB-PBI) after breast-conserving surgery (BCS). Methods Eighteen breast cancer patients suitable for EB-PBI were enrolled. Preoperative prone diagnostic MRI and postoperative prone CT scan sets for all the patients were acquired during free breathing. Target volumes and ipsilateral breast were all contoured by the same radiation oncologist. The gross tumor volume (GTV) delineated on the preoperative MRI images was denoted as the GTVpreMR and the tumor bed (TB) delineated on the postoperative prone CT images was denoted as the GTVpostCT. The MIM software system was used to deformably register the MRI and CT images. Results When based on the coincidence of the compared target centers, there were statistically significant increases in the conformity index (CI) and degree of inclusion (DI) values for GTVpostCT-GTVpreMR, GTVpostCT-CTVpreMR + 10, CTVpostCT + 10-GTVpreMR, and CTVpostCT + 10-CTVpreMR + 10 when compared with those based on the DIR of the thorax (Z = − 3.724, − 3.724, − 2.591, − 3.593, all P < 0.05; Z = -3.724, − 3.724, − 3.201, − 3.724, all P < 0.05, respectively). Conclusions Although based on DIR, there was relatively poor spatial overlap between the preoperative prone diagnostic MRI images and the postoperative prone CT simulation images for either the whole breast or the target volumes. Therefore, it is unreasonable to use preoperative prone diagnostic MRI images to guide postoperative target delineation for EB-PBI.


Background
Breast-conserving therapy (BCT), which involves a wide local excision followed by radiotherapy to the whole breast, has become the standard treatment for early-stage breast cancer [1]. However, for patients with a low risk of recurrence, accelerated partial breast irradiation (APBI) is now gaining acceptance as an alternative to whole breast irradiation (WBI) for early-stage cancer [2][3][4][5][6]. In addition, external-beam partial breast irradiation (EB-PBI) is an important approach to APBI. Polgar et al. [2] have reported that the efficacy of EB-PBI is equivalent to that of WBI. However, there are conflicting data regarding the acute and late toxicity of APBI. An Italian randomized trial has indicated [7] that the rates of Grades 1 and 2 acute skin toxicity in a APBI cohort were remarkably lower than those in a WBI group with decreases of 17 and 18.2%, respectively. However, in a prospective trial of 2135 patients from Canada [8], poor cosmesis at 3 years was significantly increased among those treated with APBI compared with WBI treatment, with 29% vs 17% as determined by trained nurses and 26% vs 18% as determined by the patients. Meanwhile, the rates of Grades 1 and 2 toxicity in the EB-PBI patients were also significantly higher than those in the WBI group.
Hence, based on these results, researchers are rethinking all aspects of postoperative EB-PBI. A potential factor that explains the increase in toxicity observed in the APBI group is the irradiation of a larger volume of breast tissue in those patients with poor cosmesis. Whether to ensure therapeutic efficacy or to reduce toxicity and side effects, an essential prerequisite for APBI is the accurate delineation of the target volume. However, defining the target in postoperative EB-PBI varies widely depending on the specimen volume, seroma size, clarity, surgical clips, simulation image, inter-observer variability [9] and other aspects. In addition, there is a volumetric difference for EB-PBI between the prone and supine positions [10]. Moreover, preoperative EB-PBI might be an effective approach to reducing the target volume compared to that in postoperative EB-PBI. It has been reported that both the gross tumor volume (GTV) and planning target volume (PTV) are significantly smaller in preoperative EB-PBI than in postoperative EB-PBI [11,12].
Preoperative image-guided techniques have been considered effective tools for improving the detection of tumors [13], and due to its high spatial resolution, preoperative magnetic resonance imaging (MRI) has the advantages of detecting occult tumors and providing additional valuable information regarding the primary tumor [14,15]. At present, there are few reports that focus on the feasibility of preoperative prone diagnostic MRI in guiding postoperative target delineation for prone EB-PBI. Therefore, the aim of this study was to provide a reference for how to use preoperative diagnostic MRI to guide the delineation of postoperative EB-PBI in the prone position.

Patient selection
Breast cancer patients who were suitable for EB-PBI after BCS were enrolled in this study. All the patients underwent preoperative diagnostic MRI in the prone position. Patients who had oncoplastic BCS were excluded from the trial, and equal or more than 5 surgical clips (2 mm in diameter) were used to mark the boundaries of the lumpectomy cavity. All of the enrolled patients had either no seroma or a seroma clarity score of ≦3 in the surgical cavity. None of the patients had chronic lung disease, and all exhibited normal arm movement after surgery. This research was performed in accordance with the relevant regulations, and all the patients in our research joined this study with informed consent and voluntarily underwent prone 3D CT simulation scanning. The study was approved by the institutional research ethics board of the Shandong Tumor Hospital Ethics Committee.

Image simulation and acquisition
Patients underwent preoperative prone diagnostic MRI that was performed with a Philips Achieva 3.0-T scanner (Amsterdam, Netherlands). The diagnostic MRI protocol began with preliminary imaging using fast-spin echo sagittal T2 with fat saturation, T2 weighted (T2w) turbo spin echo (TSE) with fat suppression [spectral adiabatic inversion recovery (SPAIR)]and axial T1 sequences. This was followed by dynamic high resolution simultaneous imaging of both breasts using the THRIVE sequence with 8 dynamic scans with fat saturation, performed after intravenous administration of a contrast agent (gadopentetate dimeglumine, 0.1 mmol/kg). Postprocessing consisted of 2 series of subtraction images. The total acquisition time of the preoperative diagnostic MRI protocols was 18 min. All the patients were placed in the prone position on the dedicated bilateral breast coil with no degree of incline. The coil contained two apertures open all sides to allow the bilateral breasts to hang freely away from the chest wall. The hands were naturally extended and placed on both sides of the head.
While undergoing postoperative CT simulation scanning with standard resolution, matrix 512 × 512, the patients were placed in the prone position on a dedicated treatment board (CIVCO Horizon™ Prone Breast Bracket-MTHPBB01) with no degree of incline using an arm support (with both arms above the head). The board contained an open aperture on one side to allow the ipsilateral breast to hang freely away from the chest wall. The MRI and CT images that were transferred to the MIM version 6.7.6 software (Cleveland, USA) were 3-mm thick.

Target volume delineation
All structures were delineated by the same radiation oncologist on both the preoperative diagnostic MRI and postoperative CT simulation images using the MIM system. MRI delineations were performed based on the preoperative T2WI images with voxel size 1 mm × 1.25 mm × 3 mm; the gross tumor was delineated based on hyperintense T2WI area (excluding the hypersignal gland around the primary tumor) and denoted as the GTV preMR, . The clinical target volumes (CTVs) consisted of the GTV preMR, plus 10-mm and 20-mm margins and were denoted as the CTV preMR + 10 and CTV preMR + 20 , respectively. All of the CTVs were limited to 5 mm from the skin surface and the gland-pectoralis interface. The PTVs were expanded by 15-mm and 25-mm margins from the GTV preMR, and were denoted as the PTV preMR + 15 and PTV preMR + 25 , respectively. All of the PTVs were limited to 5 mm from the skin surface and lung-chest wall interface. While delineating the CTV Breast-MRI , the delineator considered referenced breast at time of MRI, including the apparent MRI glandular breast tissue and incorporating consensus definitions of anatomical range. After delineating, the MRI physician was asked to confirmed the target volumes defined using the preoperative diagnostic MRI (Fig. 1a).
On the postoperative prone simulation CT images, the tumor bed was delineated based on the surgical clips alone and was defined as the GTV postCT . The CTV postCT + 10 was created by adding 10 mm to the GTV postCT and was limited to 5 mm from the skin surface and the gland-pectoralis interface. The PTV postCT + 15 was produced by equally extending the GTV postCT by a 15-mm margin and was limited to 5 mm from the skin surface and the lung-chest wall interface. The ipsilateral breast was contoured over the obtained MRI and CT images.
And the CTV Breast-CT was delineated by considering referenced breast at time of CT, including the apparent CT glandular breast tissue and incorporating consensus definitions of anatomical range (Fig. 1b).

Deformable image registration
This study applied the MIM system to perform the deformation registration. The VoxAlign Deformation Engine™ provided a registration algorithm for converting local registration into deformation registration in different modality images registrations. Meanwhile, the MIM system was a commercial software, so the algorithm details were not public. First, the main sequence and a subordinated sequence were selected for rigid registration. On this basis, the automatic deformation registration was implemented with set reference points, including the thorax and the center-coincidence of the compared targets. During the registration process of this study, the prone CT simulation was set to the main sequence, and the MRI T2WI image was used as the subordinated sequence. After the automatic deformation registration was completed, the Reg Reveal and Reg Refine tools were used to evaluate and revise the registration quality of the images to achieve the best visual effect (Fig. 2). The Reg Reveal tool was used for evaluating an image's final deformable registration results in the primary area of concern and the Reg Refine tool would only be used in the event that, while evaluating the initial deformation with Reg Reveal, it was determined a poor alignment was identified that needs to fixed.

Parameter evaluation
The target volumes defined using the preoperative diagnostic MRI and postoperative prone CT images were calculated separately. In addition, the correlations between the target volumes defined using the preoperative diagnostic MRI and the corresponding target volumes based on postoperative prone CT images were evaluated respectively. The degree of inclusion (DI), the conformity index (CI) and Dice's similarity coefficient (DSC) were calculated for the CTV Breast-MRI and CTV Breast-CT , the GTV postCT and GTV preMR, the GTV postCT and CTV preMR + 10 , the CTV postCT + 10 and GTV preMR, and the CTV postCT + 10 and CTV preMR + 10 . The DI was defined as follows: The definition of the DI of volume A included in volume B [DI (A in B)] was the percentage of the overlap between volume A and B in volume A [16]. The CI of volume A and B [CI (A, B)] was computed according to Struikmans et al. [17] The formula was as follows: which is defined as the ratio of the intersection of A with B to the union of A and B. DSC [18] is a commonly used metric in medical imaging and contouring studies and is defined as follows: The three-dimensional coordinates of the targets were recorded for each patient. Next, the displacements between the targets in the left-right (LR), anterior-posterior (AP) and superior-inferior (SI) directions were obtained and were defined as Δx, Δy and Δz, respectively. The distance between of the centers of mass (COMs) of the targets was calculated using the following formula:

Statistical analyses
Statistical analysis was performed using the SPSS 19.0 software (IBM Corporation, Armonk, NY, USA). The data that did not follow a normal distribution are described using medians and ranges. The Wilcoxon signed-rank test was used to compare the target volumes and relevant parameters. The Spearman rank correlation analysis was performed to establish the relevance of differences between the target volumes. The data were considered statistically significant at P < 0.05.

Patient characteristics
The study population consisted of 30 patients with early-stage breast cancer who were suitable for EB-PBI after BCS from July 2016 to April 2017. Eighteen of the 30 patients who underwent preoperative diagnostic MRI enrolled in this study. The patients had a median age of 43 years (range, 39-69 years) and had cancer of the breast with a pathological stage of T1-T2N0M0. Seven of the 18 patients had left-sided breast cancer, and the remaining eleven had right-sided breast cancer. The patients underwent a lumpectomy, which was performed with a circumferential margin of at least 1.0 cm [19], with sentinel lymph node dissection (SLND) or axillary lymph node dissection (ALND), and tumor-negative margins were ensured during a single operation. The patient and tumor characteristics are presented in Table 1.

Comparison of the parameters of the target volumes defined using MRI and CT
When based on the DIR of the thorax, the median values of the CI, DI and DSC between the CTV Breast-MRI and CTV Breast-CT were 0.56, 0.82 and 0.71, respectively. The distance between the COM of the CTV Breast-MRI and CTV Breast-CT was 1.81 cm. When based on the DIR of the thorax, the median CIs for GTV postCT -GTV preMR, GTV postCT -CTV preMR + 10 , CTV postCT + 10 -GTV preMR, and CTV postCT + 10 -CTV preMR + 10 were slightly lower than those based on the center-coincidence of the GTV preMR and GTV postCT ; the median CI values for these volumes were 0.02, 0.07, 0.04 and 0.17; 0.19, 0.31, 0.05 and 0.38, respectively (Z = − 3.724, − 3.724, − 2.591, − 3.593, respectively; all P < 0.05). The DI and DSC median values for the GTV postCT -GTV preMR , the GTV postCT -CTV preMR + 10 , the CTV postCT + 10 -GTV preMR and the CTV postCT + 10 -CTV preMR + 10 were generally low; however, there were statistically significant increases in these parameters based on the center-coincidence of the GTV preMR and GTV postCT when compared with those based on the DIR of the thorax (Z = − 3.724, − 3.724, − 3.201, − 3.724, all P < 0.05; Z = -3.724, − 3.724, − 2.591, − 3.636, all P < 0.05, respectively, Table 4).

Discussion
APBI, as a possible alternative to WBI, offers less overall treatment time and the delivery of a reduced dose to uninvolved portions of the breast and adjacent organs at risk [20,21]. Undoubtedly, the irradiation of normal breast tissue would be decreased by reducing the EB-PBI target volume, which provides conditions for reducing toxicity or other side effects and improving the cosmetic outcome [22]. During the EB-PBI target definition, first and foremost is the identification and contouring of the GTV. MRI is recognized as an excellent imaging tool in    [14,15]. In the preoperative APBI study of van der Leij et al. [11], a virtual plan was made for preoperative EB-PBI, which resulted in a reduction in the GTV compared to that with postoperative EB-PBI. Hence, we aimed to clarify whether the delineation of the target volumes for postoperative prone EB-PBI might benefit from preoperative prone diagnostic MRI and to provide a reference for how to use preoperative diagnostic MRI to guide the delineation of target volumes for postoperative EB-PBI in the prone position. Based on our analysis, the results showed that the GTV postCT was significantly larger than the GTV preMR by 12.58 cm 3 . Van der Leij et al. [11] also confirmed that a statistically significant difference was evident between the preoperative GTV and the postoperative tumor bed (7.71 cc lower in the preoperative EB-PBI target volume).
In addition, our study demonstrated that the CTV preMR + 10 and PTV preMR + 15 were significantly smaller than the CTV postCT + 10 and PTV postCT + 15 , respectively. Compared to the CTV postCT + 10 and PTV postCT + 15 , the CTV preMR + 20 and PTV preMR + 25 were significantly greater by 51.81 cm 3 and 43.99 cm 3 , respectively. Hence, if the expanded margin was too large, the preoperative EB-PBI would lose the advantage of reducing the dose to the ipsilateral breast, and the main factor to consider for the margin extension is the subclinical range. Controversy exists regarding EB-PBI treatment in terms of the subclinical range. Faverly et al. [23] have shown that a 10-mm tumor-free margin gives the best positive predictive value for breast cancers of limited extent. At present, a lumpectomy is performed with an intended macroscopic margin of at least 1.0 cm [24], and this value also represented a subclinical range that had been covered by previous studies [25,26]. Therefore, it is resonable to reconstruct the CTV MRI by adding a 1.0-cm margin around the GTV MRI . However, Schmitz et al. [13] indicated that typical treatment margins of 10 mm around the GTV MRI might include occult disease in 52% of patients for MRI-guided BCT. When expanded with a 20-mm margin around the GTV MRI , a subclinical lesion could also be found in one-fourth of the patients. This might have been a consideration for van der Leij et al. [27] in delineating the CTV MRI and PTV MRI by expanding around the GTV MRI with 20-mm and 25-mm margins, respectively. But, based on our result, the CTV preMR + 20 and PTV preMR + 25 were significantly greater, so the CTV MRI and PTV MRI by expanding around the GTV MRI with 20-mm and 25-mm margins should not be advised. In our study, in comparison with the CTV postCT + 10 based on the postoperative TB, the  11 have shown that the difference was not statistically significant. In theory, when there is equilateral excision around the primary tumor, there would be no significant difference between the CTV preMR + 10 and CTV postCT + 10 , which are defined by expansion from the primary tumor based on preoperative MRI and by the postoperative TB, respectively. This seemingly contradictory difference can be explained by the asymmetric resection of the primary tumors [28]. If the anisotropic surgical margin caused by asymmetrical resection is taken into account, the volumes of the CTV pre and CTV post are comparable [11]. Furthermore, Zhang AP et al. [28] and den Hartogh et al. [29] indicated that because the majority of surgeons subjectively perform BCS based on palpation of the boundaries, which results in the asymmetric resection of primary tumors, neither the resection specimen volume nor the TB correlate with the visible tumor volume based on preoperative MRI. However, in our study, the Spearman rank correlation demonstrated that a statistically significant positive correlation exists between the GTV preMR and GTV postCT and between the CTV preMR + 10 and GTV postCT. This finding might be explained by the fact that in addition to the determination of the resection range based on palpation, the preoperative imaging data, such as MRI scans, have recently played an increasingly significant role in BCS, and the anisotropy between the specimen edge and tumor edge has been reduced.
MRI is not only the basis for the implementation of preoperative EB-PBI but also helpful for selecting patients suitable for postoperative EB-PBI, guiding postoperative target delineation for EB-PBI [30,31]. The preoperative diagnostic MRI image was obtained in the prone treatment position, which is the same position as that of prone EB-PBI after BCS. Theoretically, DIR between the preoperative diagnostic MRI and postoperative prone CT simulation images should be conducive to the determination of the targets for postoperative prone EB-PBI. However, our study concluded that when based on the DIR of the thorax, the CI, DI and DSC were all poor for both the GTV preMR -GTV postCT and CTV preMR + 10 -GTV postCT comparisons. The distances between the COMs of the GTV preMR -GTV postCT and GTV preMR + 10 -GTV postCT were 2.71 cm and 2.67 cm, respectively. Moreover, the breast spatial matching between preoperative diagnostic MRI and postoperative CT simulation was not ideal, showing that the CI, DI and DSC values for the CTV Breast-MRI -CTV Breast-CT did not reach 1, for perfect agreement between volumes.
The poor breast spatial matching might be mainly caused by the difference between the dedicated MRI bilateral breast coil and the dedicated treatment board for prone CT simulation. For preoperative diagnostic MRI, there are two apertures open on all sides to allow the bilateral breasts to hang freely away from the chest wall; however, the postoperative CT treatment board only contains an open aperture on one side, whereby only the ipsilateral breast can move away from the chest wall due to gravity in the prone position. Meanwhile, the contralateral breast is pulled away from the ipsilateral side by the baffle as much as possible, which might affect the natural overhang of the ipsilateral breast. Our results indicated that there was relatively poor spatial overlap between both the GTV preMR and GTV postCT and between the CTV preMR + 10 and GTV postCT . This result may be attributable to the poor breast spatial matching and could also be the result of the same spatial morphology among the GTV preMR , CTV preMR + 10 and GTV postCT after DIR. Furthermore, from our analysis that was based on the DIR of the center-coincidence of the GTV preMR and GTV postCT , the CI, DI and DSC values for the GTV preMR -GTV postCT and the CTV preMR + 10 -GTV postCT were significantly improved compared with those based on the DIR of the thorax; however, these values were still poor. Therefore, it is unreasonable to use preoperative prone diagnostic MRI images to guide the postoperative target delineation for EB-PBI.

Conclusions
Overall, in Chinese early-stage breast cancer patients enrolled to undergo prone EB-PBI, when defining the target based on the preoperative prone MRI images, the target volumes were significantly smaller when compared to those based on postoperative prone CT images. However, a statistically significant positive correlation was found between the MRIand CT-based target volumes. Although based on DIR, there was relatively poor spatial overlap between the preoperative prone diagnostic MRI images and the postoperative prone CT simulation images for both the whole breast and the target volumes. Hence, it is unreasonable to use preoperative prone diagnostic MRI to guide postoperative target fusion delineation for EB-PBI. In fact, it is feasible to optimize the delineation of the postoperative EB-PBI target volumes by other means, such as clipping the surgical cavity by the surgical team in the presence of the radiation oncologist responsible for contouring for PBI and using respiratory gating with daily on board image verification before delivery of treatment can help in reducing the PTV margins. Further, studies on patterns of failure and adverse cosmetic outcome after EB-PBI can aid in refining the delineation techniques.