Dosimetric evaluation of bone marrow sparing in proton radiotherapy for cervical cancer guided by MR functional imaging

Background To segment the pelvic active bone marrow (PABM) using magnetic resonance (MR) functional imaging and investigate the feasibility and dosimetric characteristics of cervical cancer proton radiotherapy for active bone marrow (ABM) sparing. Methods We collected CT and MR simulation images of 33 patients with cervical cancer retrospectively. The PBM was contoured on the MRI FatFrac images; the PBM was divided into high-active bone marrow (ABM-high) and low-active bone marrow based on the fat content of the PBM. Four radiotherapy plans were created for each patient, which included intensity-modulated photon therapy (IMRT), bone marrow sparing IMRT (IMRT-BMS), intensity-modulated proton therapy (IMPT), and bone marrow sparing IMPT (IMPT-BMS). The dosimetric differences among the four plans were compared. Results The ABM-high volume in the enrolled patients accounted for 45.2% of the total ABM volume. The target coverage was similar among the four radiotherapy plans. IMRT-BMS, IMPT, and IMPT-BMS reduced the Dmean of ABM-high by 16.6%, 14.2%, and 44.5%, respectively, compared to the Dmean of IMRT (p < 0.05). IMPT-BMS had the best protective effect on the bone marrow. Compared to IMRT, the volume of ABM-high receiving an irradiation dose of 5–40 Gy decreased by 10.2%, 36.8%, 58.8%, 67.4%, 64.9%, and 44.5%, respectively (p < 0.001). Conclusions The MR functional imaging technique helped in the grading and segmentation of PABM. MR functional image-guided proton radiotherapy for cervical cancer can achieve optimal BMS. Supplementary Information The online version contains supplementary material available at 10.1186/s13014-022-02175-3.


Background
Concurrent chemoradiotherapy can improve the clinical efficacy of cervical cancer treatment, but hematological toxicity (HT) can considerably decrease the survival time and quality of life of patients [1,2]. Compared to radiotherapy alone, concurrent chemoradiotherapy affects the systemic immune response, resulting in a decrease in compensatory responses in the bone marrow (BM) outside the pelvis and an increase in the incidence of HT [3][4][5][6][7]. The unavoidable dose of radiation to the pelvic active bone marrow (PABM) in cervical cancer radiotherapy is an important cause of HT. Low-dose irradiation of 10-20 Gy to PBM is the primary cause of HT [8][9][10]. In our previous study, we found a significant positive correlation between fat content changes in the 5-10 Gy irradiated PBM region and the nadir of lymphocytes and neutrophils during chemoradiotherapy [11]. A metaanalysis showed that bone marrow sparing (BMS) techniques could effectively reduce the bone marrow dose Qin et al. Radiation Oncology (2022) 17:207 and the incidence of grade 2 or 3 HT by approximately 70% [12].
BM is divided into hematopoietic active bone marrow (ABM) and inactive BM, each accounting for approximately 50%. Approximately half of the ABM is distributed in the pelvic region. A dose of radiation to the ABM during radiotherapy is a major factor in the occurrence of acute HT. Various quantitative and qualitative imaging techniques help to accurately identify active fractions in the BM [13]. The accurate localization and sparing of ABM can reduce HT and improve tolerance to chemotherapy and survival rates [14,15]. Magnetic resonance (MR) FatFrac imaging can be used to quantitatively analyze the fat content of the BM and can reflect the fat content changes in PBM during cervical cancer radiotherapy [11]. MR functional imaging provides a feasible method for ABM segmentation.
A photon-based BMS plan can effectively reduce the exposed volume of BM. The proton Bragg peak can further concentrate the dose to the target relative to the photon plan. Only a few fields are needed to meet the dose coverage of the target. Proton therapy has great potential for BMS due to its physical dose deposition properties that can effectively protect normal tissues adjacent to the target [16,17]. Therefore, we used MR functional imaging to segment the PABM. We also analyzed the dosimetric differences between photon and proton BMS radiotherapy plans for cervical cancer. In this study, we determined the feasibility and dosimetric characteristics of MR functional imaging-guided BMS proton radiotherapy for cervical cancer.

Materials and methods
In total, 33 patients with cervical cancer who received concurrent chemoradiotherapy in Shandong Cancer Hospital from 2019 to 2020 were recruited. All patients received 45-50 Gy/25 fractions of pelvic intensity-modulated photon therapy (IMRT) combined with concurrent cisplatin chemotherapy. Each patient underwent CT and multi-sequence MR simulation scans. CT scans were performed using a Philips 16-slice Brilliance bigbore computed tomography scanner (Philips Medical Systems, Amsterdam, Netherlands) with a 3-mm slice gap thickness and 3-mm slice. The patients were immobilized in the supine position with thermoplastic molds or in the prone position with an abdominal pelvic fixator. Magnetic resonance imaging (MRI) was performed using a 3.0T superconducting MR scanner (Discovery 750w, GE Healthcare, USA) with the same position and fixed device as those used for the CT scans. All patients underwent T1WI, T2fs, and IDEAL IQ sequence scans with 3 mm slice thickness. The IDEAL IQ sequence scans were reconstructed to obtain the fat fraction (FatFrac IDEAL IQ) images.

Target and BM contouring
The targets and organ at risk (OAR) were contoured on CT and MR fusion registration images. The gross tumor volume (GTV) of the uterus, cervix, parametrium, upper third of the vagina, and locoregional lymph nodes, which included common, internal and external iliac, obturator, and presacral lymph nodes, were all included in the clinical target volume (CTV). Then, 5 mm margins were added to the CTV while creating the planning target volume (PTV). OARs included the BM, bladder, rectum, spinal cord, left femoral head, and right femoral head.
The CT and MR simulation images were transmitted into MIM 7.1.7 (MIM Software Inc., Cleveland, OH, USA). Total ABM was contoured on the FatFrac images from the L4 vertebral body to the ischial tuberosities, including the sacrum, L4-5, and pelvis [18]. The signal values of urine in the bladder (defined as the water signal) and subcutaneous fat (defined as the fat signal) of each patient were measured separately on the FatFrac images. The average of the two signal values was defined as the segmentation threshold between high-active and lowactive BM.
The threshold tool in MIM was used to contour highactive and low-active BM. The BM region with a signal value between the water signal and the segmentation threshold was defined as the high-active BM (ABMhigh ), and that with a signal value between the segmentation threshold and the fat signal was defined as the low-active BM (ABMlow ) (Fig. 1). Segmented BM and ABMhigh regions were superimposed (rigid registered) onto the CT simulation images. Finally, the CT images with BM and ABMhigh structures were imported into the RayStation planning system.
For the IMRT and IMRT-BMS treatment plans, nine coplanar fields were generated. All photon treatment plans were calculated using the Monte Carlo dose calculation algorithm and the DMLC technique based on the TureBeam linear accelerator (Varian Medical Systems, Palo Alto, CA, USA) with a 6-MV photon beam. As the primary objective, we aimed to reduce the percentage volume of ABMhigh receiving 5-40 Gy as much as feasible when the target and other OARs were met the objectives.
For the IMPT and IMPT-BMS treatment plans, two posterior oblique beams (gantry angles of 150° and 210°) and two semi-lateral beams (gantry angles of 85° and 275°) were used for the patients in the supine position ( Fig. 2A). Two anterior oblique beams (gantry angles of 30° and 330°) and two semi-lateral beams (gantry angles of 85° and 275°) were used for the patients in the prone position (Fig. 2B). The pelvic target received a dose coverage from all four beams. Robust optimization was performed for the proton treatment plans, with a 3% range and a 0.3 cm set-up uncertainty in all orthogonal directions [20]. The Monte Carlo dose calculation algorithm with 1% uncertainty was used to calculate clinical dose distributions. The proton treatment plans were generated  For IMPT and IMPT-BMS, there were nine set-up errors and three range errors creating 27 individually "perturbed doses". For each target and OAR dose-volume objective evaluated, the static scenario (no error) and worst scenario were extracted from all the "perturbed doses", respectively.

Dosimetric evaluation and statistical analysis
CTV was used as the evaluation target volume for robustness optimization because PTV was used to account for delivery uncertainties that might change the intended CTV dose. For CTV, the dose of 2% and 98% volume (D 2% , D 98% ) and the mean dose (D mean ) were analyzed. For OARs, the V X (volume percent irradiated with xGy) and the D mean were analyzed. To compare the results, the coverage of each treatment plan was normalized to the same level of 95% volume of the PTV received 100% prescription dose.
For evaluating the robustness of CTV coverage in each scenario, the median value of D 98% , D 2% , and D mean in this scenario was compared to the prescription dose (45 Gy) by conducting one-sample Wilcoxon signed-rank tests. For evaluating the robustness of OAR and ABM in each scenario, the median values of various dose-volume metrics in this scenario were compared to the IMRT values by conducting paired Wilcoxon signed-rank tests. Onesample Wilcoxon signed-rank tests were conducted to analyze the dose volume parameters of the CTV and OARs for different body position.
Friedman's statistical test was conducted to determine whether there were any differences among the four treatment plans. Because the results of all Friedman's tests were significant, the differences among dose parameters were assessed using paired Wilcoxon signed-rank tests. All analyses were performed using IBM SPSS Version 25 (IBM SPSS Inc, Chicago, IL), and all differences were considered to be statistically significant at p < 0.05.

BM location
The mean volume of BM in all patients was 299.37 cm 3 (191.32-417.54 cm 3 ). The average volume of ABMhigh was 135.44 cm 3 (19.09-267.78 cm 3 ), accounting for 45.2% of the total ABM volume.

Robustness evaluation and body position evaluation
In all perturbed scenarios, the worst values for CTV D 98% , D 2% , and D mean were similar to the prescribed dose. All the IMPT and IMPT-BMS worst scenario values for OAR were significantly better than or not different from IMRT values. The worst scenario values of ABM and ABMhigh V 5 -V 40 and D mean were all significantly lower than IMRT values. The data are shown in Additional file 1: Tables S1, S2, and S3.
The dose differences in different body positions are shown in Additional file 1: Tables S4, S5, and S6. In this study, there were 15 patients in the supine position and 18 patients in the prone position. The median volume of CTV was 521.26 mL and 585.88 mL, respectively. The target coverage between different body positions was similar. Compared to that in the supine position, the left and right femoral head and rectum V 30 of the patients in the prone position increased slightly, and the doses of other OARs were similar. In IMPT, the V 20 , V 30 and D mean of ABM and ABMhigh in the prone position were higher than those in the supine position.

Dosimetric comparison between different treatment plans IMRT versus IMPT
The CTV D 98% of the IMRT and IMPT treatment plans were similar. For IMPT, the CTV D 2% and D mean were 0.6% and 0.4% higher than those for IMRT, respectively (Table 1). Compared to IMRT, IMPT decreased the D mean of the bladder, the left, and right femoral heads, and the rectum by 17.1%, 2.6%, 2.6%, and 4.2%, respectively; the V40 of the bladder and the V 30 of the rectum decreased by 3.2% and 8.6%, respectively (p < 0.001) ( Table 2). The median reductions of the BM by IMPT were 0.8% for V 5 , 8.6% for V 10 , 25.3% for V 20 , 15.5% for V 30 , 9.6% for V 40 , and 12.1% for D mean (all p < 0.05) compared to the reduction by IMRT. ABMhigh also decreased simultaneously by 0.6%, 8.2%, 23.5%, 15.8%, 6%, and 14.2% (p < 0.05) ( Table 3).

IMPT versus IMPT-BMS
Compared to that in the IMPT treatment plans, the CTV D 2% and D mean in IMPT-BMS increased by 1.1% and 0.7%, respectively (p < 0.05) ( Table 1). The V 30 of the rectum and femoral heads decreased by 5.5% and 1.9%, respectively, in the IMPT-BMS and the V 40 of the bladder decreased by 11.5% compared to that in the IMPT   17:207 plan (p < 0.05) ( Table 2). IMPT-BMS decreased the BM V 10 -V 40 and the D mean , and the reduction range of each index was 15.5%-47.1%. The ABMhigh showed a similar trend, where the V 10 -V 40 and the D mean decreased by 9.7%, 31.2%, 46.2%, 61.3%, 66.9%, and 35.3%, respectively (p < 0.001) ( Table 3).

IMRT-BMS versus IMPT-BMS
Compared to that in the IMRT-BMS treatment plans, the CTV D 98% , D 2% , and D mean increased by 0.2%, 1.5%, and 0.9% in the IMPT-BMS treatment plans, respectively ( Table 1). The dose-volume parameters of all OARs in the IMPT-BMS plan were lower than those in the IMRT-BMS plan, with the greatest reductions in bladder-V 40 and rectum-V 30 by 22.7% and 24.4%, respectively (Table 2). IMPT-BMS performed better than IMRT-BMS. The V 5 -V 40 and D mean of BM and ABMhigh were reduced by the IMPT-BMS plans compared to their values in the IMRT-BMS plans. The reductions were as follows: 8.7% and 10.2% for V 5 , 24.7% and 29.7% for V 10 , 41.2% and 44.2% for V 20 , 54.2% and 57% for V 30 , 54.6% and 59.4% for V 40 , and 31.2% and 33.5% for D mean (p < 0.001) ( Table 3).

IMRT-BMS versus IMPT
Compared to the IMRT-BMS plans, the IMPT plans had a similar CTV D 2% and D mean , and the CTV D 98% increased in IMPT with differences within the 0.2% range (Table 1). IMPT plans significantly decreased the V 40 of the bladder and V 30 of the rectum by 12.6% and 20%, respectively (p < 0.001). ABMhigh V 10 -V 40 and D mean increased in IMPT plans compared to that in IMRT-BMS, and the V 40 had a larger increase of 22.6% (p < 0.001). The BM showed a similar trend of change (Table 3).
Although the CTV parameters differed significantly among the four plans, the absolute difference was low, and all four types of plans had the ideal target volume coverage. IMPT-BMS showed the highest BM sparing without compromising target volume dose coverage, which significantly decreased the V 10 -V 40 and D mean of ABMhigh . Additionally, IMRT-BMS reduced the dose for ABMhigh than IMRT but increased the dose in the bladder and rectum (Fig. 3).

Discussion
In this study, we investigated the role of functional MR imaging combined with proton therapy on BMS in cervical cancer radiotherapy. Our results confirmed the feasibility of using functional MR imaging to segment high-active BM and the dosimetric advantage of proton therapy for BMS, especially PABM. Huang et al. [21] reported that 69.5% of the patients with cervical cancer who received concurrent chemoradiotherapy experienced G2 + HT. For every 1 Gy increase in the D mean of PBM, the neutrophils and white blood cells of the patient decreased by 9.6/μL and 7.8/μL each week, respectively [22]. Radiation-induced lymphopenia can be associated with lower survival [23,24]. Kumar et al. [10] reported that G4 HT was associated with PBM V 20 > 45%. In many studies, PBM V 20 was significantly associated with HT, and V 40 was an important predictor of G2 + HT [12,14]. The HT produced during concurrent chemoradiotherapy can last for at least three months after treatment is completed [25,26].
The BM dose-volume parameter is the main index for predicting HT. However, many studies have used different prediction indices and dose-volume parameters, and no ideal radiation dose limitation for BM has been suggested yet. This can be associated with different definitions of the BM region used in various studies. Accurate identification and delineation of PABM are important. Most studies use CT bone window images to contour BM. However, no unified standard is available for BM delineation. CT images cannot distinguish BM activity, and there is a large error in delineation. PET/CT can segment actively proliferating BM regions based on tissue metabolism. However, the cost of PET/CT examination is high, the clinical popularity is poor, and the risk of radiation exists. MR functional imaging not only accurately delineates the BM region but also distinguishes and segments the ABM through the water-fat separation technique [27][28][29]. In another study, we reported that IDEAL IQ FatFrac imaging could be used to measure PBM fat content changes during concurrent chemoradiotherapy. The fat content of PBM increased with the increase in radiotherapy dose, especially in the PBM region close to the target region receiving high-dose irradiation (the fat fraction increased from 48.5% to 74.2%), and the change was significantly related to the HT [11]. Liang et al. [18] used the FatFrac image to contour the PBM for the IMRT-BMS plan design, which can reduce the incidence of G3 + HT from 47 to 30%. In this study, we established a method for ABM segmentation. The average value of the water and fat signal values in the FatFrac IDEAL IQ image was selected as the segmentation threshold of the ABM, and the threshold contour tool was used to manually segment the ABM. The results showed that highly active bone marrow accounted for about 45% of the total ABM volume and was dispersed, which was consistent with the results of studies by Liang and other researchers [18].
Many previous studies have been based on photon radiotherapy planning for BMS. However, some limitations are present due to the radiation dose constraints of photon beams on BM. The physical properties of protons can reduce proximal dose deposition compared with photons. The finite range and sharp distal fall-off characteristics of protons allow radiation dose to be delivered to a specific depth using the spread-out Bragg peak. The distant normal tissue receives a negligible radiation dose [16,30]. More conformal dose distributions can be achieved due to the ability of IMPT to control the beamlet energies and intensities. The irradiation dose of BM can be further reduced based on IMRT-BMS. Dinges et al. [20] used FLT-PET/CT to identify PABM in patients with cervical cancer. Compared to IMRT, the V 10 -V 40 reduction of ABM in the IMPT plan ranged from 32 to 60%. This was consistent with the results of this study that IMPT-BMS can reduce ABMhigh V 10 -V 40 by 29.7%-59.4% compared to IMRT-BMS. Lin et al. [31] demonstrated the dosimetric advantages and clinical feasibility of pencil beam scanning, and the PBM volume was significantly reduced by 24% and 17% in 10-20 Gy low-dose irradiation. Meier et al. [32] compared IMPT and VMAT cervical cancer BMS plan and found that the D mean of BM decreased from 30.76 Gy to 17.42 Gy. This was similar to the results of this study, where we found that IMPT-BMS reduced the D mean of ABMhigh from 33.7 Gy to 18.7 Gy (by 44.5%).
By comparing four types of radiotherapy plans in this study, we found that the use of proton therapy increased the protection of PABM in the low-dose region. IMRT-BMS plans set dose constraints on ABMhigh based on IMRT plans, reducing ABMhigh V 10 -V 40 and D mean .