Dosimetric evaluation of magnetic resonance imaging-guided adaptive radiation therapy in pancreatic cancer by extent of re-contouring of organs-at-risk

Article information

Radiat Oncol J. 2022;40(4):242-250

Publication date (electronic) : 2022 December 26

doi : https://doi.org/10.3857/roj.2022.00332

Jun Yeong Song^1,2, Eui Kyu Chie^1,2, Seong-Hee Kang¹, Yeon-Jun Jeon³, Yoon-Ah Ko³, Dong-Yun Kim^2,4, Hyun-Cheol Kang^,1,2

¹Department of Radiation Oncology, Seoul National University Hospital, Seoul, Korea

²Department of Radiation Oncology, Seoul National University College of Medicine, Seoul, Korea

³Seoul National University College of Medicine, Seoul, Korea

⁴Department of Radiation Oncology, Kyung Hee University Hospital at Gangdong, Seoul, Korea

Correspondence: Hyun-Cheol Kang Department of Radiation Oncology, Seoul National University Hospital, 101 Daehak-ro, Jongno-gu, Seoul 03080, Korea. Tel: +82-2-2072-2526 E-mail: shule@snu.ac.kr

Received 2022 June 13; Revised 2022 October 31; Accepted 2022 November 14.

Abstract

Introduction

Among hundreds of cancers, pancreatic cancer has one of the worst prognoses. The median overall survival is approximately 13–24 months in patients with localized disease and 9–16 months in locally advanced cases [1–4]. Trimodality therapy is increasingly being utilized in the setting of borderline resectable pancreatic cancer (BRPC) or locally advanced pancreatic cancer (LAPC) to improve resectability, local control, and survival [5–8]. Moreover, the PREOPANC trial investigated the oncologic outcomes of resectable pancreatic cancer (RPC) and BRPC cases who received neoadjuvant chemoradiotherapy before surgery and showed a significant survival benefit in BRPC patients [7].

Among many radiation therapy (RT) techniques, stereotactic ablative radiotherapy (SABR) offers many advantages compared to conventional fractionated RT, such as a high radiation dose [9,10], a relatively short period of treatment [11], and an excellent conformality [12]. In the SABR delivered to the pancreas, the stomach, duodenum, and bowel are considered organs-at-risk (OAR) with clinical significance. These organs are mobile and radiosensitive, posing challenges when planning RT [13,14]. One study investigated the inter-fractional motion of these OARs and showed it reaches 17 to 36 mm [13]. Stereotactic magnetic resonance image-guided adaptive radiotherapy (SMART) is a tool to overcome such concerns in SABR to a tumor surrounded by these organs. It considers the daily organ migration by re-contouring OARs. Also, real-time magnetic resonance imaging (MRI) offers a superior spatial resolution of the soft tissue of intra-abdominal organs during RT and considers the intra-fractional respiratory motion.

In the current practice of SMART to the pancreas at our institution, only a portion of the OARs close to the planning target volume (PTV) is re-contoured because OARs near the PTV are enough to produce an online adaptive plan, and its safety and reproducibility were also published to reduce the most time-consuming step of the treatment [14–16]. However, the extent of re-contouring has not been thoroughly discussed and has uncertainties. Therefore, in this study, we evaluated the dosimetric difference between the radiation plan that re-contoured the entire OARs (Full OAR) and that of the actual treatment with roughly contoured OARs (Rough OAR) in SMART.

Methods and Materials

We retrospectively reviewed the medical records and treatment plans of 19 patients who received SMART at our institution from February 2018 to May 2019. The Institutional Review Board of Seoul National University Hospital (IRB No. 2003-094-1109) approved the study and waived the requirement for informed consent. All patients were treated using the ViewRay MRIdian system (ViewRay Inc., Oakwood Village, OH, USA), which incorporates a 0.35T MRI scanner with three cobalt-60 γ-ray sources [17]. The patients were treated with preoperative, radical, and salvage aims.

The simulation process of SMART included MRI simulation with corresponding non-enhanced computed tomography (CT) simulations. Each scan was obtained in the end-expiratory phase to minimize respiratory motion. After simulation, we contoured the target volume and OARs and performed treatment planning. We defined the clinical target volume (CTV) as the pancreatic mass and its infiltration to neighboring structures. A 3–5 mm margin was expanded to all directions from the CTV to make up the PTV. The prescription dose for PTV was 40–50 Gy in 5 fractions, which was adjusted considering the relative location and distance from the surrounding OARs.

The OARs for the treatment course included the stomach, duodenum, and bowel. The institutional dose constraint policy required the V₃₅ of each OAR to be less than 0.5 mL in the initial planning process. Also, in the re-optimization process for SMART, the constraint required the V₃₅ of each OAR to be less than 1 mL. The UK consensus on normal tissue dose constraints for stereotactic radiotherapy [18], which was brought from the protocols of the ABC-07 and SPARC trials, was used as guidelines for the dose constraints [19,20].

All treatment plans consisted of 5 fractions of SMART. In each fraction, a daily set-up MRI was obtained using the same protocol as for the simulation. The attending physician decided whether to proceed to the treatment without making amendments to the OAR contours by comparing the anatomy between the MRI taken during the simulation and daily set-up. If the contours needed adjustment, the physician re-contoured them in real-time. For re-contouring, our institutional policy is to adjust OARs within 2 cm from the PTV in the craniocaudal direction or 3 cm, depending on the situation. Also, the attending physician decided which OARs to re-contour, considering the inter-fractional variation and proximity to the PTV. The initial plan was immediately re-optimized based on the daily contours by modifying the dose rate and beam angle.

The entire OAR was re-contoured retrospectively for every fraction of SMART regardless of the distance from the PTV (Full OAR). We contoured the duodenum from the pylorus of the stomach to the fourth portion, where it merges with the bowel. We defined the contour of the bowel as the sum of bowel loops, excluding the duodenum. It was delineated distally to the inferior aspect of the third portion of the duodenum and had no limits proximally. A representative image of the daily MR image with OAR contours using both of the contouring methods is shown in Fig. 1. Dosimetric parameters such as the maximum point dose to a volume (D_max), the minimal dose received by the highest irradiated volume of 1 mL (D_1cc), the volume receiving 35 Gy or more (V₃₅), and the volume receiving 33 Gy or more (V₃₃) were calculated. We used the same dose constraint (V₃₅ <1 mL) to evaluate the Full OAR contouring. We reviewed the dose distributions of cases that violated the dose constraint in the Full OAR contouring only.

Fig. 1.

MR image of OAR contours according to methods of re-contouring of one treatment fraction, with PTV (light blue) and the contours of Rough OAR (red) and Full OAR (yellow). MR, magnetic resonance; OAR, organ-at-risk; PTV, planning target volume.

Statistical analysis was performed using SPSS 26.0 (IBM SPSS Statistics for Windows; IBM Corp, Armonk, NY, USA). The chi-square test was used for comparing categorical variables, and the paired t-test was used for comparing continuous variables. A p-value of less than 0.05 was considered statistically significant. Side effects were graded according to the Common Terminology Criteria for Adverse Events v5.0 (CTCAE v5.0). Side effects assessed in grade 3 or higher were termed “severe side effects.”

Results

Table 1 describes the patient and treatment characteristics. A total of 19 patients were included in the analysis. Eight patients (42.1%) had a tumor located in the uncinate process or head, and the rest had a tumor in the body or tail. The median tumor size was 3.0 cm, ranging from 2.2 to 13.2 cm. All but one patient underwent 5 fractions of adaptive radiotherapy for a total of 94 fractions. One patient received 4 fractions of RT as neoadjuvant treatment, the last fraction was skipped because the plan's quality was unacceptable, and the physician decided not to treat it after considering the risk and benefit. For every patient, we measured the distance from OARs to the gross tumor volume (GTV). The mean value was 3.5 mm, ranging from 0 to 11.9 mm. There were no patients with severe side effects related to RT.

Table 1.

Patient and treatment characteristics (n = 19)

When the three OARs were evaluated together, the dose constraint was violated in 10 fractions (10.6%), and 41 fractions (43.6%) out of 94 fractions in the Rough OAR and Full OAR methods, respectively (p = 0.075). Detailed descriptions of the violation of the dose constraint evaluated by individual OARs are listed in Table 2. Also, the dose distributions were reviewed in the treatment fractions where dose constraint violation was identified only in the Full OAR. Interestingly, no violation occurred outside 2 cm of the PTV.

Table 2.

Dose constraint violation by OAR re-contouring method

In addition, various clinicopathologic factors were analyzed, and BRPC (p = 0.023), tumor size of 2 cm or more (p = 0.027), and stage group I or II (p = 0.008) were associated with a higher proportion of dose constraint violation in Full OAR. Similarly, other factors such as a distance from GTV to OAR of less than 5 mm (p = 0.061) and primary tumor in the body and tail (p = 0.054) were associated with a higher proportion of dose constraint violations in Full OAR with marginal significance.

After considering the clinical context and applicability, tumor size, distance from GTV to OAR, and primary tumor site were chosen as factors for the subgroup analysis. Analyses were performed by dividing the factors into subgroups grouped by two of the three factors and comparing the V₃₅ of each subgroup. The V₃₅ was used in place of dose constraint violation to evaluate the dosimetric difference more comprehensively with a continuous variable. The results are shown in Table 3. For all patients, the mean value of the V₃₅ (mL) for Rough OAR and Full OAR showed a significant difference with 0.748 and 2.101, respectively (p < 0.001). A difference in the V₃₅ was consistently seen across the subgroups. Except for a marginally significant difference demonstrated in the subgroup grouped by tumor size and distance from GTV to OAR (p = 0.063).

Table 3.

Mean value of V₃₅ by OAR re-contouring method in subgroups

Other dosimetric parameters, including dose constraint, were investigated for the individual OARs. For each OAR, the differences of dose constraint violations, D_max, D_1cc, V₃₅, and V₃₃ between two contouring methods were investigated (Table 4). In the duodenum, there were no violated fractions in the Rough OAR method, but in the Full OAR, 13% of the fractions violated the dose constraint. This trend was consistent with the other dosimetric parameters—D_max (p < 0.001), D_1cc (p=0.009), V₃₅ (p = 0.040), V₃₃ (p = 0.039). However, the violated fraction in the stomach was 10.3% in the Rough OAR method and 16.1% in the Full OAR method, showing no statistically significant difference (p = 0.597). Other dosimetric parameters did not show any significant difference—D_max (p = 0.949), D_1cc (p = 0.393), V₃₅ (p = 0.628), V₃₃ (p = 0.629). In the bowel, there was no significant difference in the proportion of violated fractions between the OAR contouring methods (p = 0.109), but other dose-volume parameters indicated larger volume or higher dose in the Full OAR method—D_max (p = 0.006), V₃₅ (p < 0.001), V₃₃ (p < 0.001). To summarize, the V₃₅ of each OAR and overall OARs are depicted as a box-and-whisker plot in Fig. 2. Also, a representative image of the contours of Rough OAR and Full OAR of a SMART fraction that had substantial discrepancy of V₃₅ is depicted in Fig. 3: maximum V₃₅ of individual OARs in Rough OAR (0.24 mL) and Full OAR (9.54 mL).

Table 4.

Comparison of dosimetric parameters calculated by using each re-contouring methods

Fig. 2.

Box-and-whisker plots showing values of V₃₅ (mL) of individual OARs (A) and maximum V₃₅ values of overall OARs (B) for all the adaptive fractions according to contouring methods. V₃₅, volume receiving 35 Gy or more; OAR, organ-at-risk.

Fig. 3.

MR image of OAR contours according to methods of re-contouring of one treatment fraction Showing V₃₅ discrepancy, with PTV (light blue) and the contours of Rough OAR (red), Full OAR (blue) and isodose line of 35 Gy (light green). MR, magnetic resonance; OAR, organ-at-risk; PTV, planning target volume; V₃₅, volume receiving 35 Gy or more.

Discussion and Conclusion

In SMART, daily re-contouring of the OARs provides an accurate evaluation of the intra-abdominal OARs, in which significant inter- and intra-fractional variability exist. To the best of our knowledge, this is the first study to investigate the safety of SMART for pancreatic cancer by comparing different extents of OAR re-contouring. An ideal method to evaluate dose distribution would be to draw Full OARs for all patients. However, in practice, issues such as time constraints and patient compliance (i.e., unable to maintain a supine position) are challenging. Accordingly, this study evaluated the dosimetric difference between the two distinct OAR contouring methods and investigated which patient population differed using Full OAR contouring. We could find some significant clinical factors that can be used for patient selection for the time-consuming procedure, such as tumor size, stage, distance from the GTV to OAR, and tumor location. The dosimetric difference was evident in the subgroups defined by two of the factors listed and suggested patient subgroups that could benefit from the extensive re-contouring. Regarding the individual OARs, the dosimetric parameters did not differ significantly between contouring methods for the stomach. However, there were significant differences in the duodenum and bowel. Therefore, for patients with the risk factors described above, the duodenum and bowel rather than the stomach could benefit from Full OAR contouring.

In reviewing the dose distribution of the cases that violated the dose constraint in the Full OAR, but not in the Rough OAR, we found that violations were located solely within 2 cm from the PTV contour. This indicates that the additional detection of dose constraints in the Full OAR were not due to the delineation of the contours outside the 2 cm distance from the PTV, but could rather be attributed to the re-contouring of all of the OARs, regardless of the physician's decision. The sparing of the areas outside the 2 cm distance from the PTV is due to the co-planar technique and the steep dose gradient of SABR planning system, making delivery of high dose to remote areas unlikely [21,22]. Although these violations did not result in severe RT-related side effects, constant efforts in re-contouring the OARs could benefit in accurately evaluating the dose distribution, detecting dose constraint violation and re-planning according to the dose distribution.

We evaluated the dosimetric profiles for each fraction. Ten of 94 fractions (10.6%) showed a violation of dose constraints, even in the Rough OAR. However, one SMART course is usually made up of 5 fractions, and a dose constraint violation in a single fraction does not necessarily result in a violation for the whole treatment course. Areas that received the highest level of radiation may differ throughout the treatment course. Thus, there may not have been a violation in terms of the entire course. Daily anatomic variations of the OAR should be considered to make an accurate evaluation of the accumulated dose of the treatment course. Dose accumulation of all the fractions using deformable image registration (DIR) would be helpful. Bohoudi et al. [23] reported a study of DIR-based dose accumulation in prostate cancer treated with SMART. The authors showed that V_20Gy-32Gy from the total accumulated dose delivered to the bladder rather than the dose of each fraction correlates well with the patient's symptom score. In the pancreas, a similar approach as in this study could be used to evaluate the dosimetric profile of the entire treatment course and to find a potential surrogate index associated with dosimetry of the total treatment or the incidence of adverse events.

Recently, auto-segmentation for OARs using deep learning for RT has gained attention in the field, and techniques are evolving [24–26]. Although challenging, many studies deal with auto-segmentation in intra-abdominal organs, and they report non-inferior results compared with expert-drawn contours [27,28]. These studies are based on cone-beam CT images, but studies from other organs such as the prostate employ MRI-based auto-segmentation [29]. Although this method is not mature enough to apply to daily practice, if applied to SMART, it would reduce re-contouring from the most time-consuming step to a few seconds and needs constant attention.

The current study's analysis is based on our institution's dose constraint (V₃₅ <1 mL). Thus, when we apply a different dose constraint, it would alter the proportions of the violation. Several published dose tolerance guidelines for abdominal OARs exist for 5 fraction-SABR, each employing various dose parameter [30–33]. Also, the cutoffs of the constraint vary widely. For instance, the cutoff of D_max ranges from 30 Gy to 45 Gy [33–37]. The reason why this discrepancy occurs is that the dose constraint is drawn from normal tissue complication probability (NTCP) models, which are derived from separate datasets of complications after RT [15,38,39]. Furthermore, there is evidence that the NTCP can vary according to the irradiated volume [40], use of chemotherapeutic agents [41], and fractionation schemes [42]. Thus, a dose constraint should be chosen accordingly to the patient and tumor setting and tailored individually.

This study had several limitations. First, this study was designed as a retrospective review and contains the inherent potential for selection bias. To minimize selection bias, we included 19 consecutive patients without further selection. Secondly, the Rough OARs were occasionally omitted for some OARs under the judgement of the attending physician, thus hindering clear comparison between the two distinct re-contouring methods. Third, to apply the results of the current study to clinical practice, clinical evidence with a high level of relevance is needed for the dose constraint of SMART. We could generate more solid factors associated with OAR contouring from the prospective data with a larger number of patients. In addition, as mentioned above, the dose constraint violation from a single fraction does not necessarily translate to a violation of the total treatment. Therefore, it is ambiguous to interpret dose constraint violations concerning clinical adverse events. No severe side effects were observed among the patients in the study, even in the group with a violation. Therefore, further analysis of dose constraint violation using DIR-based dose accumulation is warranted. Moreover, the clinical significance of dose constraint violation should be clarified by assessing how many adverse events occur.

In conclusion, Full OAR differed in the proportion of violated fractions from the Rough OAR in SMART for pancreatic cancer. Patient groups with a large tumor, a short distance from the OAR to GTV, and a tumor in the body or tail showed benefit in further discriminating occult dose constraint violations and should be considered for Full OAR contours, especially in the duodenum and bowel. Also, delineating all of the OARs within 2 cm from the PTV, rather than outside 2 cm from the PTV, would help to discover occult violations. However, it should be weighted between the clinical usefulness and its cost as the dosimetric profile of SMART cannot be represented by a single fraction.

Article information Continued