Online MRI contour differences concealed duodenal dose exceedances in pancreatic SBRT

In 50 pancreatic SBRT fractions, expert recontouring identified duodenal dose-limit exceedances in two patients despite largely preserved target coverage.

KEY POINTS

  • This retrospective single-center study from Institut Jules Bordet, Belgium, analyzed 50 treatment fractions from ten patients with pancreatic ductal adenocarcinoma treated on a 1.5-T Elekta Unity MR-Linac between July 2023 and October 2024. All patients received isotoxic 50 Gy in five fractions, with online target and OAR contours generated by seven radiation oncologists during routine treatment. Each fraction was subsequently recontoured offline by a single pancreatic-expert radiation oncologist, creating an expert reference against which the original clinical contours and delivered treatment plans were evaluated.
  • Daily online adaptation was performed predominantly using Adapt-to-Shape (ATS), with 47 of 50 fractions undergoing contour review or modification and full plan reoptimization, while only three used Adapt-to-Position. Online delineation covered the gross tumor volume, tumor–vessel interface, and OARs within 3 cm of the target, including the stomach, duodenum, small bowel, colon, bile duct, and major vessels. The treatment protocol incorporated a 2-mm ITV expansion and 3-mm PTV margin, with GI OAR evaluation using 3-mm planning risk volumes.
  • Contouring variability depended strongly on anatomical structure. Online GTV contours were systematically larger than expert reference contours, with median volumes of 75.4 versus 61.4 cm³, while the mean paired per-fraction volume difference was +8.6 cm³ (range −6.7 to +38.0 cm³). Median Dice similarity for the GTV and tumor–vessel interface was approximately 0.77, indicating moderate geometric agreement. Larger structures such as the stomach and duodenum generally showed better concordance, whereas the target, major vessels, and individual bowel segments were more variable.
  • Target coverage remained largely unaffected by differences between online and expert contours. Recalculation of target metrics using the clinically delivered plans showed no statistically significant differences in V50Gy or V35Gy between contour sets. This suggests that the online tendency toward larger GTV contours helped preserve target coverage despite geometric uncertainty. However, larger target contours may also increase OAR dose exposure when treatment is optimized against incorrectly delineated bowel or other adjacent structures.
  • Expert recontouring revealed gastrointestinal dose discrepancies that had not been apparent during online planning. In two patients, the sum of per-fraction duodenal D0.5cc exceeded the nominal 40-Gy limit when evaluated on expert reference contours. For one patient, the value increased from 34.6 Gy online to 41.4 Gy on reference contours; for another, from 34.0 to 40.5 Gy. These estimates assumed that the most highly irradiated 0.5-cm³ region remained spatially coincident across fractions and therefore represent conservative summed hotspot estimates, not deformably accumulated delivered dose.
  • Two distinct anatomical problems explained the most clinically relevant discrepancies. In one patient, the bile duct was repeatedly confused with the adjacent duodenum across three fractions, while another patient had difficulty distinguishing the duodenal wall from the tumor and surrounding pancreatic tissues. Small contouring differences in these steep dose-gradient regions resulted in meaningful changes in calculated bowel exposure. Differences involving the great vessels were statistically significant but remained within the evaluated tolerance limits, and no grade ≥3 gastrointestinal toxicity had been reported in the two patients with duodenal constraint exceedances at the time of analysis.
  • The findings are limited by the small ten-patient cohort, unequal contribution of fractions by individual physicians, and use of a single expert reference rather than a multiobserver consensus. The reference physician also participated in the online treatment workflow, and the offline review had access to more imaging and time than routine daily adaptation. Importantly, true voxel-level accumulated dose and intrafraction anatomical changes were not assessed, and the absence of observed severe toxicity in two patients does not establish that the identified dose exceedances were clinically harmless.

CLINICAL TAKEAWAY

Online adaptive MR-guided SBRT for pancreatic cancer can maintain favorable target coverage despite moderate interobserver contouring variability, but errors involving adjacent gastrointestinal structures may cause clinically relevant OAR dose discrepancies. These findings support structured contouring protocols, additional attention to the duodenum–bile duct interface, and robust OAR verification during adaptation, rather than assuming that daily replanning automatically eliminates contour-related risks.

SOURCE

Radiotherapy and Oncology

Browse more research Suggest a correction