Geometric Variation in Gastrointestinal Organs-At-Risk in MR-Guided Online Treatment Adaptation

Abstract

MR-guided RT has enabled online treatment adaptation by evaluating dose to anatomy defined on pre-treatment MRI. Adaptation is especially compelling for treating GI cancers where MRI allows OARs to be reliably delineated. This study characterizes daily anatomical variation observed during adaptive treatments and identifies geometric measures that correlate with the decision of whether or not to adapt treatment. These quantities for characterizing clinically meaningful anatomy changes will be useful for better understanding potential for margin reduction and dose escalation while preserving OAR sparing. Pre-treatment imaging was acquired for 56 patients treated for GI cancers. All patients were treated via SBRT in five fractions. Adaptation was performed depending on dose calculated to anatomy on a given day. OARs were retrospectively re-contoured by a radiation oncologist to within a 2 cm expansion superior and inferior to the GTV. Per OAR volume (a) variation was summarized for the cohort. Shape-based quantities including surface area (b), volume-to-surface area ratio (c), compactness (d), sphericity (e), disproportion (f) and maximum diameter (g) were calculated for all fractions. Metrics describing GTV-to-OAR proximity – including centroid-to-centroid distance (h), Hausdorff distance (i) and solid angle subtended by the OAR to the GTV centroid (j) – were also computed. Each fraction was labeled as either 1) not being adapted, or being adapted due to 2) an OAR constraint being exceeded or 3) PTV coverage being improved. Analysis of variance (ANOVA) was performed to identify significant associations between calculated metrics and labels. Median (inter-quartile range, IQR) GTV volume for the cohort was 25 (49) cc. Duodenum (DU), stomach (ST), small bowel (SB) and large bowel (LB) were contoured whenever sufficiently close to the GTV. Median volumes for OARs were measured to be 37 (27), 188 (172), 142 (209), and 261 (317) cc for DU, ST, SB and LB respectively. Largest-to-smallest variation in OAR volume per treatment course was observed to be ST, LB, SB, and then DU. Heeding notation in Methods, significant (p < 0.05) associations were measured for ST (i), DU (d), SB (a, b, c, d, h, i) and LB (h, i, j). MR-guided radiotherapy provides fruitful imaging data for assessing inter-fractional variation in OARs. Intuitive associations between various geometric quantities (volume and OAR-to-GTV proximity) and the decision of whether to adapt a treatment were confirmed to be statistically significant especially for small and large bowel. Substantial variation in OAR volume and overlap was calculated across the cohort. While this data was derived from an MR-IGRT system, these metrics constitute an initial step toward understanding GI variation over a treatment course and enabling OAR margins to be determined for achieving dosimetrically robust plans across platforms.