Atlas-based auto-contouring of breast tissue in young female lymphoma patients

Abstract

Purpose or Objective Manual contouring adds to the RT planning workload, thus atlas-based auto-contouring tools (ABAC) have been explored for several sites. Young women (<36yo) treated for lymphoma with supradiaphragmatic RT have an increased risk of developing breast cancer (BC) compared to the general population, and would benefit from time of treatment breast contouring via breast ABAC to minimise RT dose. However, there is limited published research in breast ABAC for non-BC patients. Here, we evaluate the performance of ABAC developed on BC patients to contour breast tissue on young lymphoma patients, and hypothesise that they will perform worse than ABAC specifically developed for lymphoma patients. Materials and Methods ABAC were built on 20 templates from 2 patient groups, including 10 female lymphoma patients (18-36yo, 8/10 arms down) treated with supradiaphragmatic RT (ABAClymph) and 10 female BC patients (36-75yo, all arms up, 8/10 right-sided breast RT) from the REQUITE study (ABACbreast). All contours were manually drawn by expert breast oncologists following ESTRO guidelines without the 5mm crop from skin. A leave-one-out (LOO) approach in each group was implemented in RayStation v6.99. Additionally, each LOO tool was applied to the other diagnostic group, generating 10 extra contours per patient (e.g. each lymphoma patient had 1 ABAClymph and 10 ABACbreast contours). ABAC performance was evaluated using mean distance to agreement (meanDTA) and Dice similarity coefficient (DSC). Metrics for the contour generated using the LOO approach (i.e. using the other 9 patients in the same group) were compared to the averaged metric of the 10 extra contours (each generated using 9 templates from the other group). Results Results are reported for left and right breasts combined (see table 1). Despite different age ranges, no significant difference in breast volume or average breast density was observed between the two diagnostic groups. ABACbreast applied to BC patients performed best; DSC median 0.89 (range 0.75-0.96), meanDTA 0.25cm (0.13-0.90) (see fig 1). Unexpectedly, when applied to lymphoma patients, ABAClymph did not appear to perform better than ABACbreast; DSC 0.86 (0.46-0.91) vs 0.86 (0.56-0.94), meanDTA 0.28cm (0.14-0.82) vs 0.24cm (0.18-0.65). We also observed poorer performance on the left breast than the right in all ABACs Conclusion We could not confirm our hypothesis that lymphoma-specific ABAC performs better than ABAC developed using BC patients. The disappointing performance of ABAC tools for breast contouring in young lymphoma patients suggests more sophisticated methods might be needed in this complex population, e.g. specific atlases to account for variations in arm position and different dense tissue distribution in the breast. We also found ESTRO guidelines need small adaptations to contour ‘breast as an OAR’, which needs further exploration. The different performance between right and left breasts may be due to differences in heart and lung shape and is the subject of ongoing work.