LAUSR

Simple Summary When treating patients with head-and-neck cancer (HNC), in addition to the primary tumour, commonly involved lymph node (LN) levels are often electively irradiated. This requires the definition of the elective LN target volume. Because the LN levels that will be included in the target depend on the clinical situation, and because manual contouring is a laborious task that can also introduce inter- and intra-observer variation, being able to automate the segmentation of individual LN levels would reduce the clinical burden and would allow use of contours regardless of the primary tumor location. We trained and evaluated three patch- and/or voxel-based deep learning frameworks to segment elective LN levels. Our results suggest that accurate segmentations can be obtained using an ensemble of patch-based UNets and that this result can be further refined by sequentially applying a 2.5D, multi-view voxel classification network. Abstract Depending on the clinical situation, different combinations of lymph node (LN) levels define the elective LN target volume in head-and-neck cancer (HNC) radiotherapy. The accurate auto-contouring of individual LN levels could reduce the burden and variability of manual segmentation and be used regardless of the primary tumor location. We evaluated three deep learning approaches for the segmenting individual LN levels I–V, which were manually contoured on CT scans from 70 HNC patients. The networks were trained and evaluated using five-fold cross-validation and ensemble learning for 60 patients with (1) 3D patch-based UNets, (2) multi-view (MV) voxel classification networks and (3) sequential UNet+MV. The performances were evaluated using Dice similarity coefficients (DSC) for automated and manual segmentations for individual levels, and the planning target volumes were extrapolated from the combined levels I–V and II–IV, both for the cross-validation and for an independent test set of 10 patients. The median DSC were 0.80, 0.66 and 0.82 for UNet, MV and UNet+MV, respectively. Overall, UNet+MV significantly (p < 0.0001) outperformed other arrangements and yielded DSC = 0.87, 0.85, 0.86, 0.82, 0.77, 0.77 for the combined and individual level I–V structures, respectively. Both PTVs were also significantly (p < 0.0001) more accurate with UNet+MV, with DSC = 0.91 and 0.90, respectively. The accurate segmentation of individual LN levels I–V can be achieved using an ensemble of UNets. UNet+MV can further refine this result.