Target volume delineation for adaptive treatment in HNSCC is highly variable among experts

Purpose or Objective
Inter-observer variability (IOV) in target volume delineation is a well-documented phenomenon and a major source of uncertainty in radiation treatment (RT) planning. The increasing adoption of adaptive RT adds a dynamic component to IOV, which is largely unknown. We analysed IOV in the pre- and mid-treatment (PT and MT) setting using expert primary gross tumour volume (GTV) and clinical target volume (CTV) delineations in locally advanced HNSCC.

Material and Methods
Five patients from routine clinical practice, who underwent repeat imaging, were selected such that a variety of features likely to prompt adaptation were included. Brief case reports and CT imaging data were sent to five observers, the latter supplemented with pre-therapeutic FDG-PET in four cases. Observers received PT and MT imaging data at least one week apart and were asked to delineate the GTV and CTV in their own treatment planning system and to comment on their delineation process, i.e. on how GTV to CTV expansion and adaptation were performed. Delineations were rasterised on a 1mm³ grid and their compatibility assessed with the Generalised Conformity Index (CIgen). Differences in IOV between PT and MT were probed with Wilcoxon signed-rank tests and the correlation between GTV and CTV IOV evolution with Spearman rank correlation. While it is respectively impossible and very difficult for these tests to show two-sided α=0.05 significance at n=5, more powerful parametric alternatives cannot credibly be employed. Delineations were processed with ITK and statistical analyses performed in R.

Results
A total of 82/100 delineations were received and analysed. Figure 1 shows an overview of IOV in terms of volume overlap. All cases and volumes suffered a reduction of CIgen when transitioning from PT to MT (p=0.063 for both CTVs and GTVs). There was generally better agreement in CTV than GTV delineations at individual time points, and the correlation between GTV and CTV CIgen changes was very weak (ρ=0.5, p=0.45). This comparative robustness of CTV delineations might stem from GTV to CTV expansion practices, which overwhelmingly employed sizeable isotropic margins (mean: 7.6 mm) and additional editing for anatomical boundaries at both time points. Figure 2 shows a case in which CIgen for CTVs remains stable despite deteriorating IOV in the primary GTV. MT GTV delineations were often based on PT delineations after image registration (14/24 analysed observers and cases), the remainder contoured de novo. Post-transfer adaptation to changed anatomy was performed by two observers and inclusion of the PT GTV in the MT CTV by one. Shortcomings of this contouring challenge are a lack of MRI and contrast-enhanced CT at both time points.

Conclusion
IOV in target volume delineation increases during treatment, where a disparity in institutional adaptation practices adds to the static causes of IOV. Consensus guidelines are urgently needed and should recommend scope, frequency, and quality of MT imaging alongside adaptation strategies.