Fig. 1.

Schematic overview of the contour sets considered in this study. A: Sub-sampled DL-contours; B: Manually adjusted DL-contours; C: Manually adjusted DL-contours followed by interpolation; D: Final clinical contour following adjustment of C. In this schematic figure, we assumed no edits were made to the blue contours from C, but in clinical practice it is possible some minor edits are made on these slices after interpolation; E: Interpolated DL-contours for evaluation purposes, for subsampled OARs only to simulate a full DL-contour; F: Subsampling of the clinical contour set (D) at the location of the original DL-contours for analysis, to catch also edits made in the second round of adjusting. This means F is not necessarily the same as B, however only minor differences are expected; G: Interpolation of F to infer edits only on the interpolated contours. Note: amount of edits are computed as signed Euclidean distances between the contour surfaces (in 3D). Abbreviations: DL = Deep-learning, RTSS = RT Structure Set. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)