FIGURE 3.

Schematic illustration of traditional WBRT field definition. (a) Digitally reconstructed radiograph (DRR) with clinically used beam aperture (green) and two‐dimensional projection of a patient's three‐dimensional body contour (blue). (b) Illustration of the inputs (DRR and contour) for deep learning‐based auto‐segmentation where the contour (yellow) is the intersection of the beam aperture and body contour projection. (c) Post‐processing step used to add skin flashing to the predicted mask (dashed yellow) to generate the final beam apertures (red). During testing, a value of 10 mm was chosen for skin flash for posterior, anterior, and cranial expansions (a value of 30 mm was used to highlight this step on this figure) based on clinical practice