1 Oncology, Department of Clinical Research, Det Sundhedsvidenskabelige Fakultet, SDU2 Oncology, Department of Clinical Research, Det Sundhedsvidenskabelige Fakultet, SDU
Background. The aim of this study was to evaluate two fully automatic segmentation methods in comparison with manual delineations for their use in delineating the heart on planning computed tomography (CT) used in radiotherapy for breast cancer. Material and methods. Automatic delineation of heart in 15 breast cancer patients was performed by two different automatic delineation systems. Analysis of accuracy and precision of the differences between manual and automatic delineations were evaluated on volume, mean dose, maximum dose and spatial distance differences. Two sets of manual delineations were used in the evaluation: 1) a set prior to common delineation guidelines; and 2) a second set repeated with a common set of guidelines. Results. Systematic differences between automatic and manual delineations were small for volume as well as dose. The uncertainty of the difference in volume was smaller than or similar to the inter-observer variation in manual delineations. For dose, the uncertainty was similar to manual delineations performed without common guidelines but slightly higher than the variation in manual delineations with common guidelines. Spatial differences between average manual and automatic delineations were largest at the base of the heart, where also large variations are observed in the manual delineations. Both algorithms could be improved slightly at the apex of the heart where the variation of automatic delineation was larger than for the manual delineations. Conclusion. Automatic delineation is an equal alternative to manual delineation when compared to the inter-observer variation. The reduction in precision of measured dose was small compared to other uncertainties affecting the estimated heart dose and would for most applications be outweighed by the benefits of fully automated delineations.
Acta Oncologica, 2014, Vol 53, Issue 10, p. 1366-1372