The utility of atlas-assisted segmentation in the male pelvis is dependent on the interobserver agreement of the structures segmented.

Abstract

To investigate the relationship between the ability of atlas-based autosegmentation to reduce outlining time in the male pelvis (body, bladder, rectum, femoral heads, prostate and seminal vesicles) and the interobserver agreement in the delineation of these structures. To examine any increase of the interobserver agreement with the use of an autosegmentation tool. We created atlases in the ABAS™ system v. 2.0 (Elekta, Crawley, UK) and recorded the time to delineate the above structures on eight patients with and without its aid. We also measured the interobserver agreement in the structure definitions using several metrics [Dice's similarity coefficient (DSC), mean distance to conformity, percentage volume difference] with and without the aid of ABAS. There is a high degree of correlation between the time saving with the use of ABAS and the degree of interobserver agreement (r = 0.90 for DSC). This indicates that for structures where the interobserver agreement is low (DSC < 0.65), the ABAS does not reduce outlining time. We found that the interobserver agreement is increased with ABAS only for the prostate. Outlining time saved in the male pelvis is highly correlated with the interobserver agreement of the structures. Only for the prostate does the use of ABAS significantly reduce the amount of interobserver variation in contouring. The use of autosegmentation software increases the outlining time for structures where the interobserver agreement is low. Any increase in the interobserver agreement in contouring with the aid of such software may be limited to those structures where there is currently mid-range agreement between observers.