Testing of auto segmentation to improve the workflow for stereotactic radiosurgery

Auto segmentation (contouring) is used to improve the efficiency of the radiotherapy workflow. Stereotactic Radiosurgery (SRS) patients often require short patient pathways. The aim of this study was to compare 3 different commercial auto segmentation software packages for SRS. The first software package uses a single contour set from a previous patients CT. The second uses an atlas built from a selection of previous patients CT. The final uses a generic model of a patient and utilises MRI and CT images. These 3 auto segmented datasets, and the original contours that were used during treatment, were assessed by 1 consultant clinical oncologist and 3 medical physicists from the SRS team. 12 contours, across two patients were analysed and ranked on a 3-step scoring system; 1-clinically acceptable, 2-needs small amendments, 3-needs large amendments. DICE coefficients were also performed. All auto generated contour sets required some amendments. Some of the original contours drawn by the physician were ranked as requiring amendments by all reviewers. Two of the auto segmentation tools had a median score of 2 across all the reviewed contours. For the contours that were originally physician approved, the median score was 1, although some of the contours were rated as 2. Statistical analysis indicated no significant difference between software packages 2 and 3 when comparing all contours that were included in both systems. This study demonstrated that our implementation of software packages 2 & 3 could be used to improve the workflow in the department by creating autogenerated structures.