ESTRO 2024 - Abstract Book

S3072

Physics - Autosegmentation

ESTRO 2024

implementation [4]. Changes in performance may occur due to changes in imaging protocols, patient population or changing observer bias. Here we reviewed autocontouring performance approximately 6 months after clinical implementation, focussing on agreement between clinician accepted (edited) contours and the original autocontours and investigating how much editing of autocontours was taking place.

Material/Methods:

The Cardiac Avoidance Area (CAA) chosen at our institution includes the right atrium, aortic valve root, and proximal portions of the left and right coronary arteries, based on the work of McWilliam et al. [5]. Since April 2023 this CAA has been included as an additional OAR for radiotherapy planning of non-SABR, radical lung treatments in the routine setting at our centre. Autocontours for the CAA are imported to the RT planning system where the clinician checks and manually edits as necessary prior to plan optimisation. This study includes 150 patients planned between April and September 2023, comparing the original CAA autocontour to the final, edited contour accepted by the clinician. Dice similarity coefficient (DSC) and mean surface distance (MSD) were used to quantify the agreement between the original and edited contours, to show the magnitude of editing of the autocontours. The percentage of contour points edited by > 2mm was also evaluated. In addition, similarity metrics were plotted as a function of time to investigate whether contour quality was consistent throughout the period. Contour similarity was also compared between clinicians to compare contouring practice between observers.

Results:

Average Dice Similarity Coefficient between original autocontours and clinician edited contours was 0.93 ± 0.07 (range 0.58-1.00) and average Mean Surface Distance was 1.1 ± 1.2 mm (range 0.0-5.73mm). On average 80% of points on the edited contour were within 2mm of the original autocontour. Figure 1 shows the DSC values for individual cases plotted against date of contouring. No time trends are apparent, and linear regression analysis shows the data are consistent with zero change over time (P=0.90).

Figure 1: DSC between auto and clinician edited contours by date.

Figure 2 shows mean and standard deviation of DSC between auto- and clinician edited contours for 12 clinicians. There are clear differences in how much editing is done by different clinicians. 8 of 12 clinicians had DSC between 0.90 and 0.95.