MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

The Correlation Between Treatment Plan Dosimetry Metrics and the DICE Similarity Coefficient and Target Displacement, in a Prostate Cancer Treatment Planning Study

D Wang*, W Smith , M Phillips , University of Washington, Seattle, WA

Presentations

(Monday, 7/15/2019) 3:45 PM - 4:15 PM

Room: Exhibit Hall | Forum 7

Purpose: DICE similarity coefficient has been commonly used to evaluate the similarity between two contours. We have previously found that the DICE had a linear correlation with target dosimetry metrics, but not strongly correlated with OAR dosimetry metrics from treatment plans developed on autocontours. We introduce autocontour target displacement, which is defined as the distance change of target (i.e. PTV) centroid to OAR centroid (for bladder) or OAR central axis (for rectum). We evaluate the suitability of these metrics for auto-contouring of anatomy by examining how changes in the contours affect the developed treatment plans.

Methods: A machine learning (ML) algorithm based on decision forests was used to contour prostate GTV, seminal vesicles, bladder, rectum and femurs. 24 subjects were contoured by both physicians and ML software. We then generated additional contours needed for planning, such as PTV expansions. The plans based on the physician contours and autocontours were optimized using the same plan geometry and optimization objectives. Dosimetry metrics for both plans were calculated based on the physician’s original ‘gold standard’ contours. GTV-V100%, PTV-V100%/V95%, Rectal Wall-V70Gy/V50Gy, Rectal Volume-V70Gy, and Bladder Wall-V75Gy were used to evaluate the dose distributions of both plans and study their correlations with PTV/GTV DICE and displacements.

Results: The target DICE and PTV/GTV metrics have a good linear correlation. The OAR metrics have linear correlation with PTV DICE and PTV displacement.

Conclusion: The DICE is suitable to evaluate ML autocontour performance, since it predicts target coverage due to the high degree of target conformality in the accepted treatment plans. The OAR dosimetry metrics can be predicted by target DICE and displacements and may increase ML autocontouring performance if introduced at the learning stage. These metrics have the potential to aid in assessing the necessity of adaptive replanning by predicting the dosimetry metrics changes on adjusted contours.

Keywords

Dosimetry, Treatment Planning, Segmentation

Taxonomy

TH- External beam- photons: treatment planning/virtual clinical studies

Contact Email