Clinical Evaluation of Deep Learning Based Auto-Contouring Software for Prostate Radiotherapy
M Kirk1*, B Anderson1, H Prichard1, L Ryan1, X Zhang1, Y Wang2, (1) Mass General/North Shore, Danvers, MA,(2) Massachusetts General Hospital, Boston, MA
Presentations
PO-GePV-M-1 (Sunday, 7/25/2021) [Eastern Time (GMT-4)]
Purpose: Deep learning-based auto-contouring (DLC) may be used to reduce the time required to generate organs-at-risk (OAR). In this study, we investigate both the time and accuracy of the DLC OARs compared to manually drawn OARs for prostate radiotherapy.
Methods: The OARs of ten prostate patients were contoured by the DLC software and four experts (EC) (Senior dosimetrists with an average of 15 years of experience). The time to manually draw the Penile Bulb, Femurs, Bladder and Rectum was recorded. Additionally, the experts adjusted the DLC contours (User Adj DLC) and recorded the time. The Dice Similarity Coefficient (DSC) and the Mean Distance to Agreement (MDA) were used for comparisons.
Results: The use of DLC resulted in a statistically significant (p<0.05) reduction in contouring time, from a mean time for EC contours of 28 minutes (SD=10min) to a mean time of 10 minutes (SD=2min) for the User Adj DLC. The mean DSC values for the DLCs were: Penile Bulb = 0.62, Femurs = 0.94, Bladder = 0.94, and Rectum = 0.83. The DSC values for the mean User Adj DLC were: Penile Bulb = 0.69, Left and Right Femur = 0.96, Bladder = 0.94, and Rectum = 0.88.The mean MDA values for the DLCs were: Penile Bulb = 2.43mm, Femurs = 0.85mm, Bladder = 0.92 and Rectum = 2.31 mm. The MDA values for the mean User Adj DLC were: Penile Bulb = 1.90 mm, Left and Right Femur = 0.77 mm, Bladder = 0.81 mm, and Rectum = 1.57 mm.
Conclusion: The use of auto contouring provides a starting point for the creation of OARs allowing for a significant time reduction. User adjustment of the DLC based contours resulted in a significant increase (p<0.05) in DSC values and decrease in MDA values for all the OARs.
ePosters
Keywords
Not Applicable / None Entered.
Taxonomy
IM/TH- Image Segmentation Techniques: Machine Learning
Contact Email