Deep learning predicts CyberKnife prostate stereotactic radiotherapy doses

U-Net model trained on 462 CyberKnife prostate plans predicted mean doses with absolute error of 0.63 Gy (rectum) and 1.04 Gy (bladder) on a 93-patient test set, showing promise for automating treatment planning.

• The U-Net-based convolutional neural network used CT images, delineated structures, and distance from the planning target volume to predict clinically achievable dose distributions.
• On the independent test set (n=93), mean absolute error between predicted and clinical mean doses was 0.63 Gy for the rectum and 1.04 Gy for the bladder.
• The model may provide patient-specific dose estimates for initial planning objectives, potentially improving inter-planner consistency.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.