An adversarial machine-learning-based approach and biomechanically guided validation for improving deformable image registration accuracy between a planning CT and cone-beam CT for adaptive prostate radiotherapy applications.