SU‐E‐T‐466: A New Evaluation Method of Deformable Image Registration Algorithms for Image‐Guided Radiation Therapy

Purpose: Non‐rigid image registration (NIR) is an essential image processing tool for image‐guided adapted radiation therapy. The current radiotherapy process requires three‐dimensional (3D) quantification of the registration error, which is not accurately achieved by existing evaluation methods. The objective of this study is to develop a method for 3D evaluation of NIR algorithms. Methods: Cone‐beam (reference) and planning (moving) CT scan images of five prostate cancer patients were scanned using the VelocityAI (Velocity Medical Solutions) platform, which employs B‐spline‐based NIR algorithms. The NIR algorithms were evaluated by measuring the distance between two factors: (1) the outlines (reference contours) of the bladder, the dorsal right and left seminal vesicles (SV), the proximal SV, the prostate, and the rectum in the reference images; and (2) other outlines (deformed contours) in the deformed images of the same CT slice. This distance is the difference between two points that indicate the point of intersection of the reference and the deformed contours that cross a straight line every 10° from the center of gravity of the reference contour. However, the difference in direction was determined as being negative if it was inside the reference contours for the deformed contours, and positive if it was outside the reference contours for the deformed contours. The measurement values were displayed on a two‐dimension (2D) difference map. Results: The 2D map showing differences in the prostate indicated an error of −3 to +3 mm on the bladder side, −3 to +3 mm on the proximal SV side, −2 to +8 mm on the rectum side, and −5 to +1 mm on the apex side. Conclusion: This method measures the distance and the difference in direction between the contours in the reference and those in deformable images, and it can be used to accurately evaluate the non‐rigid image registration.