Normalized Distance Discordance (DD): A Quantitative and Position-Specific Metric to Evaluate the Uncertainty of Inter- and Intra-Patient Deformable Registrations
Z Saleh1*, A Apte1, G Sharp2, y wang1, J Deasy1, (1) Memorial Sloan Kettering Cancer Center, Manhattan, NY, (2) Massachusetts General Hospital, Boston, MA,SU-C-WAB-7 Sunday 1:00PM - 1:55PM Room: Wabash Ballroom
Quantitative tools to estimate the uncertainty in deformable image registration are lacking. Distance discordance(DD) is a quantity that we have defined, and is an estimate of the mean error (measured in terms of distance) due to deformable image registration. This abstract presents a refined method of estimating distance discordance that improves accuracy.
Starting from an arbitrary reference image [i] in a set of images [1...n], voxels are traced from any two moving images [j] and [k] that happened to be registered at the same voxel on [i] to other reference images in the image set. The difference between each pair of points is recorded in the Cartesian coordinate system (dx,dy,dz). The mean of the difference (
The 3D spatial map of the mean DD metric showed a variation among anatomical sites in the head and neck region. Regions of high contrast (bones) showed a lower mean DD value (2-13 mm) while regions of low contrast (soft tissue) had a larger mean DD value (6-19 mm). The shape of the dose discordance histogram (DDH) for some anatomical structures followed a log normal distribution with a long tail while others had irregular shape.
The updated DD metric gives an unbiased measure of the spatial uncertainty in image registration in the absence of a ground truth. For instance, the DD maps can be used to set a level of confidence for dose accumulation in adaptive radiotherapy. However, the DD metric provides a necessary but not a sufficient condition to ensure good registration.
Funding Support, Disclosures, and Conflict of Interest: Partially supported by NIH grant R01 CA85181 download free
Add this talk to vcal | ical | Contact Email: