Deformable Image Registration - A Critical Evaluation: Demons, B-Spline FFD and Spring Mass System

Abstract

This paper describes results of a quantitative evaluation of a flexible spring mass system image registration technique previously proposed by the authors. The method is assessed against two well-known registration algorithms namely the Demons and the B-spline free form deformations (FFD) implemented in Insight Segmentation and Registration Toolkit (ITK). The evaluation has been performed using simulated data as well as real CT images of the radiotherapy prostate and the head and neck patients. Whereas for the simulated data the quality of registration has been measured using the dense displacement field, the discrete anatomical landmarks have been used with the real CT images. The results show the method using spring mass system achieves comparable registration accuracy with the Demons and B-spline FFD for the data with no noise or Gaussian noise, but it outperforms these methods when structured noise is present in the data. Moreover, the method using spring mass system can offer more accurate registration quality if some additional information, in the form of feature landmarks and/or segmented anatomical structures, is available. Throughout the paper a special attention has been given to the effective visualisation of the results.