Evaluation of various deformable image registrations for point and volume variations

Abstract

The accuracy of deformable image registration (DIR) has a significant dosimetric impact in radiationtreatment planning. Many groups have studied the accuracy of DIR. In this study, we evaluatedthe accuracy of various DIR algorithms by using variations of the deformation point and volume.The reference image (Iref) and volume (Vref) were first generated by using virtual deformation QAsoftware (ImSimQA, Oncology System Limited, UK). We deformed Iref with axial movement of thedeformation point and Vref, depending on the type of deformation (relaxation and contraction) inImSimQA software. The deformed image (Idef) and volume (Vdef) acquired by using the ImSimQAsoftware were inversely deformed relative to Iref and Vref by using DIR algorithms. As a result,we acquired a deformed image (Iid) from Idef and volume (Vid) from Vref. Four intensity-basedalgorithms were tested by following the horn-schunk optical flow (HS), iterative optical flow (IOF),modified demons (MD) and fast demons (FD) with the Deformable Image Registration and AdaptiveRadiotherapy Toolkit (DIRART) of MATLAB. The image similarity between Iref and Iid wascalculated to evaluate the accuracy of DIR algorithms using by Normalized Mutual Information(NMI) and Normalized Cross Correlation (NCC) metrics, when the distance of point deformationwas moved 4 mm, the value of NMI was above 1.81 and that of NCC was above 0.99 in all DIRalgorithms. As the degree of deformation was increased, the degree of image similarity decreased.When the Vref was increased or decreased by about 12%, the difference between Vref and Vid waswithin ±5% regardless of the type of deformation, the deformation was classified into two types:deformation 1 increased the Vref (relaxation) and deformation 2 decreased the Vref (contraction).The value of the Dice Similarity Coefficient (DSC) was above 0.95 in deformation 1 except for theMD algorithm. In the case of deformation 2, the value of the DSC was above 0.95 in all DIR algorithms.The Idef and the Vdef were not completely restored to Iref and Vref, and the accuracy ofthe DIR algorithms were different, depending on the degree of deformation. Hence, the performanceof DIR algorithms should be verified for the desired applications