An ITK-based Implementation of the Stochastic Rank Correlation (SRC) Metric

Abstract

Recently, Birkfellner et al. proposed a novel image-to-image merit function (stochastic rank correlation, SRC) for robust intensity-based 2D/3D image registration. In this work, we summarize the basic idea of SRC, and present a generic ITK-based implementation of this image-to-image metric including tests for software verification. Moreover, we provide two simple examples that demonstrate the usage of this metric: a) within the native ITK 2D/3D image registration method, and b) within a recently published extended ITK-based 2D/3D registration framework. It is, however, important to note, that this paper neither covers a comprehensive evaluation of SRC, nor a comparison with other metrics. It rather shows that SRC appears to succeed on a femoral and a porcine data set in the course of ITK-based 2D/3D image registration.

Influence of imaging source and panel position uncertainties on the accuracy of 2D/3D image registration of cranial images

“…5 As an alternative, intensity-based 2D/3D image registration has been proposed by several groups as comprehensively reviewed by Markelj et al 6 As opposed to 2D/3D slice-tovolume image registration, 7 where a 3D computed tomography (CT) is aligned with extracted 2D slices of a volumetric image, this work focuses on projection-based registration of a 3D CT to 2D kilovoltage (kV) x-rays utilizing intensitybased similarity measures for assessing image alignment. [8][9][10][11][12][13][14] This method generally requires a small number of, ideally simultaneously, acquired kV x-ray images from different viewing perspectives as intratreatment patient pose measurement. Besides delivering a relatively small amount of dose to the patient, the method does not require additional image reconstruction time and might, therefore, allow for comparably fast patient setup procedures as exemplarily shown in Ref.…”

“…The rigid transformations required to correctly position the patient on the treatment couch are based on the intensity differences between the x-ray images and corresponding, iteratively recomputed DRRs from the reference CT. Recently, Steininger et al [8][9][10] proposed an algorithm for graphics processing unit (GPU) accelerated 2D/3D image registration which is included in the open source toolkit plastimatch 15 as Reg23 module. Basically, Reg23 is generically configurable and easily adaptable, allowing it to tackle a wide range of concrete 2D/3D registration issues.…”

Deformable registration of lateral cephalogram and cone‐beam computed tomography image

“…The publicly available Elastix‐based registration toolkit 54 provides the rigid 2D–3D registration with 6‐DOFs. The rigid 2D–3D registration by the Elastix takes less than 1 min, the same as the Inten‐MS 32 for the rigid transformation. The nonrigid 2D–3D registration of the Inten‐MS is implemented by adopting the nonrigid registration toolkit, 55 and takes 960 s to optimize B‐spline‐based deformation parameters.…”

“…For instance, the public Elastix 54 provides the rigid 2D–3D registration with 6‐DOFs. In experiments, we adapt the optimization‐based rigid 2D–3D registration 3,32 using the B‐spline‐based nonrigid transformation model 55 . We compared with the one‐shot CNN‐based regression models with short and long residual connections 24,50 .…”

“…Liao et al 31 trained the agent via greedy deep supervised learning for the 3D rigid‐body registration. The above 2D–3D registration mostly handled the rigid transformation of 3D rotation and translation 3,4,32–34 . Considering the nonrigid structure deformation caused by treatments and growth in the longitudinal orthodontic treatments, the rigid 2D–3D registration could not construct reasonable 3D anatomical structures.…”