Implementation and evaluation of various demons deformable image registration algorithms on a GPU

Abstract

Online adaptive radiation therapy (ART) promises the ability to deliver an 20 optimal treatment in response to daily patient anatomic variation. A major technical barrier for the clinical implementation of online ART is the requirement of rapid image segmentation. Deformable image registration (DIR) has been used as an automated segmentation method to transfer tumor/organ contours from the planning image to daily images. However, the current 25 computational time of DIR is insufficient for online ART. In this work, this issue is addressed by using computer graphics processing units (GPUs). A greyscale based DIR algorithm called demons and five of its variants were implemented on GPUs using the Compute Unified Device Architecture (CUDA) programming environment. The spatial accuracy of these algorithms was 30 evaluated over five sets of pulmonary 4DCT images with an average size of 256×256×100 and more than 1,100 expert-determined landmark point pairs each. For all the testing scenarios presented in this paper, the GPU-based DIR computation required around 7 to 11 seconds to yield an average 3D error ranging from 1.5 to 1.8 mm. It is interesting to find out that the original passive 35 force demons algorithms outperform subsequently proposed variants based on the combination of accuracy, efficiency, and ease of implementation. X. Gu et al.2

Deformable image registration of CT images for automatic contour propagation in radiation therapy

“…Demons is a process of optimizing the displacement of every pixel in the floating image to accurately construct point-to-point correspondence of the registered images, which can be mathematically represented by two terms of a similarity measurement and regularization of DVF: (2) where transforms the points of the moving image into the coordinate of the fixed image by the spatial transformation ; accounts for the noise on the image intensity; and controls the amount of necessary regularization. An image similarity term measures the similarity of the two registered images.…”

“…investigated four ITK based Demons variant algorithms and found that the performance of the symmetric Demons implemented by Vercautern [2] was advantageous over the original Demons and the other two Demons variants. The symmetric Demons defines the displacement field as follows: (1) where and respectively represent the moving image after the ith iteration and the fixed image; and denote the gradient image; and is the mean-squared value of the image voxel size.…”

An optimization algorithm for 3D real-time lung tumor tracking during arc therapy using kV projection images

“…Li et al 38 developed a 3D lung tumor localization method based on principal component analysis (PCA), where the accuracy of the tracking algorithm is limited by the accuracy of training deformable vector field derived using the demons algorithm. 44 Lewis et al 42, 43 developed a phase-binned tumor trajectory reconstruction method. The method can only provide one representative phase-binned tumor trajectory from a CBCT scan, however, with the aid of a breathing signal provided by an additional device.…”

Analysis of periodic anisotropic media by means of split-field FDTD method and GPU computing

“…Many authors are focusing their efforts in how to adapt the implementation of numerical methods under the CUDA compliance. Some examples can be found in literature for different topics such as physics 7,8 , astronomy 9, 10 , images [11][12][13] , biomedical applications 14,15 and computer science 16 . The acceleration of the SF-FDTD by means of GPU computing has been recently introduced by Shahmansouri 17 et al In their work GPU computing is applied to accelerate the SF-FDTD implementation for Drude-Lorentz dispersive media.…”