Sum Of Squared Intensity Differences Research Articles

Deformation vector fields (DVF)-driven image reconstruction for 4D-CBCT.

High quality 4D-CBCT can be obtained by deforming a planning CT (pCT), where the deformation vector fields (DVF) are estimated by matching the forward projections of pCT and 4D-CBCT projections. The matching metric used in the previous study is the sum of squared intensity differences (SSID). The scatter signal level in CBCT projections is much higher than pCT, the SSID metric may not lead to optimal DVF. To improve the DVF estimation accuracy, we develop a new matching metric that is less sensitive to the intensity level difference caused by the scatter signal. The negative logarithm of correlation coefficient (NLCC) is used as the matching metric. A non-linear conjugate gradient optimization algorithm is used to estimate the DVF. A 4D NCAT phantom and an anthropomorphic thoracic phantom were used to evaluate the NLCC-based algorithm. In the NCAT phantom study, the relative reconstruction error is reduced from 18.0% in SSID to 14.13% in NLCC. In the thoracic phantom study, the root mean square error of the tumor motion is reduced from 1.16 mm in SSID to 0.43 mm in NLCC. NLCC metric can improve the image reconstruction and motion estimation accuracy of DVF-driven image reconstruction for 4D-CBCT.

A Practical Approach Based on Analytic Deformable Algorithm for Scenic Image Registration

BackgroundImage registration is to produce an entire scene by aligning all the acquired image sequences. A registration algorithm is necessary to tolerance as much as possible for intensity and geometric variation among images. However, captured image views of real scene usually produce unexpected distortions. They are generally derived from the optic characteristics of image sensors or caused by the specific scenes and objects.Methods and FindingsAn analytic registration algorithm considering the deformation is proposed for scenic image applications in this study. After extracting important features by the wavelet-based edge correlation method, an analytic registration approach is then proposed to achieve deformable and accurate matching of point sets. Finally, the registration accuracy is further refined to obtain subpixel precision by a feature-based Levenberg-Marquardt (FLM) method. It converges evidently faster than most other methods because of its feature-based characteristic.ConclusionsWe validate the performance of proposed method by testing with synthetic and real image sequences acquired by a hand-held digital still camera (DSC) and in comparison with an optical flow-based motion technique in terms of the squared sum of intensity differences (SSD) and correlation coefficient (CC). The results indicate that the proposed method is satisfactory in the registration accuracy and quality of DSC images.

TU‐G‐141‐06: Deformation Vector Fields (DVF)‐Driven Image Reconstruction for 4D‐CBCT

Purpose: High‐quality four dimensional cone‐beam CT (4D‐CBCT) can be obtained by deforming the planning CT in radiotherapy, where the deformation vector fields (DVF) is estimated by matching the forward projection of the deformed planning CT and measured 4D‐CBCT projections. Due to the presence of scatter signal in CBCT projection images, the accuracy of DVF estimation is degraded when the sum of squared intensity difference (SSID) is used as the matching criterion. The goal of this work is to improve the estimation accuracy of DVF by using normalized correlation coefficient (NCC) as the matching criterion to obtain high quality on‐treatment 4D‐CBCT of lung cancer patients in radiation therapy. Methods: The DVF used to deform the planning CT is estimated by maximizing the NCC between the projections of on‐treatment 4D‐CBCT and the forward projection of the deformed planning CT. A non‐linear conjugate gradient (NLCG) algorithm was used for the optimization. To obtain better initial DVF for NLCG optimization, demons registration was first performed between planning CT and 4D‐CBCT reconstructed by total variation (TV) minimization. A 4D NCAT phantom was used to quantitatively evaluate the performance of the algorithm, where the reconstruction error is calculated as the sum of squared difference between the reconstructed 4D‐CBCT and the phantom image. Results: NCC based DVF estimation improves the image reconstruction accuracy as compared to the results obtained using SSID criterion. The image reconstruction error of 4D‐CBCT obtained by NCC criterion is 61.5% smaller than that is obtained by SSID criterion. Conclusion: The accuracy DVF estimation and image reconstruction is substantially improved by using the NCC criterion, as compared to the results obtained by SSID criterion. High quality 4D‐CBCT can be a valuable tool in image‐guided radiation therapy for lung cancer patients. Cancer Prevention and Research Institute of Texas (RP110562)

GPU parallel implementation of B-spline non-rigid grid registration using free-form deformations

Image registration is a vital problem in medical imaging. Fast deformable registration methods are needed especially in intra-operative settings characterised by their time-critical applications. In this paper, we present our work on B-spline non-rigid registration using free-form deformations. Our first implementation which was serial on standard desktop machine showed the algorithm is computed intensive and not real time; even multithread implementation in CPU was not real time because of high execution time. We have broken the registration process in threads and blocks with the help of Compute Unified Device Architecture (CUDA) for handling GPU. Our work employs extensive usage of highly multithreaded architecture of multicored GPU. An efficient use of shared memory is required to optimise parallel reduction in CUDA. Experimental results show that parallel implementation achieved an average speed-up of 82x over the serial implementation when run on a GPU GeForce 9500 GT having 32 cores and around 56x over multicore architecture CPU. Registration method is also being evaluated using SSD (Squared Sum of Intensity Differences) and CC (Cross Correlation).

Registration Accuracy and Quality of Real-Life Images

BackgroundA common registration problem for the application of consumer device is to align all the acquired image sequences into a complete scene. Image alignment requires a registration algorithm that will compensate as much as possible for geometric variability among images. However, images captured views from a real scene usually produce different distortions. Some are derived from the optic characteristics of image sensors, and others are caused by the specific scenes and objects.Methodology/Principal FindingsAn image registration algorithm considering the perspective projection is proposed for the application of consumer devices in this study. It exploits a multiresolution wavelet-based method to extract significant features. An analytic differential approach is then proposed to achieve fast convergence of point matching. Finally, the registration accuracy is further refined to obtain subpixel precision by a feature-based modified Levenberg-Marquardt method. Due to its feature-based and nonlinear characteristic, it converges considerably faster than most other methods. In addition, vignette compensation and color difference adjustment are also performed to further improve the quality of registration results.Conclusions/SignificanceThe performance of the proposed method is evaluated by testing the synthetic and real images acquired by a hand-held digital still camera and in comparison with two registration techniques in terms of the squared sum of intensity differences (SSD) and correlation coefficient (CC). The results indicate that the proposed method is promising in registration accuracy and quality, which are statistically significantly better than other two approaches.

Elastic models application for thorax image registration

This work consist of the implementation and evaluation of elastic alignment algorithms of biomedical images, which were taken at thorax level and simulated with the 4D NCAT digital phantom. Radial Basis Functions spatial transformations (RBF), a kind of spline, which allows carrying out not only global rigid deformations but also local elastic ones were applied, using a point-matching method. The applied functions were: Thin Plate Spline (TPS), Multiquadric (MQ) Gaussian and B-Spline, which were evaluated and compared by means of calculating the Target Registration Error and similarity measures between the registered images (the squared sum of intensity differences (SSD) and correlation coefficient (CC)). In order to value the user incurred error in the point-matching and segmentation tasks, two algorithms were also designed that calculate the Fiduciary Localization Error. TPS and MQ were demonstrated to have better performance than the others. It was proved RBF represent an adequate model for approximating the thorax deformable behaviour. Validation algorithms showed the user error was not significant.