| Lung four-dimensional computed tomography plays an important role in lung cancer radiotherapy, because it provides breathing movement information which is necessary for precise radiotherapy in lung cancer treatment. Lung 4D-CT data acquisition is usually a collection of CT images over the entire respiratory cycle, then CT images are analyzed retrospectively combined with respiratory signal by synchronous acquisition and sorted according to the timeline, finally reconstructed to 3d images. Lung 4D-CT data can not only represent the spatial structure and movement information, but also reduce artifacts caused by respiratory motion. With those advantages, radiation therapy guided by lung 4D-CT data allows high dose radiation in tumors, at the same time decreases the radiation dose to normal tissue. Based on movement information, physicists can design personal radiotherapy for patients which has great values for precise radiotherapy in moving targets. Besides, compared to normally used breath control techniques(such as breath holding techniques, abdominal pressure techniques, breathing gating techniques), lung 4D-CT technique is scanning in the condition of free breathing and has lower requirements for patients’ health so that the patient’s tolerance is strong. However, lung 4D-CT technique also has some shortcomings. First, in order to acquire lung 4D-CT data, scanner needs to collect images at different breathing phase which greatly prolonged scanning time. As a consequence, patients would receive dose several time higher than normal CT scanning. Second, lung 4D-CT data is big and one set lung 4D-CT data usually contains 1000～3000 images. The task of target sketching is heavy and excessive labor will affect doctor’s judgement to the patient’s condition. Third, lung 4D-CT data collection needs to use helical CT scanner or CT scanner in cine mode and heavily relies on respiration monitoring system. It has high requests on CT equipment and can’t carry out on normal CT in which way limiting its promotion to small-scale hospitals.To solve above questions in lung 4D-CT practical application, this paper propose a registration based method to reconstruct Images at Intermediate phases of lung 4D-CT data. The proposed method fully uses breathing movement information contained by lung 4D-CT data:first use registration method to estimate the movement between two different phases, then reconstruct immediate phases based on the assumption of lung respiratory movement model. The proposed method only needs to collect a few phases using normal CT scanner to reconstruct intermediate phases instead of collecting all phase binned images in at least one breathing cycle. As a result, the proposed method shorts the CT scanning time and restricts 4D-CT radiation dose. Based on the above idea, we develop the following research.The paper applies Active Demons rigid registration method to estimate the sub-pixel movement of two phases. Based on the movement estimation, reconstruct the intermediate phases according to lung breathing linear motion hypothesis. Then we use lung 4D-CT data to do experiments to validate the effectiveness of the method. More details are as follows:(1) First, we introduce the Active Demons algorithm and adopt the algorithm to estimate motion displacement between the two specific phase images. Inspired by Maxwell Demons thermodynamics theory and optical flow method, Thirion proposed Demons algorithm based on partial differential equation. But Demons algorithm may get wrong registration results when reference image gradient information is not sufficient. To improve the robustness of the algorithm, Wang drew into floating image gradient information and further proposed Active Demons algorithm. The experiment results show that the algorithm can accurately estimate the displacement field between the two different phase images and prepare for the immediate phases image reconstruction in next step.(2) Based on the motion field, we can compute interpolating coefficients by comparing the phase correlation of image already known and to be reconstructed, then reconstruct immediate phase images according to lung linear motion hypothesis. Lung 4D-CT data contains multiple phase 3D-CT images, covering the inhale process and the exhale process. Lung respiratory movement is very complex, the movement of heart and diaphragm will cause certain influence to it, so the actual lung breathing process is difficult to use model to accurately simulate. In this problem, all phases can be seen as an image linearly deformed by extreme inhale phase. Therefore, we adopt an assumption that lung movement linearly change in the process of expiratory. So after obtaining the movement vector between extreme inhale phase and extreme exhale phase, we can use time continuity of lung 4D-CT data to reconstruct other phases. The experiment results show that the reconstructed intermediate phase images are very close to the real images, so the proposed method could accurately reconstruct intermediate phase images.The proposed method is based on registration, so the accuracy of registration algorithm directly affects the quality of the reconstruction of the middle phase images. The problem of Active Demons algorithm is that optical flow theory has an assumption that lung image intensity does not change with time during the breathing process, but the lung density changed due to the effect of inspired air, in which case intensity constant hypothesis is inapplicable. This effect will have an impact on registration accuracy, especially in the corresponding areas whose intensity difference is big. To solve this problem, intensity correction is an effective preprocessing step. But it’s not easy to get the full intensity correction field. This paper uses block matching based registration strategy, then implement of lung image registration through image blocking, block matching of sample points, full deformation field computing. The main advantage of blocking is:for the corresponding image block, intensity correction is easier to realize. So we further study an outlier rejection scheme to improve registration accuracy. Based on registration deformation field, we can reconstruct intermediate phases. The detail process is as follows:(1) First, select voxel points with certain distance and we can get some blocks with those voxels as the center point. Then we use full search block matching algorithm to estimate the movement vector of those points after intensity correction. Finally, we can get a sparse deformation filed. Full search block matching algorithm search the most similar patch in reference image with the target block according to certain matching principle. The motion vector can be computed from the position of target block and matching block. The algorithm is simple and fast. So it was widely used.(2) In step (1), we only get movement vectors of part voxels. To get full deformation field, Gaussian convolution is employed to interpolate the original sparse deformation field. The standard deviation σ allows to compute a smoother or shaper deformation field. It is very important to choose a proper value for σ.(3) Because of the shortcomings of full search block matching algorithm, there may be some error estimation in original deformation filed. Those errors can be enlarged through Gaussian convolution and reduce the accuracy of motion estimation. Therefore, we employ an outlier rejection scheme to reject error estimation in original deformation field and keep the right deformation field.(4) Repeat step (2) and step (3), then get the final registration results. Based on the deformation field, we reconstruct intermediate phase images according to lung motion assumption. The proposed method is easy to achieve and stable. The reconstruction accuracy is higher than the former one.We select a public available lung 4D-CT data as the experimental data which is provided by DIR laboratory at Anderson university of Texas. The data contains 10 phases of 3D-CT images, collected from a complete respiratory cycle. During the breathing cycle, the condition of extreme exhale (inhale) phase is relatively stable and their images have little artifacts, so we choose extreme inhale phase and extreme exhale phase as benchmark data to reconstruct other phases. What’s more, lung 4D-CT data set contains 75 landmark points selected by experts on 0-5 phase. We can use landmark point error to evaluate our reconstruction method.The experiment results showed that the reconstruction method based on Active Demons algorithm can obtain intermediate phase images similar with true images. The mean errors of landmark points are within 3 mm and the errors of 15 typical landmark points are within 2 mm. Besides, we compared the reconstruction method based on Active Demons with the reconstruction method based on block matching algorithm with outlier rejection. Reconstruction results is more accurate in some regions and the mean errors of landmark points are smaller when using the latter one. Both the visual and quantitative results showed that the proposed methods could accurately reconstruct intermediate phase images and effectively reduce patient radiation dose. |
PDF Full Text Request