Non-rigid 2D-3D Registration Based on Support Vector Regression Estimated Similarity Metric
In this paper, we proposed a novel non-rigid 2D-3D registration framework, which is based on Support Vector Regression (SVR) to compensate the disadvantages of generating large amounts of Digitally Rendered Radiographs (DRRs) in the stage of intra-operati
- PDF / 568,538 Bytes
- 10 Pages / 430 x 660 pts Page_size
- 49 Downloads / 199 Views
Computer Science, Shanghai Jiaotong University, 800 Dongchuan Road, Shanghai 200240, China 2 Shanghai Renji Hospital, Shanghai {jimmyqwy,gu-lx}@sjtu.edu.cn
Abstract. In this paper, we proposed a novel non-rigid 2D-3D registration framework, which is based on Support Vector Regression (SVR) to compensate the disadvantages of generating large amounts of Digitally Rendered Radiographs (DRRs) in the stage of intra-operation for radiotherapy. It is successfully used to estimate similarity metric distribution from prior sparse target metric values against different featured transforming parameters of non-rigid registration. Through applying the appropriate selected features and kernel of SVR solution to our registration framework, experiments provide a precise registration result efficiently in order to assist radiologists locating the accurate positions of radiation surgery. Meanwhile, a medical diagnosis database is also built up to reduce the therapy cost and accelerate the procedure of radiotherapy in the case of future scheduling of multiple treatments. Keywords: 2D-3D Registration, Non-rigid, Support Vector Regression, DRR, Registration Framework, Radiotherapy.
1 Introduction Nowadays, non-rigid registration [1] algorithm is widely employed into many kinds of modern surgery, diagnosis and operation planning in order to combine and enhance the information of two or several different modality data sets at different times. Especially, in the field of radiation surgery [2], most radiologists traditionally diagnose diseases through viewing 2D X-ray film only. It is very hard for a radiologist to imagine the complex 3D shapes of tissue or organ various from different patients and difficult for them to locate the surgical position accurately. To this point, during radiotherapy, we should introduce information of a 3D model reconstructed from pre-operative data obtaining by CT or MRI machine into 2D X-ray image to aid radiologists to diagnose various diseases and locate the surgical position easily and accurately in real time [3]. Because many surgical objects are soft tissues, we have to develop the non-rigid registration to reach the above goal. ∗
Corresponding authors.
T. Dohi, I. Sakuma, and H. Liao (Eds.): MIAR 2008, LNCS 5128, pp. 339–348, 2008. © Springer-Verlag Berlin Heidelberg 2008
340
W. Qi, L. Gu, and J. Xu
Our target is to utilize an effective 2D-3D registration algorithm [4,5] to decide the physical space position of 3D model for matching the intra-operative 2D X-ray image with deformation as accurately as possible. Scholars had engaged into developing many effective, highly evaluated, deeply proved and widely used cutting-edge non-rigid 2D-3D registration algorithm. Few of them involve the area of matching the intensities between 3D data sets and X-ray images in elasticity deformation by minimizing a similarity measure to reach the goal of registration. On the other hand, Voxel-based registration [4] had been widely used for its simplicity and robustness. As the key technology in this kind of 2D-3D reg
Data Loading...
