Research On Lung Tumor Localization Based On 3D/2D Registration Of Medical Images

A tumor localization method based on 3D/2D registration of preoperative CT images and intraoperative dynamic X-ray images is proposed to solve such two problems that the current CT-based lung tumor puncture and ablation procedures with low positioning accuracy,inability to compensate patients for lung breathing during surgery.Our proposed method is an improvement on the original method.After the accurate localization information of the lung tumor was obtained from the CT,the intraoperative 3D and 2D position information of the lung tumor were calculated by CT and X-ray registration.Our method is divided into the following three steps,first use 3D/2D registration to align the CT and X-rays,then achieve static localization of tumors in X-rays by fuse the lung tumor information from CT to X-rays through respiratory phase discrimination,finally,we achieved the dynamic localization of lung tumors in X-rays by 3D/2D registration of lung tumor regions.We improve the algorithm according to the actual requirements of the surgery while completing the method we proposed.First,we propose the algorithm test data set and experimental verification data set related to the task of this paper,study the 3D/2D registration algorithm components suitable for lung images,and realize the whole and local 3D/2D registration of lung image under breathing motion;Then,we improve the digital reconstructed radiography algorithm to achieve a high-resolution digital reconstructed image generation rate of 50 frames per second,which is 200 times faster than the traditional algorithm.We therefore achieve the overall contour alignment of 4s and the spine alignment of 12s in the overall registration of lung images.In the dynamic registration of lung tumors,we achieve real-time localization with an average time of 0.87s and an average error of 1.41mm;At last,we propose an innovative breathing phase discrimination method based on similarity measures,which can quickly and accurately match pre-operative CT images and intra-operative X-ray images.