2D-3D medical image parallel registration method based on combination similarity measure

Abstract

The invention discloses a 2D-3D medical image parallel registration method based on combination similarity measure. The method comprises the following steps: firstly, using a CUDA (Compute Unified Device Architecture) parallel computing model to finish a quick generation process of a DRR (Digitally Reconstructed Radiograph) image; combining a SAD (Sum of Absolute Difference) with PI (pattern intensity) as new similarity measure to carry out parallel computation on GPU (Graphics Processing Unit); and finally, transferring a combination similarity measure value to CPU (Central Processing Unit), and adopting a fruit fly optimization algorithm based on bacterial chemotaxis behaviors to optimize for looking for an optimal registration parameter. An experiment verifies the performance of the method to show that the execution speed of the method is effectively improved since DRR high-speed generation and the mixed similarity measure are realized in the GPU. Meanwhile, compared with the single similarity measure, the invention adopts the mixed similarity measure to improve the accuracy of a registration result.

Description

technical field [0001] The invention relates to rapid registration of medical images, in particular to a method for parallel registration of 2D-3D medical images based on combined similarity measure. Background technique [0002] Image-Guided Radiotherapy (IGRT) is a new radiotherapy technology after three-dimensional conformal radiotherapy (3DCRT) and intensity-modulated radiotherapy (IMRT). In image-guided radiation therapy, in order to determine the actual position of the fractionation treatment, the physicist uses an electronic portal imaging device to obtain X-ray portal images. The X-ray image reflects the actual position during treatment, while the CT volume data before treatment represents the ideal position for treatment. Through the 2D-3D registration of the X-ray field image and the CT volume data, the offset of the treatment bed is calculated, and finally Adjust the position of the treatment bed according to the offset to ensure the accuracy of the positioning d...

AI Technical Summary

Problems solved by technology

[0003] In 2D-3D image registration, it is necessary to reconstruct the 3D CT volume data to generate DRR. Since the generation process of DRR images is very time-consuming, and in the 2D-3D image registration algorithm, multiple DRR images need to be generated from different angles. Therefore, the generation process of DRR directly affects the efficiency of the entire 2D-3D registration algorithm. In addition, the calculation of the similarity measure also has an important impact on the operation efficiency and result accuracy of the entire registration process. Therefore, how to improve the 2D-3D registration algorithm The operating efficiency and accuracy to meet the needs of IGRT has become a hot research topic for many scholars.

GraemeP.Penney et al. compared six grayscale-based similarity measures applied to 2D-3D medical image registration. Even when there are differences in soft tissues in the two images to be registered, the pattern intensity and gradient The difference between the two similarity measures can still achieve accurate registration results and has good robustness; W.Birkfellner et al. proposed a DRR image generation method based on splat rendering and applied it to 2D-3D medical image registration In the standard, for volume data with a size of about 30MB on a PC, the time to draw a slightly blurred DRR image that can be used for 2D-3D registration is 100ms; Daniel B. Russakoff et al. proposed a method based on the attenuation domain The DRR rapid generation method is applied to 2D-3D registration, and its 2D-3D registration method has certain advantages in terms of execution efficiency and accuracy compared with the registration algorithm based on ray casting to generate DRR images; OsamaDorgham et al. Humans use the octree structure to compress the volume data, reducing the number of intersections between the projected rays and the internal space of the volume data, thereby increasing the speed of ray casting to generate DRR images, and applying it to 2D-3D registration can also achieve Effective registration accuracy can meet the accuracy requirements for correcting patient positioning errors; the above registration algorithms use different methods to improve the generation process of DRR images and improve the efficiency of registration algorithms, but they cannot meet the real-time requirements in image-guided radiotherapy. requirements; kubiasA et al. implemented a GPU-based 2D-3D medical image registration algorithm. This algorithm completes DRR generation and similarity measure calculation on the GPU, which is greatly compared with the traditional CPU-based registration algorithm. The registration speed has been greatly improved, but due to the limitation of the pipeline, the acceleration capability of the GPU is still not fully utilized; OsamaM.Dorgham et al. realized the parallel accelerated generation of DRR images on the GPU supporting CUDA, and at the same time, the ray casting process was performed. Compressed sampling and other processing to further improve the DRR generation speed, and finally applied the DRR accelerated generation method to 2D-3D medical image registration, and achieved acceptable accuracy for clinical radiotherapy, but this method does not use the similarity measure calculation process Also moved to GPU to run

Images