Research On Virtual Endoscopy Technology Based On GPU

Virtual endoscopy technology which is non-invasive develops a new method for medical diagnosis, it can detect organs which are extremely small or unable to explore for traditional endoscopy, such as brain and heart. Although the virtual endoscopy technology is developing at high pace now, it still has problems such as the slow speed of centerline extraction and the low efficiency of interaction. Therefore, the main purpose of this article is to integrate the GPU ’s parallel computing performance into virtual endoscopy technology implementation, which can improve the efficiency of centerline extraction. Furthermore, this article aims to develop a virtual endoscopy system with highly interactive performance to help clinical diagnosis and treatment.In terms of centerline extraction, potential field method has been used. To reduce the time complexity, this article presents a parallel potential field skeleton extraction method which is suitable for implementation on GPU, and then improves it by using constant memory and shared memory which is unique in CUDA. The implementation has been tested on several complex 3D models in CUDA framework. The results show that our method has excellent performance especially on large data scale. When processing the volume data with the scale of 256 ×256×487, this improved method achieves speedups of 18 x. It solves the problems of the slow speed of centerline extraction.In terms of 3D visualization, surface rendering based on marching cubes and the volume rendering based on ray-casting has been realized. However, the volume rendering often fails to meet the requirements of real-time rendering because of the large computation when only using CPU. Therefore, this article presents a parallel volume rendering method based on GPU. It uses the GPU’s parallel computing performance to improve frames per second. The test results show that for s everal 3D models of different sizes, the frames per second of this parallel method reaches more than 40 which meets the demands of real-time rendering. In addition, this article also provides real-time interactive features. Users can set their own transfer function to achieve the desired rendering results. Moreover, they can zoom, rotate or drag the visual results.In terms of virtual roaming, the foreshortening effects has been implemented by transforming the parallel projection to perspective projection. It provides guided navigation. Users can control the virtual camera walk along the centerline or put it on a fixed position for careful observation. They can zoom or rotate the virtual camera lenses by clicking the mouse or buttons.Finally, this article designs and implements a virtual endoscopy system by mixing the above three technologies. This system has good interactive performance which can improve the user experience, and provides an effective assistant platform of clinical diagnosis and treatment.