Real-time Imaging Of Freehand 3D Ultrasound Based On GPU

In medical diagnostics, ultrasonic testing has been widely used in clinical practice for it has the advantages of low-cost, fast- imaging and non- invasive. Traditional two-dimensional ultrasound(2D US) imaging could provide cross-sectional information of the tissue in real-time, but is unable to help the doctors get hold of the anatomy of the tissue. The doctor should image the three-dimensional structure of the scanning area with his experience, and thus reduce the diagnosis accuracy. To address this limitation, three-dimensional ultrasound(3D US) imaging has been proposed and applied to the actual medical diagnosis. 3D US gives doctors the intuitive information, so that the doctors have good understanding of the spatial information and anatomical structure of the region of interest(ROI), and get more accuracy result. However, the disadvantages of expensive volume reconstruction complexity and reconstruction time make it difficult to be applied in some real-time application scenario. It will make the 3D US play a greater role in medical diagnostic applications if the imaging al ows real-time volume reconstruction and volume visualization of the ROI while scanning.This paper presents a real-time freehand 3D US imaging system based on graphics processing unit(GPU). In terms of selecting reconstruction algorithms, we used the squared distance weighted(SDW) interpolation algorithm and Bezier interpolation algorithm by taking into account the balance between reconstruction speed and accuracy. In order to reach the goal of real-time volume reconstruction, we parallelized the two reconstruct ion algorithms and handled the expensive computation, i.e. volume reconstruction and volume rendering, with GPU’s parallel computing power, expecting to complete the real-time incremental volume reconstruction and volume visualization while scanning.In order to verify the feasibility of real-time visualized reconstruction with the selected mapping algorithms, we used a set of data in vitro and another set of data in vivo to conduct the offline simulation experiment of real-time imaging and found that the parallelized SDW cannot get the effects of real-time imaging while the Bezier one enabled the real-time visualized reconstruction with a speed of up to 119 frames/second(the acquisition rate of B-scan image is typically 25frames/s). Therefore, we selected the Bezier interpolation algorithm as the reconstruction method of the real-time 3D US imaging system. In the online real-time experiment, the fetal resolution phantom’s fingers as well as a human’s fingers were scanned. The result showed that the real- time freehand 3D US imaging system based on GPU achieved the effects of real-time volume reconstruction and volume visualization during the scanning, indicating a high application value in medical diagnosis.