Real-Time 3D Ultrasound Reconstruction Using Octrees

Ultrasound is a major medical imaging modality that is widely used in healthcare because of advantages such as the use of nonionizing radiation, ease of operation, real-time imaging from different perspectives, and low operation costs. The most common ultrasound modality produces two-dimensional images (2D) with a 1D-array transducer. However, in recent years, three-dimensional ultrasound (3D US) imaging has become increasingly relevant. There are many reasons behind this shift. For example, 3D US images are easier to register with 3D images from another modality while patients undergo procedures or during presurgical planning. In particular, 3D freehand ultrasound (FUS) imaging yields 3D US images of large anatomical regions at low cost. An area of interest is scanned with a conventional 1D-array transducer, which is tracked with an attached device; the resulting 2D US images are input into a reconstruction algorithm; and the brightness values are assigned to a 3D image. Several 3D reconstruction algorithms in FUS imaging have been proposed and clinically used, and in the present work, we report a new neighbor search-based approach for reconstructing 3D FUS images based on hierarchical octrees with Morton key coding that can be implemented on GPUs using CUDA((R)) kernels to exploit multithreading. Our approach achieves considerably faster throughput for high-resolution 3D images and can reconstruct 3D US images with dimensions of 128 x 128 x 128 voxels in approximately 0.5 s. The proposed approach is a viable option for obtaining 3D US images in real time based on sets of freehand 2D ultrasound images acquired with 1D-array transducers.