12292

Improved Distance Weighted GPU-based 3D Ultrasound Reconstruction Methods

Tord Oygard
Norwegian University of Science and Technology, Faculty of Information Technology, Mathematics and Electrical Engineering, Department of Computer and Information Science
Norwegian University of Science and Technology, 2014

@article{oygard2014improved,

   title={Improved Distance Weighted GPU-based 3D Ultrasound Reconstruction Methods},

   author={Oygard, Tord},

   year={2014},

   publisher={Institutt for datateknikk og informasjonsvitenskap}

}

Download Download (PDF)   View View   Source Source   

2330

views

Ultrasound is a flexible medical imaging modality with many uses, one of them being intra-operative imaging for use in navigation. In order to obtain the highest possible spatial resolution and avoiding big, clunky 3D ultra-sound probes, reconstruction of many 2D ultrasound images obtained by a conventional 2D ultrasound probe with a tracking system attached has been employed.Earlier work in this field has yielded fast Graphical Processing Unit(GPU)-based implementations of voxel-based reconstruction algorithms such as VoxelNearest Neighbor(VNN), Pixel Nearest Neighbor(PNN), VNN2 and Distance Weighted(DW) reconstruction. However, it is desirable to improve upon the reconstruction quality of the methods mentioned above. To do so, we propose in this thesis an adaptive algorithm called VGDW, which tries to intelligently smooth away speckles and noise, yet retains detail in high-frequency regions, while being not being much slower than the above mentioned algorithms. It also has a tunable weight function enabling value collisions to be handled gracefully.Using our novel adaptive algorithm, we are able to produce very high-quality reconstructions, which are unanimously preferred over the output of the above mentioned algorithms by both a group of medical personnel and agroup of technologists working with ultrasound, while having comparable computation time to VNN2 and DW, i.e. 16%, 10% and 5% difference fromDW when computing a volume with 128 millions of voxels from a small,medium-sized and very large input data set using an AMD Radeon 6470MGPU. The algorithm also performs especially well with complex scanning patterns with overlapping data when using a customized weight function. As for future work, there are some aspects of the weight function that can benefit from improvement. Also, turning the problem upside down and looking at it from a pixel-based perspective could potentially give huge benefits bothin terms of probe movement robustness and performance.
No votes yet.
Please wait...

* * *

* * *

HGPU group © 2010-2024 hgpu.org

All rights belong to the respective authors

Contact us: