Abdominal Image Segmentation

Patented HPC technology was applied to the analysis of computed tomography.* A test was performed for segmentation of regions of computed tomography of the abdomen. The image region contained 6.5M pixels. In this case, a solution with 1% precision was obtained in 7 seconds. By comparision the fastest serial computing implementation of the graph cut required 175 seconds.** HPC provides a 25X speedup.



Cardiac Image Segmentation

Patented HPC technology was applied to the segmentation of a cardiac image with 58M voxels. The graph used for performing the segmentation contained 58M nodes and 745M edges.*** In this case, a solution with 3% precision was obtained in 28 seconds. By comparision the fastest serial computing implementation of the graph cut required 322 seconds.** HPC provides an 11X speedup.



Prenatal Ultrasound Segmentation

Patented HPC technology was applied to the segmentation of prenatal ultrasound.* The reconstruction required finding the minimum cut of a weighted directed graph with 5M nodes and 65M edges. A solution with 10% precision was obtained in 13 seconds. By comparision the fastest serial computing implementation of the graph cut required 82 seconds.** HPC provides a 6X speedup.



Liver Segmentation

Patented HPC technology was applied to the segmentation of the liver in contrast computed tomography.***** The reconstruction required finding the minimum cut of a weighted directed graph with 46M nodes and 590M edges. A solution with 1% precision was obtained in 36 seconds. By comparision the fastest serial computing implementation of the graph cut required 255 seconds.** HPC provides a 7X speedup.



Photo Editing

Patented HPC technology was applied to photo editing.* A test was performed on a photograph with 3.5M pixels. In this case, a solution with 1% precision was obtained in 1.5 seconds. By comparision the fastest serial computing implementation of the graph cut required 12.0 seconds.** HPC provides an 8X speedup.



Depth from Multiview Photography

Patented HPC technology was applied to construct the disparity surface from three photographic views.**** The reconstruction required finding the minimum cut of a weighted graph with 16M nodes and 81M edges. A solution with 1% precision was obtained in 3.7 seconds. By comparision the fastest serial computing implementation of the graph cut required 47 seconds.** HPC provides an 12X speedup.




*The HPC algorithm was run on an Infiniband QDR cluster using 400 Intel Haswell cores. **Serial processing using algorithm in Boykov, Kolmogorov, IEEE PAMI, 26(9)1124-37,2004 ***The HPC algorithm was run on an Infiniband QDR cluster using 800 Intel Westmere cores. . ****The HPC algorithm was run on an Infiniband QDR cluster using 300 Intel Westmere cores. *****The HPC algorithm was run on an Infiniband QDR cluster using 1008 Intel Haswell cores.



Copyright © 2012 Virtual Scalpel, Inc., All rights reserved