Research Article

Instant Feedback Rapid Prototyping for GPU-Accelerated Computation, Manipulation, and Visualization of Multidimensional Data

Table 1

Timings for selected operations on 3 different volumes. The speedup factor between CPU and GPU is calculated as the lowest CPU time (8 threads) divided by the GPU time. The 3D Euclidian distance transform operates on normalized (pixel values in ) volumes with a solidity threshold of 0.5. The CPU implementation is our own, including the multithreading. Intensity rescaling and median filtering use ITK’s CPU-based implementation. The median filter takes the direct neighborhood of each voxel into account, that is, a box of 27 voxels in total.

Volume CPU (1 thread)CPU (8 threads)GPUSpeedup 1 versus 8 threadsSpeedup CPU versus GPU

Distance transform
134.5 s45.9 s630 ms2.9x73x
63.2 s20.3 s217 ms3.1x93x
1631 s627 s4900 ms2.6x128x

Rescale intensity to
88 ms68 ms10.1 ms1.29x6.7x
38 ms30 ms7.2 ms1.27x4.17x
603 ms457 ms97.5 ms1.31x4.7x

Median filter
4155 ms1544 ms97 ms2.7x15.9x
2120 ms671 ms48 ms3.15x14x
14 s4800 ms715 ms2.9x6.7x