COBRA software development kit. Integration COBRA into customer’s software.

Recommended PC 

We recommend the following PC giving the best price/performance ratio
CPU: Intel quad-core i7 / 2.4+ GHz
RAM: 6 GB
GPU: nVIDIA Geforce GTX480
OS: Windows7 64 bit


COBRA speed for the basic CT reconstruction task

Lets assume that the used PC is what is recommended above. Lets also assume that our scanner is built corresponding a classic Felfkamp scheme (symmetric detector setup); input projections (views) are 512x512 and our CT scan contains 512 input views. All projections are located on the hard disk. Then it takes approximately 15 seconds to get a reconstructed volume 512x512x512 in RAM and additional 5-8 seconds to save the volume to the disk.
The reconstruction speed is 25-30 msec per input view 

Performing on various hardware platforms
The reconstruction speed obviously depends on PC hardware.

PC with a graphics card. For cards with a decent number of processing units (192 and up) and a decent size of built-in RAM (768 MB and up) a reconstruction speed may vary from 52 msec per view with 8800GTX to 25 msec per view with GTX480.
Desktop PC. If PC specified above does not have a decent graphics card then the reconstruction speed is approximately 120 msec per input view.
Laptops. If the reconstruction is being run on Intel i7 laptop or powerful 3 years old dual core desktop with 2GB RAM then the reconstruction speed is approximately 200 msec per input view.

All numbers in this section are relevant for the basic CT reconstruction task (512x512x512 → 512x512x512).

Dependence on geometry setup and corrections


The reconstruction speed dramatically depends on input/output data and applied corrections. The following approximate rules may be taken into account
Number of input views: linear dependence.
Input view size: both doubled X and Y projection size causes factor 1.5 slow down.
Cube size: linear dependence on voxel amount.
Metal artifact reduction: factor 2 longer
Feldkamp artifact reduction: 20 (GPU) / 40 (CPU) sec extra plus doubled time of slice saving
Dynamic slice scale calibration: 20 sec extra plus doubled time of slice saving.
Half-beam mode: factor 1.2-1.5 slow down comparing to a standard full-beam setup with identical parameters.
3x View mode: factor 2 slow down comparing to a standard full-beam setup with identical parameters.
Noise reduction: factor 1.2-2.0 slow down of  slice saving

Examples

PC: Intel quad-core i7 / 2.4 GHz / Geforce GTX480

360 input views 512x512 → 512x512x512 volume:
18- 22 sec total time.

768 input views 1024x1024 → 1024x1024x1024 volume: 3-4 min total time.

5x1024 input views (5 stacked volumes)
256x256 → 512x512x700 consolidated volume, CFA/Autoscale/Noise reduction applied: approx. 6 min