Nikolay Gavrilov
personal web page
gavrilov86@gmail.com

GPU-based computed tomography

In this project I implemented the FDK algorithm (Feldkamp, Davis and Kress) for FBP (Filtered Back Projection) to reconstruct CT images by cone-beam X-Ray projections (CBCT). This algorithm was multithreaded both on CPU and GPU. In GPU case the GLSL shaders were used.

Benchmarks:

Preprocessing step includes 2D filtering of the projections on CPU (takes 30 seconds for 513 projections 616x608 pixels each).
CT reconstruction of the volume 512512442 voxels:
CPU: 18 minutes, GPU: 25 seconds.
GeForce GTS 250, Intel Cor i5-4200H 2.80GHz.


Dental dataset reconstruction. Axial scan & Volume Rendering.
Dental dataset reconstruction. Axial scan & Volume Rendering.
Raw projections
Raw projections
512512 scan reconstruction process GIF-animation
512512 scan reconstruction process GIF-animation


>Test dataset example. Rat skull scanned by X-Ray microtomograph.
Test dataset example. Rat skull scanned by X-Ray microtomograph.
Reconstruction tool GUI. The projection on the detector surface is filtered by ramp-filter.
Reconstruction tool GUI. The projection on the detector surface is filtered by ramp-filter.
Pomegranade Poppyseeds
Pomegranade & poppyseeds reconstructions.


Rat skull projections and other datasets can be found here

The video below shows the reconstruction process.



e-mail:gavrilov86@gmail.com
© 2009-2022 Nikolay Gavrilov