Description
This project aims at developing highly efficient 3D medical reconstruction
methods based on CT, MRI, etc.,
on multi-core processors. For
competitive analysis, the scope
of this research is not only
limited to Intel IA architecture
but also includes the
state-of-the-art GPGPUs from
Nvidia, ATI and other
heterogeneous platforms such as
IBM CELL BE. The research components of this project involve
parallelization efforts of various
medical imaging algorithms using OpenMP
and pthread library. We are also
studying the
performance/productivity
tradeoff by using Nvidia's CUDA, RapidMinds, and
the upcoming OpenCL .
Our current main activity investigates the Katsevich algorithm [1],
which was
developed by Alexander Katsevich in 2001 on Intel's Core 2 Quad processors and
the latest Nehalem processor.
The Katsevich algorithm reconstructs a 3-D cylindrical volume from 2-D x-ray projections of an object.
It is a type of 'filtered backprojection' algorithm used in computed tomography (CT).
Each projection is created from an x-ray source located along the path of a helix surrounding the 3-D cylindrical volume.
The x-rays move radially away from the source, pass through the object, and hit a detector opposite the source.
(a) shows a series of projections taken along a helix of a simulated mathematical test image known as the Shepp-Logan phantom (d).
These projections are differentiated and weighted appropriately, producing (b).
Next, the projections undergo a 1-D Hilbert transform (c).
Finally, backprojection is performed.
The coordinates of a desired 3-D voxel to reconstruct are projected back onto these filtered projections to generate a 2-D coordinate on the projection.
The interpolated value from each filtered projection belonging to a voxel's PI interval are added together to reconstruct the density of the desired 3-D voxel.
This is repeated for each voxel in the 3-D cylindrical volume (e).
[1] Alexander Katsevich, "Theoretically exact FBP-type inversion algorithm for spiral CT", Society for Industrial and Applied Mathematics Journal on Applied Mathematics, 62:2012-2026, 2002
|
