Programs for Evaluation of 3D PET Reconstruction Algorithms
Abstract
Evaluation of a reconstruction algorithm should be done using a sample set that is large enough to provide us with a statistically significant result. In order to carry out an evaluation, one possibility is to use a set of computer simulated phantoms, that takes into account some parameter variabilities. This technical report describes in detail programs that generate a set of 3D phantoms and projection data, reconstruct, evaluate and then compare. The main characteristics are: (1) Phantom and projection data generation, (a) Phantoms with many (69) ellipsoid features, ranging from small (4 mm) to large (40 mm) sized features; (b) Phantoms are random samples from a statistically described ensemble of 3D images resembling those to which PET would be applied in a medical situation (features with random size, orientation and activity); (c) Features are inside spheres that provide background value for some important clinical tasks such as detectability; (d) Types of features: hot, cold and normal spots, and (e) Emulation of 3D PET scanner for projection data generation, with detector field of view (FOV) blurring and a realistic 3D PET noise model. (2) Reconstruction algorithms: (a) Algebraic Reconstruction Technique using blob (ARTblob) as basis function; (b) ART using voxels (ARTvox); (c) EM-ML using blobs (EMblob), and (d) EM-ML using voxels (EMvox). (3) Evaluation for following tasks: (a) Training figure of merit (FOM); (b) Structural accuracy; (c) Hot spot detectability, and (d) Cold spot detectability. (4) Statistical comparison using paired t-test. Justifications for using some models are described in the paper and an application evaluating some reconstruction methods is reported.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1994
- Accession Number
- ADA331311
Entities
People
- G. T. Herman
- R. M. Klnahan
- S. Matej
- S. S. Furule
- T. K. Nayaran
Organizations
- University of Pennsylvania