Maximum entropy deconvolution of low count nuclear medicine images
Maximum entropy deconvolution of low count nuclear medicine images
Maximum entropy is applied to the problem of deconvolving nuclear medicine images, with special consideration for very low count data. The physical of the formation of scintigraphic images is described, illustrating the phenomena which degrade planar estimates of the tracer distribution. Various techniques which are used to restore these images are reviewed, outlining the relative merits of each.
The development and theoretical justification of maximum entropy as an image processing technique is discussed. Maximum entropy is then applied to the problem of planar deconvolution, highlighting the question of the choice of error parameters for low count data. A novel iterative version of the algorithm is suggested which allows the errors to be estimated from the predicted Poisson mean values. This method is shown to produce the exact results predicted by combining Poisson statistics and a Bayesian interpretation of the maximum entropy approach. A facility for total count preservation has also been incorporated, leading to improved quantification.
In order to evaluate this iterative maximum entropy technique, two comparable methods, Wiener filtering and a novel Bayesian maximum likelihood expectation maximisation technique, were implemented. The comparison of results obtained indicated that this maximum entropy approach may produce equivalent or better measures of image quality than the compared methods, depending upon the accuracy of the system model used.
The novel Bayesian maximum likelihood expectation maximisation technique was shown to be preferable over many existing maximum a posteriori methods due to its simplicity of implementation. A single parameter is required to define the Bayesian prior, which suppresses noise in the solution and may reduce the processing time substantially.
Finally, maximum entropy deconvolution was applied as a pre-processing step in single photon emission computed tomography reconstruction of low count data. Higher contrast results were obtained than those achieved by a Wiener pre-filtering approach and a scatter-subtracted attenuation corrected filtered back projection method.
Maximum entropy optimised for low counts holds promise for nuclear medicine applications where counts are necessarily low, and may facilitate reduction of the administered activity for other applications. The algorithm was in fact deemed advantageous for the processing of low count Poisson data in general.
University of Southampton
1999
McGrath, Deirdre Maria
(1999)
Maximum entropy deconvolution of low count nuclear medicine images.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
Maximum entropy is applied to the problem of deconvolving nuclear medicine images, with special consideration for very low count data. The physical of the formation of scintigraphic images is described, illustrating the phenomena which degrade planar estimates of the tracer distribution. Various techniques which are used to restore these images are reviewed, outlining the relative merits of each.
The development and theoretical justification of maximum entropy as an image processing technique is discussed. Maximum entropy is then applied to the problem of planar deconvolution, highlighting the question of the choice of error parameters for low count data. A novel iterative version of the algorithm is suggested which allows the errors to be estimated from the predicted Poisson mean values. This method is shown to produce the exact results predicted by combining Poisson statistics and a Bayesian interpretation of the maximum entropy approach. A facility for total count preservation has also been incorporated, leading to improved quantification.
In order to evaluate this iterative maximum entropy technique, two comparable methods, Wiener filtering and a novel Bayesian maximum likelihood expectation maximisation technique, were implemented. The comparison of results obtained indicated that this maximum entropy approach may produce equivalent or better measures of image quality than the compared methods, depending upon the accuracy of the system model used.
The novel Bayesian maximum likelihood expectation maximisation technique was shown to be preferable over many existing maximum a posteriori methods due to its simplicity of implementation. A single parameter is required to define the Bayesian prior, which suppresses noise in the solution and may reduce the processing time substantially.
Finally, maximum entropy deconvolution was applied as a pre-processing step in single photon emission computed tomography reconstruction of low count data. Higher contrast results were obtained than those achieved by a Wiener pre-filtering approach and a scatter-subtracted attenuation corrected filtered back projection method.
Maximum entropy optimised for low counts holds promise for nuclear medicine applications where counts are necessarily low, and may facilitate reduction of the administered activity for other applications. The algorithm was in fact deemed advantageous for the processing of low count Poisson data in general.
This record has no associated files available for download.
More information
Published date: 1999
Identifiers
Local EPrints ID: 463588
URI: http://eprints.soton.ac.uk/id/eprint/463588
PURE UUID: 14a0199e-0236-4e57-8d96-c9d87599a921
Catalogue record
Date deposited: 04 Jul 2022 20:54
Last modified: 04 Jul 2022 20:54
Export record
Contributors
Author:
Deirdre Maria McGrath
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics