Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

HIGH ORDER QUADRATIC PENALTY FUNCTIONS TO REDUCE RINGING ARTIFACTS IN ITERATIVE REGULARIZED IMAGE RECONSTRUCTION FOR POSITRON EMISSION TOMOGRAPHY

IP.com Disclosure Number: IPCOM000237016D
Publication Date: 2014-May-27
Document File: 7 page(s) / 33K

Publishing Venue

The IP.com Prior Art Database

Abstract

The invention proposes techniques to suppress ringing artifacts that occur when a detector PSF model is incorporated into the image reconstruction process. The goal of reducing artifacts is achieved through penalty functions in iterative regularized image reconstruction designed to minimize the side lobes of the resultant local impulse response. The techniquesconsider Poisson noise statistics through which count rate and patient dependence are systematically considered. The techniques include designing a quadratic penalty function, which suppresses ringing artifacts. A quadratic penalty function penalizes a quadratic function Rx of the array of image voxel values where x is a vector of image voxel values, which are lexicographically ordered, and R is the Hessian matrix of the quadratic penalty function.

This text was extracted from a Microsoft Word document.
This is the abbreviated version, containing approximately 30% of the total text.

HIGH ORDER QUADRATIC PENALTY FUNCTIONS TO REDUCE RINGING ARTIFACTS IN ITERATIVE REGULARIZED IMAGE RECONSTRUCTION FOR POSITRON EMISSION TOMOGRAPHY

BRIEF ABSTRACT

The invention proposes techniques to suppress ringing artifacts that occur when a detector PSF model is incorporated into the image reconstruction process. The goal of reducing artifacts is achieved through penalty functions in iterative regularized image reconstruction designed to minimize the side lobes of the resultant local impulse response. The techniquesconsider Poisson noise statistics through which count rate and patient dependence are systematically considered. The techniques include designing a quadratic penalty function, which suppresses ringing artifacts. A quadratic penalty function penalizes a quadratic function Rx of the array of image voxel values where x is a vector of image voxel values, which are lexicographically ordered, and R is the Hessian matrix of the quadratic penalty function.

KEYWORDS

Quadratic penalty function, ringing artifacts, PET, PSF model, image reconstruction, iterative regularized reconstruction


DETAILED DESCRIPTION

Positron emission tomography involves reconstructing the spatial distribution of a positron-emitting radioisotope inside a patient from photons measured at detectors surrounding the patient. The reconstructed images of the radioisotope distribution provide in vivo functional information about physiological processes. Accurate quantitation in emission tomography is important in cancer diagnosis and staging, and monitoring cancer response to therapy, and in studying in vivo physiological and neurological processes. An accurate data model, which maps radioisotope distribution into detected photons, enables accurate quantitation. Therefore, efforts are made to model data formation process more accurately by incorporating detector point spread functions (PSFs), which model such physical processes as crystal penetration, inter-crystal scattering and photon noncolinearity [1-3]. However, such attempts at more accurate models are found ironically to result in more pronounced ringing artifacts particularly at sharp edges in images [2-4].

In emission tomography, iterative image reconstruction techniques provide better image quality, compared to analytical techniques such as filtered backprojection (FBP), by using accurate models for data formation processes. There are two representative iterative reconstruction techniques. One of them iteratively finds an image, which maximizes the likelihood of measured data. This technique is referred to as maximum likelihood (ML) and usually implemented using expectation maximization (EM) algorithm or its variant OSEM. The other iteratively finds an image maximizing a likelihood penalized by a penalty function, which controls the resolution property in reconstructed images. This technique is referred to as penalized (maximum) likelihood, maximum a posteriori (MAP) or regularized reconstruction, and is implemen...