Lesion quantification in oncological positron emission tomography: A maximum likelihood partial volume correction strategy

• Published in: Medical physics (Lancaster) Vol. 36; no. 7; pp. 3040 - 3049
• Main Authors: De Bernardi, Elisabetta, Faggiano, Elena, Zito, Felicia, Gerundini, Paolo, Baselli, Giuseppe
• Format: Journal Article
• Language: English
• Published: United States American Association of Physicists in Medicine 01.07.2009
• Subjects: Algorithms; Cancer; Fluorodeoxyglucose F18; Image Interpretation, Computer-Assisted - methods; image reconstruction; image resolution; Image scanners; image segmentation; Iteration theory; iterative methods; Likelihood Functions; maximum likelihood estimation; maximum likelihood reconstruction; Medical image artifacts; Medical image noise; medical image processing; Medical image reconstruction; Medical image smoothing; Medical imaging; Neoplasms - diagnostic imaging; Numerical approximation and analysis; partial volume effect correction; PET; phantoms; Phantoms, Imaging; positron emission tomography; Positron emission tomography (PET); Positron-Emission Tomography - methods; Positron-Emission Tomography - statistics & numerical data; Probability theory, stochastic processes, and statistics; Radiography; Reconstruction; regional basis functions; Segmentation; tumours; uptake quantification; volume estimate; wounds; partial volume effect correction; volume estimate; maximum likelihood reconstruction; uptake quantification; PET; regional basis functions
• ISSN: 0094-2405; 2473-4209
• DOI: 10.1118/1.3130019

A maximum likelihood (ML) partial volume effect correction (PVEC) strategy for the quantification of uptake and volume of oncological lesions in F 18 -FDG positron emission tomography is proposed. The algorithm is based on the application of ML reconstruction on volumetric regional basis functions initially defined on a smooth standard clinical image and iteratively updated in terms of their activity and volume. The volume of interest (VOI) containing a previously detected region is segmented by a k -means algorithm in three regions: A central region surrounded by a partial volume region and a spill-out region. All volume outside the VOI (background with all other structures) is handled as a unique basis function and therefore “frozen” in the reconstruction process except for a gain coefficient. The coefficients of the regional basis functions are iteratively estimated with an attenuation-weighted ordered subset expectation maximization (AWOSEM) algorithm in which a 3D, anisotropic, space variant model of point spread function (PSF) is included for resolution recovery. The reconstruction-segmentation process is iterated until convergence; at each iteration, segmentation is performed on the reconstructed image blurred by the system PSF in order to update the partial volume and spill-out regions. The developed PVEC strategy was tested on sphere phantom studies with activity contrasts of 7.5 and 4 and compared to a conventional recovery coefficient method. Improved volume and activity estimates were obtained with low computational costs, thanks to blur recovery and to a better local approximation to ML convergence.