Effect of Varying Number of OSEM Subsets on PET Lesion Detectability

Journal of Nuclear Medicine Technology

A. Michael Morey¹  Dan J. Kadrmas¹ 

[1] Department of Radiology, Utah Center for Advanced Imaging Research (UCAIR), and Department of Bioengineering, University of Utah, Salt Lake City, Utah Department of Radiology, Utah Center for Advanced Imaging Research (UCAIR), and Department of Bioengineering, University of Utah, Salt Lake City, Utah Department of Radiology, Utah Center for Advanced Imaging Research (UCAIR), and Department of Bioengineering, University of Utah, Salt Lake City, Utah

关键词: positron emission tomography (PET);    OSEM;    lesion detection;    observer study;    LROC;   

DOI  :  10.2967/jnmt.113.131904

学科分类：医学（综合）

来源: Society of Nuclear Medicine

【 摘 要 】

Iterative reconstruction has become the standard for routine clinical PET imaging. However, iterative reconstruction is computationally expensive, especially for time-of-flight (TOF) data. Block-iterative algorithms such as ordered-subsets expectation maximization (OSEM) are commonly used to accelerate the reconstruction. There is a tradeoff between the number of subsets and reconstructed image quality. The objective of this work was to evaluate the effect of varying the number of OSEM subsets on lesion detection for general oncologic PET imaging. Methods: Experimental phantom data were taken from the Utah PET Lesion Detection Database, modeling whole-body oncologic 18F-FDG PET imaging of a 92-kg patient. The experiment consisted of 24 scans over 4 d on a TOF PET/CT scanner, with up to 23 lesions (diameter, 6–16 mm) distributed throughout the thorax, abdomen, and pelvis. Images were reconstructed with maximum-likelihood expectation maximization (MLEM) and with OSEM using 2–84 subsets. The reconstructions were repeated both with and without TOF. Localization receiver-operating-characteristic (LROC) analysis was applied using the channelized nonprewhitened observer. The observer was first used to optimize the number of iterations and smoothing filter for each case that maximized lesion-detection performance for these data; this was done to ensure that fair comparisons were made with each test case operating near its optimal performance. The probability of correct localization and the area under the LROC curve were then analyzed as functions of the number of subsets to characterize the effect of OSEM on lesion-detection performance. Results: Compared with the baseline MLEM algorithm, lesion-detection performance with OSEM declined as the number of subsets increased. The decline was moderate out to about 12–14 subsets and then became progressively steeper as the number of subsets increased. Comparing TOF with non-TOF results, the magnitude of the performance drop was larger for TOF reconstructions. Conclusion: PET lesion-detection performance is degraded when OSEM is used with a large number of subsets. This loss of image quality can be controlled using a moderate number of subsets (e.g., 12–14 or fewer), retaining a large degree of acceleration while maintaining high image quality. The use of more aggressive subsetting can result in image quality degradations that offset the benefits of using TOF or longer scan times.

【 授权许可】

Unknown