RT Journal Article
SR Electronic
T1 Analysis of Biased PET Images Caused by Inaccurate Attenuation Coefficients
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 753
OP 760
DO 10.2967/jnumed.109.070326
VO 51
IS 5
A1 Young-Don Son
A1 Hang-Keun Kim
A1 Sung-Tae Kim
A1 Nam-Beom Kim
A1 Young-Bo Kim
A1 Zang-Hee Cho
YR 2010
UL http://jnm.snmjournals.org/content/51/5/753.abstract
AB PET scanners with an elongated axial field of view intended to increase overall system sensitivity, such as the high-resolution research tomograph (HRRT) scanner, have been reported to produce images with decreased signals in the brain stem and cerebellum. The cause of this negative bias of the images was analyzed, and the effects of an inaccurate linear attenuation coefficient (μ-value) of tissue and bones were separately examined. Methods: A new phantom was manufactured, and 18 human subjects were recruited for the study. 18F-FDG PET images were reconstructed using attenuation coefficient maps generated by various algorithms. The algorithms included maximum a posteriori reconstruction for transmission data (MAP-TR) with default priors, MAP-TR with adjusted priors for bone (MAP-TRadj-b), MAP-TR with adjusted priors for tissue (MAP-TRadj-t), and noise-equivalent count TR and CT-TR. Results: With the CT-TR and MAP-TRadj-t algorithms, increased intensity in the brain stem and cerebellum was seen, and negative bias was reduced. With the MAP-TRadj-t algorithm, however, positive bias increased in the central region. Inappropriate attenuation coefficients of brain tissue increased the positive or negative bias of reconstructed images, especially for the central regions of the volume. Poor representation of the skull or bone also locally increased the bias in the near regions where bone detection had failed. Conclusion: An inaccurate μ-map obtained from the MAP-TR algorithm caused the bias problem for the HRRT system. The CT-TR algorithm provided a relatively more reliable μ-map that demonstrated a small degree of intensity bias. Appropriate priors for μ-values of each tissue compartment and better classification to distinguish bone from tissue are necessary for accurate attenuation correction.