Raped calculation of detectability in Bayesian SPECT

Yuxiang Xing1, G. Gindi1
1Departments of Electrical & Computer Engineering and Radiology, SUNY Stony Brook, USA

Tóm tắt

To speed up the laborious process of optimizing and comparing SPECT imaging systems, computational methods using model observers can be used to predict human performance in the context of a lesion detection task. We address the case of SPECT images reconstructed by Bayesian MAP methods and the use of a channelized Hotelling observer (CHO) for a detection task. It is possible to write a closed-form theoretical expression to evaluate an image quality figure-of-merit (FOM) for this case. The evaluation of this FOM expression is intractably computationally intensive due, it turns out, to the nature of attenuation and depth-dependent blur in SPECT. We make several approximations that make this computation tractable, and validate our new expressions using Monte-Carlo simulations.

Từ khóa

#Bayesian methods #Optimization methods #Computational modeling #Context modeling #Predictive models #Humans #Lesions #Image reconstruction #Image quality #Attenuation

Tài liệu tham khảo

qi, 1999, A Theoretical Study of the Contrast Recovery and Variance of MAP Reconstructions from PET Data, IEEE Trans Med Imag, 18 10.1109/83.491322 xing, 2001, Efficient Calculation of Resolution and Variance in 2D Circular-Orbit SPECT, Conf Rec IEEE Nuc Sci Sym Med lmag Conf 10.1109/83.535846 10.1109/42.938249 10.1109/42.370406 10.1364/JOSAA.4.002447 10.1109/23.873017 10.1073/pnas.90.21.9758