Simultaneous Tc-99m/I-123 SPECT brain imaging using generalized spectral factor analysis

In brain SPECT, simultaneous Tc-99m/I-l23 acquisitions allow comparison of the distribution of two radiotracers in brain diseases, while avoiding image misregistration issues. However, there is no solution to the crosstalk caused by Tc-99m and I-123 photopeak overlap and by downscatter of I-123 photons into the Tc-99m spectral window and accurate quantification cannot be achieved. We describe a generalized spectral factor analysis (GSFA) method for solving crosstalk and downscatter problems in simultaneous Tc-99m/I-123 SPECT to improve quantitative accuracy. In GSFA, the spectrum of the photons detected in each pixel is expressed as a linear combination of two photopeaks and K-2 scattered spectra common to all pixels. These basis spectra are estimated using a PCA of all spectra from which a Q-dimensional study space 3 is derived. The basis spectra are then identified in 3 using priors. Conventional SFA assumes that the number K of basis spectra is equal to the dimension Q of 3. We generalized SFA to estimate K factors in a Q-dimensional space, Q>K. Using Monte Carlo simulations of a Tc-99m/I-123 brain phantom, we show that unlike SFA or conventional spectral windows (WIN), GSFA yields accurate quantitative measurements both from the I-123 images (errors <5% against errors between -13% and +14% with SFA and WIN) and from the Tc-99m images with a maximum error less than 4.5%, against errors between 28% and 49% with SFA and WIN.