Generalized five-dimensional dynamic and spectral factor analysis.

We have generalized the spectral factor analysis and the factor analysis of dynamic sequences (FADS) in SPECT imaging to a five-dimensional general factor analysis model (5D-GFA), where the five dimensions are the three spatial dimensions, photon energy, and time. The generalized model yields a significant advantage in terms of the ratio of the number of equations to that of unknowns in the factor analysis problem in dynamic SPECT studies. We solved the 5D model using a least-squares approach. In addition to the traditional non-negativity constraints, we constrained the solution using a priori knowledge of both time and energy, assuming that primary factors (spectra) are Gaussian-shaped with full-width at half-maximum equal to gamma camera energy resolution. 5D-GFA was validated in a simultaneous pre-/post-synaptic dual isotope dynamic phantom study where 99mTc and 123I activities were used to model early Parkinson disease studies. 5D-GFA was also applied to simultaneous perfusion/dopamine transporter (DAT) dynamic SPECT in rhesus monkeys. In the striatal phantom, 5D-GFA yielded significantly more accurate and precise estimates of both primary 99mTc (bias=6.4 % +/- 4.3 %) and 1231 (-1.7% +/- 6.9%) time activity curves (TAC) compared to conventional FADS (biases = 15.5% +/- 10.6% in 99mTc and 8.3% +/- 12.7% in 123I, p < 0.05). Our technique was also validated in two primate dynamic dual isotope perfusion/DAT transporter studies. Biases of 99mTc-HMPAO and 123I-DAT activity estimates with respect to estimates obtained in the presence of only one radionuclide (sequential imaging) were significantly lower with 5D-GFA (9.4% +/- 4.3% for 99mTc-HMPAO and 8.7% +/-4.1% for 123I-DAT) compared to biases greater than 15% for volumes of interest (VOI) over the reconstructed volumes (p < 0.05). 5D-GFA is a novel and promising approach in dynamic SPECT imaging that can also be used in other modalities. It allows accurate and precise dynamic analysis while compensating for Compton scatter and cross-talk.