Spectral analysis (SA) in PET
In comparison to compartmental models, SA has only few presumptions. Like compartmental models, SA describes the kinetics of the tracer using homogeneous compartments, but there is no need to know the number of compartments; SA can instead be used to estimate the number of compartments. Therefore SA can be used for selecting or validating a compartmental model. SA has also been applied in determination of time delay (Hinz and Turkheimer, 2006), and in discriminating brain gray and white matter uptake on voxel-level analysis (Heurling et al., 2015).
IRF represents the tissue tracer concentration curve that would be measured (with PET) after an ideal instantaneous bolus injection. In spectral analysis IRF is assumed to be the sum of M+1 (0, …, M) exponential functions:
, where αj≥0, β0=0, and βj≥0.
Simulation of tissue curve
In practice, the input function to the tissue is far from ideal bolus. Spectral analysis requires arterial blood sampling to get the tracer concentration in arterial plasma, CP(t), used as the input function. Tissue curve, CT(t), can be computed by convolution between h(t) and CP(t):
Estimation of the spectrum
First a fixed list of βj values is defined, with a range starting from zero (β0=0). Since input function CP(t) is measured, we can calculate a table of basis functions for each βj,
, to replace the nonlinear part in equation (2). Non-negative least-squares (NNLS) method (Lawson and Hanson, 1974) can then be used to solve the αj values, minimizing the weighted residuals sum of squares (WRSS) between the simulated tissue curve and measured tissue curve, CPET(t):
, where N is the number of PET time frames, and wi are the weights of the time frames. Negative αj values would be nonphysiological, and NNLS method is therefore suitable for estimating αj.
The estimated αj values can be called the spectrum of the regional tissue TAC, and the structure of the model (number of compartments, reversibility and irreversibility) can be derived from the spectrum.
NNLS method is fast to compute, but SA should still not be applied to pixel-by-pixel calculations because of its sensitivity to noise.
Spectral analysis is applicable to analysis of PET data when the tracer has a radioactive metabolite which is transported into tissue, and the plasma input curves of both parent tracer and radioactive metabolite are measured and incorporated in the model (thus the name dual-input or double-input) (Tomasi et al., 2012).
Cunningham VJ, Ashburner J, Byrne H, Jones T. Use of spectral analysis to obtain parametric images from dynamic PET studies. In: Quantification of brain function. Tracer kinetics and image analysis in brain PET. Elsevier, 1993, pp 101-111.
Cunningham VJ, Jones T. Spectral analysis of dynamic PET studies. J Cereb Blood Flow Metab. 1993; 13: 15-23.
Cunningham VJ, Gunn RN, Byrne H, Matthews JC. Suppression of noise artifacts in spectral analysis of dynamic PET data. In: Quantitative functional brain imaging with positron emission tomography, p 329-334, Academic Press, 1998.
Hinz R, Turkheimer FE. Determination of tracer arrival delay with spectral analysis. IEEE Trans Nucl Sci. 2006; 53(1): 212-219.
Hudson HM, Walsh C. Density deconvolution using spectral mixture models. In: Proceedings of the Second World Congress of the IASC, Pasadena, CA, pp 593-599.
Lawson, C. L., Hanson, R.J.: Solving Least Squares Problems, Prentice-Hall, Englewood Cliffs, New Jersey, 1974.
Meikle SR, Matthews JC, Brock CS, Wells P, Harte RJA, Cunningham VJ, Jones T. Pharmacokinetic assessment of novel anti-cancer drugs using spectral analysis and positron emission tomography: a feasibility study. Cancer Chemother Pharmacol. 1998; 42: 183-193.
Reutens DC, Andermann M. Constraints in spectral analysis. In: Quantitative functional brain imaging with positron emission tomography. Academic Press, 1998, pp 335-337.
Riaño Barros DA, McGinnity CJ, Rosso L, Heckemann RA, Howes OD, Brooks DJ, Duncan JS, Turkheimer FE, Koepp MJ, Hammers A. Test-retest reproducibility of cannabinoid-receptor type 1 availability quantified with the PET ligand [11C]MePPEP. Neuroimage 2014; 97: 151-162.
Rizzo G, Veronese M, Zanotti-Fregonara P, Bertoldo A. Voxelwise quantification of [11C](R)-rolipram PET data: a comparison between model-based and data-driven methods. J Cereb Blood Flow Metab. 2013; 33: 1032-1040.
Schmidt K. Which linear compartment systems can be analyzed by spectral analysis of PET output data summed over all compartments. J Cereb Blood Flow Metab. 1999; 19: 560-569.
Sederholm K. Using NNLS in multilinear PET problems. TPCMOD0020.pdf.
Suominen H. Yleistettyyn lokeromalliin perustuva spektraalianalyysi positroniemissiotomografia-mallintamisessa. Pro gradu, 2005.
Tomasi G, Kimberley S, Rosso L, Aboagye E, Turkheimer F. Double-input compartmental modeling and spectral analysis for the quantification of positron emission tomography data in oncology. Phys Med Biol. 2012; 57: 1889-1906.
Turkheimer F, Moresco M, Lucignani G, Sokoloff L, Fazio F, Schmidt K. The use of spectral analysis to determine regional cerebral glucose utilization with positron emission tomography and [18F]fluorodeoxyglucose: theory, implementation, and optimization procedures. J Cereb Blood Flow Metab. 1994; 14: 406-422.
Turkheimer F, Sokoloff L, Bertoldo A, Lucignani G, Reivich M, Jaggi JL, Schmidt K. Estimation of component and parameter distributions in spectral analysis. J Cereb Blood Flow Metab. 1998; 18: 1211-1222.
Turkheimer FE, Hinz R, Gunn RN, Aston JAD, Gunn SR, Cunningham VJ. Rank-shaping regularization of exponential spectral analysis for application to functional parametric mapping. Phys Med Biol. 2003; 48: 3819-3841.
Updated at: 2017-11-18
Created at: 2014-05-07
Written by: Vesa Oikonen