Factor analysis of dynamic PET images

Intrinsic tissue heterogeneity and partial volume effect lead to image voxel time-activity curves that are a mixture of time-activity curves of pure homogeneous regions of tissue and blood. Factor analysis of dynamic data is a mathematical technique that aims to automatically resolve the true time-activity curves (Di Paola et al., 1982; Geckle and Szabo, 1992), helping in tissue segmentation, spill-over correction (Lee et al., 2005; Millet et al., 2012), obtaining image-derived input function (Wu et al., 1995; Ahn et al., 2000; Schiepers et al., 2008), and computation of parametric images (Tsartsalis et al., 2018).

The dynamic data is transformed to obtain key temporal components, and factor analysis fits a linear combination of selected components to the dynamic data. Typical problems in factor analysis, when applied to noisy data, include the high computational burden of the optimization process, and non-uniqueness of the results unless some constraints are set (Sitek et al., 2002).

Non-negative matrix factorization

One obvious constraint that should be applied is non-negativity; non-negative matrix factorization (NMF, or sometimes NNMF, or non-negative matrix approximation, or self modeling curve resolution) (Lawton & Sylvestre, 1971; Paatero & Tapper, 1994; Paatero, 1997; Lee DD & Seung, 1999 and 2001; Li et al., 2014) has been shown to provide more robust segmentation of tissue and blood components than factor analysis or cluster analysis in dynamic PET (Lee JS et al., 2001a and 2001b; Kim et al., 2001; Choi et al., 2002; Ahn et al., 2004; Bödvarsson B et al., 2006 and 2007; Hwang et al., 2009; Schulz et al., 2012; Mu et al., 2013; Kopriva et al., 2017). NMF has not been successful in some PET applications (Zanotti-Fregonara et al., 2009).

NMF can be used as part of feature extraction methods in diagnostics, for example in FDG studies of Alzheimer’s disease (Segovia et al., 2016). The use of algorithms related to NMF are also used in PET image reconstruction.


See also:



References:

Ahn JH, Kim S, Oh JH, Choi S. Multiple nonnegative-matrix factorization of dynamic PET images. In: ACCV, p. 5, 2004.

Boutchko R, Mitra D, Baker SL, Jagust WJ, Gullberg GT. Clustering-initiated factor analysis application for tissue classification in dynamic brain positron emission tomography. J Cereb Blood Flow Metab. 2015; 35: 1104-1111.

Bödvarsson B, Mørkebjerg M, Hansen LK, Knudsen GM, Svarer C. Extraction of time activity curves from positron emission tomography: K-means clustering or non-negative matrix factorization. NeuroImage 2006; 31: T185-T186. doi: 10.1016/j.neuroimage.2006.04.167.

Buvat I, Benali H, Di Paola. Statistical distribution of factors and factor images in factor analysis of medical image sequences. Phys Med Biol. 1998; 43: 1695-1711.

Di Paola R, Bazin JP, Aubry F, Aurengo A, Cavailloles F, Herry JY, Kahn E. Handling of dynamic sequences in nuclear medicine. IEEE Trans Nucl Sci. 1982; NS-29(4): 1310-1321.

Foster B, Bagci U, Mansoor A, Xu Z, Mollura DJ. A review on segmentation of positron emission tomography images. Comput Biol Med. 2014; 50(1): 76-96.

Geckle WJ, Szabo Z. Physiologic factor analysis (PFA) and parametric imaging of dynamic PET images. In: Proceedings of the Annual IEEE Symposium for Computer-based Medical Systems. 1992; 5: 9-16.

Hermansen F, Bloomfield PM, Ashburner J, Camici PG, Lammertsma AA. Linear dimension reduction of sequences of medical images: II. Direct sum decomposition. Phys Med Biol. 1995; 40: 1921-1941.

Hermansen F, Lammertsma AA. Linear dimension reduction of sequences of medical images: III. Factor analysis in signal space. Phys Med Biol. 1996; 41: 1469-1481.

Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization. Nature 1999; 401: 788-791.

Lee DD, Seung HS. Algorithms for non-negative matrix factorization. Advances in Neural Information Processing Systems 13: Proceedings of the 2000 Conference. MIT Press, 2001. pp. 556-562.

Lee JS, Lee DD, Choi S, Lee DS. Application of non-negative matrix factorization to dynamic positron emission tomography. In: Lee T-W, Jung T-P, Makeig S, Sejnowski TJ (eds.), Proceedings of the 3rd International Conference on Independent Component Analysis and Signal Separation, ICA2001, San Diego, pp. 629-632 (2001b).

Lee JS, Lee DS, Ahn JY, Cheon GJ, Kim S-K, Yeo JS, Park KS, Chung J-K, Lee MC. Parametric image of myocardial blood flow generated from dynamic H215O PET using factor analysis and cluster analysis. Med Biol Eng Comput. 2005; 43: 678-685.

Millet P, Moulin-Sallanon M, Tournier BB, Dumas N, Charnay Y, Ibáñez V, Ginovart N. Quantification of dopamine D2/3 receptors in rat brain using factor analysis corrected [18F]Fallypride images. NeuroImage 2012; 62: 1455-1468. doi: 10.1016/j.neuroimage.2012.05.075.

Pedersen F, Bergström M, Bengtsson E, Långström B. Principal component analysis of dynamic positron emission tomography images. Eur J Nucl Med. 1994; 21(12): 1285-1292.

Schulz D, Tapfer A, Buck A, Reeder S, Miederer M, Weidl E, Ziegler SI, Schwaiger M, Bundschuh RA. Non-negative matrix factorization based input function extraction for mouse imaging in small animal PET - comparison with arterial blood sampling and factor analysis. J Mol Imaging Dynam. 2012; 2(2):108.

Szabo Z, Ravert HT, Gözükara I, Geckle W, Seki C, Sostre S, Peller P, Monsein L, Natarajan TK, Links JM, Wong DF, Dannals RF, Wagner HN Jr. Noncompartmental and compartmental modeling of the kinetics of carbon-11 labeled pyrilamine in the human brain. Synapse 1993; 15: 263-275.

Wu H-M, Hoh CK, Choi Y, Schelbert HR, Hawkins RA, Phelps ME, Huang S-C. Factor analysis for extraction of blood time-activity curves in dynamic FDG-PET studies. J Nucl Med. 1995; 36(9): 1714-1722.

Wu H-M, Huang S-C, Allada V, Wolfenden PJ, Schelbert HR, Phelps ME, Hoh CK. Derivation of input function from FDG-PET studies in small hearts. J Nucl Med. 1996; 37(10): 1717-1722.



Tags: , ,


Created at: 2017-02-16
Updated at: 2018-12-06
Written by: Vesa Oikonen