18F‐FLT uptake kinetics in head and neck squamous cell carcinoma: A PET imaging study. Issue 4 (20th March 2014)

• Record Type: Journal Article
• Authors: Liu, Dan; Chalkidou, Anastasia; Landau, David B.; Marsden, Paul K.; Fenwick, John D.
• Is Part Of: Medical physics. Volume 41:Issue 4(2014)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2014-03-20
• Subjects: Positron emission tomography (PET) -- Cancer -- Dose‐volume analysis -- Probability theory, stochastic processes, and statistics; Bayes methods -- biochemistry -- cancer -- cellular biophysics -- dosimetry -- drugs -- physiological models -- positron emission tomography -- reaction kinetics -- tumours; compartment model -- Bayesian information criterion -- expectation‐maximization clustering -- parametric map; Therapeutic activity of chemical compounds or medicinal preparations -- Biochemistry; Beer; Spirits; Wine; Vinegar; Microbiology; Enzymology; Mutation or genetic engineering -- Scintigraphy -- Measuring half‐life of a radioactive substance; Cancer -- Medical imaging -- Biomedical modeling -- Positron emission tomography -- Cluster analysis -- Cell growth -- Numerical modeling -- Radioactivity -- Image analysis -- Dosimetry; Medical physics -- Periodicals; Medical physics; Geneeskunde; Natuurkunde; Toepassingen; Biophysics; Periodicals; Periodicals; Electronic journals; 610.153
• DOI: 10.1118/1.4868462
• Languages: English
• ISSNs: 0094-2405

Abstract

Abstract : Purpose: : To analyze the kinetics of 3 ′ ‐deoxy‐3 ′ ‐[F‐18]‐fluorothymidine (18F‐FLT) uptake by head and neck squamous cell carcinomas and involved nodes imaged using positron emission tomography (PET). Methods: : Two‐ and three‐tissue compartment models were fitted to 12 tumor time‐activity‐curves (TACs) obtained for 6 structures (tumors or involved nodes) imaged in ten dynamic PET studies of 1 h duration, carried out for five patients. The ability of the models to describe the data was assessed using a runs test, the Akaike information criterion (AIC) and leave‐one‐out cross‐validation. To generate parametric maps the models were also fitted to TACs of individual voxels. Correlations between maps of different parameters were characterized using Pearsonˈs r coefficient; in particular the phosphorylation rate‐constants k 3 ‐ 2 tiss and k 5 of the two‐ and three‐tissue models were studied alongside the flux parameters K FLT‐ 2 tiss and K FLT of these models, and standardized uptake values (SUV). A methodology based on expectation‐maximization clustering and the Bayesian information criterion ("EM‐BIC clustering") was used to distil the information from noisy parametric images. Results: : Fits of two‐tissue models 2C3K and 2C4K and three‐tissue models 3C5K and 3C6K comprising three, four, five, and six rate‐constants, respectively, pass the runs test for 4, 8, 10, and 11 of 12 tumor TACs. The three‐tissue models have lower AIC and cross‐validation scores for nine of the 12 tumors. Overall the 3C6K model has the lowest AIC and cross‐validation scores and its fitted parameter values are of the same orders of magnitude as literature estimates. Maps of K FLT and K FLT‐ 2 tiss are strongly correlated ( r = 0.85) and also correlate closely with SUV maps ( r = 0.72 for K FLT‐ 2 tiss, 0.64 for K FLT ). Phosphorylation rate‐constant maps are moderately correlated with flux maps ( r = 0.48 for k 3 ‐ 2 tiss vs K FLT‐ 2 tiss and r = 0.68 for k 5 vs K FLT ); however, neither phosphorylation rate‐constant correlates significantly with SUV. EM‐BIC clustering reduces the parametric maps to a small number of levels—on average 5.8, 3.5, 3.4, and 1.4 for K FLT‐ 2 tiss, K FLT, k 3 ‐ 2 tiss, and k 5. This large simplification is potentially useful for radiotherapy dose‐painting, but demonstrates the high noise in some maps. Statistical simulations show that voxel level noise degrades TACs generated from the 3C6K model sufficiently that the average AIC score, parameter bias, and total uncertainty of 2C4K model fits are similar to those of 3C6K fits, whereas at the whole tumor level the scores are lower for 3C6K fits. Conclusions: : For the patients studied here, whole tumor FLT uptake time‐courses are represented better overall by a three‐tissue than by a two‐tissue model. EM‐BIC clustering simplifies noisy parametric maps, providing the best description of the underlying information they contain and is potentially useful for radiotherapy dose‐painting. However, the clustering highlights the large degree of noise present in maps of the phosphorylation rate‐constants k 5 and k 3 ‐ 2 tiss, which are conceptually tightly linked to cellular proliferation. Methods must be found to make these maps more robust—either by constraining other model parameters or modifying dynamic imaging protocols.