Abstract
371
Objectives Dynamic PET studies can be used for quantification of the uptake and binding of the hypoxia tracer [18F]-fluoromisonidazole (FMISO) in tumours. Long acquisition times are needed, as the tracer molecules typically need a long time to diffuse from the blood vessels to the hypoxic regions, where the tracer binds irreversibly in viable hypoxic cells. Time-activity curves (TACs) obtained for volumes of interest (VOIs) can be analysed using compartmental models and non-linear optimisation techniques. An irreversible 2-tissue compartment model with 3 rate-constants (2TC3k) has previously been used for FMISO data. We have compared this model with a 3TC5k model. For voxel-based analysis, it may be advantageous to use a faster algorithm, such as spectral analysis (SA) or Patlak analysis (PA). PA only produces two output parameters; the irreversible uptake rate-constant, Ki, and the volume of distribution for reversible tracer uptake, Vd. The SA output can be used to derive additional parameters; the blood-to-tissue uptake rate-constant, K1, and the irreversible binding rate-constant, k3 (assuming a 2TC3k model). SA has previously mainly been used for reversible tracers, and the algorithm will not necessarily produce an irreversible output component in the presence of noise or motion artifacts. Therefore we have developed a novel approach, combining SA and PA; we call it spectral Patlak analysis (SPLA). The objectives of this work were to optimise the methodology for kinetic analysis of FMISO PET data from colorectal tumours on a VOI- and on a voxel-basis.
Methods Nine patients with colorectal cancer were scanned after i.v. injection of 18F-FMISO in a PET/CT scanner. Each patient had three dynamic PET scans, starting at the time of injection and ~100 and ~200 min p.i., respectively. The data from the 3 scans were co-registered based on the CT images. Image derived arterial input functions were obtained for kinetic analyses using different approaches. VOIs were defined on the CT images as well as on K1- and Ki-maps obtained by voxel-based analysis using SA and PA. VOI-TACs were then analysed with the 2TC3k and the 3TC5k models, as well as with the novel SPLA approach. We compared the output from the different methods using correlation analysis.
Results The results are presented in the table below. There was a good correlation in terms of Ki and Vd between the 2TC3k and the 3TC5k models. However, the correlation was poor for K1, both in terms of slope and R2 value, suggesting that the former model is too simple to produce accurate values of K1. On the other hand, more robust estimates of k3 were obtained with the 2TC3k as compared to the 3TC5k model. The SPLA parameters had a reasonably good correlation with the 3TC5k model for Ki, Vd and K1 and with the 2TC3k model for k3.
Conclusions Our results show that the 2TC3k model does not provide accurate K1 values, so the more complex 3TC5k model is needed. The novel spectral-Patlak analysis method can be useful for voxel-based analysis. This work was supported by NIHR UCLH-BRC, CRUK KCL/UCL-CIC and GE Healthcare.