Page 209 - Quantitative Imaging of Small Tumours with Positron Emission Tomography
P. 209

                                Chapter 11 tumour boundaries (i.e. the volumes of interest defined on PET) are highly susceptible to delineation accuracy. Therefore, such methods may worsen rather than improve accuracy. We recommend the use of parametric (voxel-wise) PVC methods such as iterative deconvolution, which performed rather well regardless of tumour delineation accuracy. An advantage of an iterative deconvolution algorithm is that it is easily implemented and computationally efficient since it can be applied to standard reconstructed clinical images (3,4). It does, however, require that the PSF setting is calibrated for individual PET-CT systems. This might be easily performed using the phantom images that are routinely acquired for the EARL calibration (5). A disadvantage of such an algorithm is that it tends to amplify image noise, mandating some form of image denoising when SUVmax is used (6). PVC in dynamic PET: worth the trouble? In Chapter 3 we evaluated the impact of PVC in a dynamic pharmacokinetic [18F]FLT PET-CT study in NSCLC patient treated with tyrosine kinase inhibitors (7). Full pharmacokinetic studies are needed to technically validate simplified quantitative metrics of tracer uptake on PET (e.g. SUV or TBR) (8-11). In this regard, accuracy has a somewhat different meaning than it had it the previous paragraph. Here, the ground truth is the reference pharmacokinetic parameter (distribution volume, VT), and accuracy pertains to the correlation between VT and the simplified parameters. PVC may affect both differently, and thus needs to be considered in such pharmacokinetic PET studies. We observed that during a dynamic PET acquisition, the PVE changes over time, depending on tracer kinetics in blood and tumour. Also, the denoising algorithm required specific optimization to prevent that the temporal course of PVE was omitted (12). As expected, PVC increased both tumour kinetic parameters and SUV/TBR reads. However, as the effects of PVC on both these parameters were quite similar, it did not substantially change their correlation coefficients (despite having a significant impact on absolute values). Therefore, PVC improved nor worsened the accuracy of SUV and TBR taking VT as reference for response assessment purposes. This study provided valuable, and perhaps reassuring, insight into the impact of PVE on technical validation of simplified PET metrics for clinical use, indicating that it may not be of value for future full pharmacokinetic response monitoring validation studies in oncology. It should be noted that these findings only apply to relative treatment-induced changes in measured tumour parameters. If absolute values, 208 


































































































   207   208   209   210   211