Image Derived Input Functions for Cerebral PET Studies SUMMARY

This thesis describes the development and evaluation of a non-invasive method for obtaining the input function. Extraction of arterial input functions from dynamic brain scans may obviate the need for arterial sampling and would increase the clinical applicability of quantitative PET studies. Different strategies for extracting image derived input functions (IDIF) from dynamic PET images were evaluated. To correct for partial volume effects, a reconstruction based partial volume correction (PVC) method was implemented. Optimal settings for extracting the input function and for the reconstruction based partial volume correction were evaluated for a number of tracers with different kinetics. Methods were evaluated for both standard and high resolution scanners. For all tracers, image derived input functions were compared with measured arterial input functions. As IDIFs are sensitive to patient motion, effects of this motion, together with different motion correction methods, were evaluated.

In chapter 2, the evaluation of different methods for extracting image derived input functions from dynamic [ 11 C]flumazenil PET scans is described. Optimal settings were determined for a reconstruction based PVC method that was used for extracting the image derived input functions. Best results were obtained from reconstruction based PVC images, using 4 iterations, 16 subsets and a resolution kernel of 4.5 mm full width at half maximum (FWHM), which represents the point spread function of the scanner. It was demonstrated that a region of interest (ROI) consisting of the four hottest pixels per plane (over the carotid arteries) was the best method to extract IDIFs. Excellent peak area under the curve (AUC) ratios (0.99 ± 0.09) between IDIF and blood sampler input function (BSIF) were found. In addition, extracted IDIFs provided volume of distribution (V T ) and K 1 values that were very similar to those obtained using blood sampler input functions (BSIF). The proposed method appeared to be suitable for analysing [¹¹C]flumazenil data without the need for on-line arterial sampling.

In chapter 3, an evaluation of the optimal image derived input function extraction method for [ 11 C]flumazenil PET brain studies obtained using the high resolution research tomography (HRRT) is provided. As these PET images have higher spatial resolution, optimal settings for extracting IDIFs and for the reconstruction based PVC were determined again. In addition, the impact of high resolution on accuracy of image derived input functions was assessed. Good peak AUC ratios (0.83 ± 0.21) between IDIF and BSIF were found for IDIFs that had been extracted from standard HRRT reconstructed images and that had been scaled (calibrated) to manual samples. In addition, good slope values (1.07 ± 0.11) were found. Improved resolution, as obtained with PVC reconstruction, changed AUC ratios between BSIFs and IDIFs to 1.14 ± 0.35 and, for tracer kinetic analysis, slopes changed to 0.95 ± 0.13. For all reconstructions, non-scaled IDIFs gave poorer results (>61 ± 34% higher slopes) compared with calibrated IDIFs. The results of this study indicated that the use of IDIFs, extracted from OP-OSEM or PVC OP-OSEM images, is also feasible for dynamic HRRT data, thereby obviating the need for on-line arterial sampling.

In chapter 4, the robustness of the image derived input function extraction method, developed in chapter 2, was assessed for three additional tracers ([ 11 C]PIB, (R )-[ 11 C]verapamil and ( R )-[ 11 C]PK11195) with varying brain uptake. Again, IDIFs derived from PVC reconstructed images were compared with BSIFs using AUC ratios and outcome of tracer kinetic analyses. For ( R )-[ 11 C]verapamil, accurate IDIFs were obtained (slope: 0.96 ± 0.17; R 2 : 0.92 ± 0.07) without scaling to manual samples. However, scaling was necessary to make IDIFs comparable to BSIFs for both [ 11 C]PIB (slope: 1.04 ± 0.05; R 2 : 1.00 ± 0.01) and ( R )-[ 11 C]PK11195 (slope: 0.96 ± 0.05; R 2 : 0.99 ± 0.01). The need for calibration may be due to stickiness of both tracers, in which case BSIFs may have been affected by sticking. Nevertheless, results of this study showed that the method to extract image derived input functions is also suitable for [ 11 C]PIB, ( R )-[ 11 C]verapamil and ( R )-[ 11 C]PK11195 studies, thereby obviating the need for on-line arterial sampling.

In chapter 5, the accuracy of reconstruction based PVC was validated using both phantom and clinical studies. Phantom and healthy volunteers were scanned on both HR+ and HRRT scanners and were reconstructed using three different reconstruction methods, including a reconstruction based partial volume correction method. For phantom data, recovery of the spheres was calculated by measured radioactivity divided by true radioactivity. For clinical studies, parametric V T images were generated for both scanners and all reconstruction methods, and V T values for 14 different anatomical regions were compared with each other. For phantom data, good recovery was found for both (standard) HR+ (0.84 to 0.97) and (high resolution) HRRT (0.91 to 0.98) reconstructions. In addition, for the HR+, very good recovery was found for PVC-OSEM reconstructions (0.84 to 0.94), which corresponded well with results found for standard HRRT OP-OSEM reconstructions. In contrast, recovery of standard NAW-OSEM reconstructions was reduced (0.42 to 0.86). For clinical data, good correspondence was found between HR+ PVC-OSEM and HRRT OP-OSEM derived V T values (slope: 1.02 ± 0.08). HR+ image resolution using PVC-OSEM was comparable to the resolution of the HRRT scanner. Outcome of tracer kinetic analysis of HR+ studies reconstructed with PVC-OSEM correlated well with outcome of HRRT studies, indicating that reconstruction based partial volume correction yields quantitatively accurate images.

In chapter 6, four different off-line frame-by-frame methods to correct for patient motion were evaluated. These methods differed in the way realignment parameters were derived. Two simulation studies were performed, based on [ 11 C]flumazenil and (R) -[ 11 C]PK11195 datasets, respectively. For both simulation studies, different types (rotational, translational) and degrees of motion were added. Simulated PET scans were corrected for motion using all correction methods. The optimal method derived from these simulation studies was used to evaluate two (one with and one without visible movement) clinical datasets of [ 11 C]flumazenil, (R) -[ 11 C]PK11195 and [ 11 C]PIB. For each dataset, V T was derived using Logan analysis and values were compared before and after motion correction. For both [ 11 C]flumazenil and (R) -[ 11 C]PK11195 simulations, optimal results were obtained when realignment was based on non-attenuation corrected images. For the clinical datasets motion disappeared visually after motion correction. Regional differences of up to 433% in V T before and after motion correction were found for scans with visible movement. On the other hand, when no visual motion was present in the original dataset, overall differences in V T before and after motion correction were <1.5 ± 1.3%. In conclusion, frame-by-frame motion correction using non-attenuation corrected images improved accuracy of tracer kinetic analyses compared to non-motion corrected data.

In chapter 7 the effects of motion affected IDIFs on outcome of tracer kinetic analyses were quantified. Again, two simulation studies, one based on [ 11 C]flumazenil (high cortical uptake), the other on (R) -[ 11 C]verapamil (low cortical uptake), were performed. Different degrees of rotational and axial translational motion were added to the final frames of simulated dynamic PET scans. Extracted IDIFs from motion affected simulated scans were compared to original IDIFs and to outcome of tracer kinetic analysis (volume of distribution V T ). Differences in IDIF values of up to 239% were found for the last frames. Patient motion of more than 6º or 5 mm resulted in at least 10% higher or lower V T values for the low cortical tracer. The degrees of motion studied are commonly observed in clinical studies and hamper the extraction of accurate IDIFs. Therefore, it is essential to ensure that patient motion is minimal and corrected for.