Abstract
3284
Introduction: Positron emission tomography (PET) tracers targeting prostate specific membrane antigen (PSMA) allows for enhanced detection of metastatic prostate cancer. Accurate assessment of features related to PSMA PET images, such as tumour uptake or volume, may lead to improved disease management and prediction of patient outcome. However, PET images are highly influenced by the image acquisition and reconstruction methods. Ordered subsets expectation maximization (OSEM) is conventionally used to reconstruct images, but a trade-off between image noise and image convergence exists depending on the number of iterations. As a result, post-reconstruction filters are often used to “smooth” the image, but this comes at the expense of reduced image contrast and diagnostic sensitivity. Within this study, we aim to evaluate how the number of iterations and smoothing filter size affects lesion quantitation in PSMA PET imaging. We will use our Canadian PET Phantom for Prostate Oncology (C3PO) with embedded radioactive epoxy spheres to simulate prostate cancer metastasis. We focus on selection of reconstruction parameters for PSMA PET quantitation, though our work is also being applied to harmonize images between cancer centres.
Methods: To simulate prostate cancer metastasis, 3mm-16mm diameter (0.014mL-2.14ml) epoxy spheres were infused with a long half-life positron emitter (22Na, 50 kBq/mL). Tumour models were inserted into C3PO (without background activity) and imaged using the GE DMI PET/CT scanner (1x10min scan), to determine ground truth radioactivity of the 22Na spheres. To simulate a patient image, three 2min scans were performed with [18F]FDG injected into the bladder and background (80 kBq/mL and 1.5 kBq/mL, respectively). Images were reconstructed using OSEM (1-10 iterations, 32 subsets) with point-spread function modelling (PSF). 2-8mm Gaussian filters were applied and compared to the “Native” image without smoothing. Lesions were segmented using 41% of SUVmax fixed threshold (41% FT), as computed using Python based on masks delineated with 3D Slicer software. Recovery coefficients were computed by dividing the measured radioactivity by the ground truth. Mean absolute error (MAE) and percent noise (PN) was computed for each set of parameters (varying iterations and filter size).
Results: To determine number of OSEM iterations required for PSMA PET image convergence, MAE±PN was computed for 1-10 iterations (taken as the average of all filter sizes). MAE±PN was 51.1±2.2%, 47.5±2.3%, 44.8±2.6%, and 43.4±3.0%, for 1, 2, 4, and 8 iterations, respectively. While comparing the unfiltered, 3mm, 5mm, and 8mm-smoothed images (OSEM with 4 iterations), MAE±PN was 29.9±3.6%, 30.4±3.5%, 49.9±2.2%, and 71.3±1.1%, respectively. To compare SUVmean accuracy for different lesion sizes, MAE and MPN were calculated for specific lesion sizes. MAE±PN for 10-16mm lesions was 4.0±3.0%, 3.9±3.0%, 1.9±2.6%, and 0.6±2.1%, while MAE±PN for 3-8mm lesions was 45.7±4.0%, 46.7±3.9%, 66.4±1.9%, and 85.3±0.6%, for the unfiltered, 3mm, 5mm, and 8mm-smoothed images.
Conclusions: These results suggest reconstructing OSEM images with 3-4 iterations and applying a 2-4mm filter, to achieve accurate PSMA PET quantification of prostate cancer metastasis. Our findings differ from our currently implemented clinical protocol, which performs 2 iterations with 6.4mm filter. In a clinical setting, it is still likely that two sets of reconstruction parameters will provide the best performance – one for optimized image quality and another for lesion quantification. Using the 41% FT applied to OSEM with 4 iterations, we observed that accurate quantification of larger lesions (10-16mm) can be achieved with filter sizes up to 8mm, while accuracy for smaller lesions (3-8mm) is only maintained for filter sizes up to 3mm. In future studies, we will extend the capabilities of C3PO for multi-centre harmonization of PSMA PET imaging.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.