Abstract
1446
Introduction: The higher sensitivity of state-of-the-art PET scanners and the implementation of advanced reconstruction algorithms may allow for reduced injected activities while maintaining excellent image quality. However, [18F]FDG PET/CT images can be degraded by differences in counting statistics caused by patient anatomical characteristics such as body mass, lean body mass (LBM), or surface area (SA). Poor image quality can prevent the detection of small lesions and raise the rate of false negatives. Scaling the time-activity product (TAP) (i.e. the product of injected activity and bed duration) is one way to mitigate degradation in image quality (Everaert, 2003). The aim of this work was to achieve higher quantification accuracy and consistent image quality for all patients independent of their anatomy. This was done by 1) determining the optimum scaling of TAP based on patient anatomical characteristics and 2) optimizing PET reconstruction parameters.
Methods: A NEMA phantom was scanned utilizing list-mode. Images were reconstructed with different parameters (i.e. number of iterative updates, post-smoothing filter) using (i) OSEM and (ii) BSREM (referred to as Q.Clear) algorithms, and with different scan durations. The reconstructed images were assessed for noise and quantitative bias. Parameters that resulted in low bias of recovery coefficients and low noise (COV) were regarded as the best. The minimum scan duration to achieve a threshold of 10% COV was then determined for both the best reconstruction parameters and the parameters used in the patient study (next). A possible scaling factor to reduce the TAP was calculated by taking the ratio of the minimum scan durations for these 2 reconstructions. PET/CT images of a cohort of 200 patients that received an [18F]FDG scan were analyzed after approval by the ethics board at our institution. An equal number of patients were scanned on a 5-ring Discovery MI and a D690 PET/CT scanner (both GE Healthcare, USA). Images were reconstructed using OSEM with and without TOF with 2 iterations, 34 subsets, 6.4mm FWHM Gaussian smoothing, and PSF correction. Image quality was quantified by measuring signal-to-noise ratio in the liver (SNRL). The relationship between SNRL, TAP, and several patient anatomical characteristics was modeled using a power function (de Groot, 2013). A new TAP protocol was then derived using a fixed target SNRL according to: TAP = A*t = (SNRL/a)2*p2b, where p is a patient anatomical characteristic (body mass, LBM, SA, BMI, mass per length, fat mass), and a and b are fit parameters from the power function. The TAP scaling factor derived from the NEMA study was then used to recommend a reduced TAP protocol.
Results: The optimal reconstruction parameters were found to be (i) TOF OSEM with 3 iterations, 8 subsets, 3.2mm Gaussian filter, and (ii) Q.Clear with β = 600. The analysis of patient data showed average COV levels of ~10% in the liver. Body mass (m) was the parameter that showed the highest correlation between TAP and image quality (SNRL). Our results suggest that to maintain consistent image quality, the TAP should be scaled as m1.37 and m1.88 for the Discovery MI and D690 scanners, respectively. With our optimized reconstruction parameters, the TAP could potentially be reduced by 32.7% while maintaining COV of 10% for the reconstructed images.
Conclusions: This work suggests that to maintain a uniform level of image quality between different patients, the TAP should be scaled using body mass with a power function that follows a relation in-between linear and quadratic, and is scanner specific. Using current reconstruction parameters, the TAP for a 75 kg patient is 518 MBq*min. With the optimized reconstruction parameters, (i) Q.Clear with β = 600 (QFX), and (ii) TOF OSEM (FXS) 3i 8s 3.2mm FWHM, the TAP can be decreased to 409 MBq*min and 348 MBq*min, a reduction of ~21% and ~33%, respectively. This would allow for a reduction of dose delivered to patients, an increase in patient throughput, or a combination of the two.