Abstract
Introduction Positron Emission Tomographie (PET), coupled to X-ray Computed Tomography (PET-CT), is widely employed in the diagnosis and the therapeutical follow-up of some cancers. Technological advancements have led to Time Of Flight-PET (TOF-PET) technology that considerably improves the image quality, sensitivity and quantification of events. In combination with iterative tomographic reconstruction, the quality of reconstructed images has been largely improved. The Ordered Subset Expectation Maximization algorithm, coupled to TOF (TOF-OSEM) is usually employed in clinical routine. It led to an enhanced contrast but amplifies noise in the image. Thus, the iterations are usually stop arbitrarily before convergence, creating biases in quantification. It has been demonstrated that introduction of a smoothness contraint in the objective function limits high frequencies. This method is implemented by General Electrics (GE) in the Q.Clear [1] reconstruction algorithm, using a Bayesian penalty term, that assigns a low probability to images containing large differences in neighbouring pixels. This algorithm improves activity quantification, while reducing noise in the reconstructed image. Methods We propose to first compare different reconstruction algorithms, available on a PET-CT installation (MI-DR, GE), namely the VPFX (TOF-OSEM) and Q.Clear with a varying penalization weight. Then, we study the influence of tumor volume segmentation, using manual segmentation, relative and absolute SUV thresholding and adaptative thresholding [2] . The influence of each reconstruction and segmentation parameter is evaluated in terms of tumoral volume definition, on 12 patients suffering from a Non-Small Cell Lung Cancer, as well as on a calibration phantom (The NEMA IEC BodyPhantom Set™). Results As expected, the variability background and the signal-to-noise ratio are improved using iterative reconstruction algorithm, especially with high penalty weight. The Total Lesion Glycolysis (TLG) remains steady whatever the thresholding and reconstruction methods. The adaptive thresholding seems to improve tumoral volume segmentation. We noticed that using an adaptive thresholding method enable to take into account the reconstruction method, and therefore better reflect tumoral volume, with respect to classical thresholding. Conclusions Improvement in image quality using iterative reconstruction algorithms has been largely demonstrated, but classical thresholding methods do not integrate this trend in tumoral volume. We show the efficiency of adaptive thresholding that adapts better to the reconstruction method employed. Moreover, this adaptive thresholding is relatively easy to implement in clinical routine.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call