Evaluation of A Direct Regional Reprojection Reconstruction Method to Improve Quantitation of Amyloid and Tau PET Images using 3D Printed Phantoms

Abstract

AbstractBackgroundImage reconstruction and partial volume correction (PVC) techniques designed to account for the limited spatial resolution of PET are used inconsistently in amyloid and tau PET imaging, likely due to a lack of characterization of the techniques and their unknown effect on reliability. In this study, we aimed to test the accuracy and reliability of a variety of reconstruction and image‐based correction methods using 3D‐printed PET phantoms.MethodTwo phantoms modeled on the lateral temporal brain region of two human subjects (67‐year‐old male and 69‐year‐old female, both cognitively normal) were created using 3D‐printing technology (Formlabs Form 3). The phantoms had six thin‐walled chambers corresponding to inferior, middle, and superior temporal grey matter (GM) and white matter (WM). Phantoms were scanned (Siemens Biograph mCT) multiple times as chambers were filled with F‐18 solution to mimic various distributions of activity seen in amyloid‐positive, amyloid‐negative, or tau‐positive subjects. 16 scans were completed in total. Data from each 3‐hour scan were binned into 16 statistically equivalent frames and reconstructed with three techniques: FBP, OSEM, and TrueX, a manufacturer‐supplied method including point spread function modeling (Figure 1). Two image‐based correction techniques were also applied and evaluated: Geometric Mean Transfer Matrix (GTM) PVC and a novel Direct Regional Reprojection Reconstruction (DRRR) method, an iterative algorithm that reprojects images directly into regions‐of‐interest. To evaluate accuracy, fractional error of each region (F = M/T ‐ 1, where M is the measured regional value averaged over 16 frames and T is the true regional value) was calculated. To evaluate reliability, the coefficient of variation (CV) of M over all 16 frames in each scan was assessed for every region of the phantoms.ResultFractional errors (Figure 2) and CVs (Figure 3) are shown for each reconstruction and image‐based correction method. Uncorrected TrueX fractional error was not apparently improved relative to uncorrected FBP and OSEM fractional error. For all reconstructions, GTM and DRRR improved accuracy relative to uncorrected. For the FBP and OSEM reconstructions, GTM was more accurate than DRRR, but DRRR was more reliable than GTM.ConclusionDRRR is a promising novel method for PET quantification that improves accuracy without compromising reliability.