Clinical impact of digital and conventional PET control databases for semi-quantitative analysis of brain 18F-FDG digital PET scans

Abstract

Purpose Digital PET cameras markedly improve sensitivity and spatial resolution of brain 18F-FDG PET images compared to conventional cameras. Our study aimed to assess whether specific control databases are required to improve the diagnostic performance of these recent advances. Methods We retrospectively selected two groups of subjects, twenty-seven Alzheimer's Disease (AD) patients and twenty-two healthy control (HC) subjects. All subjects underwent a brain 18F-FDG PET on a digital camera (Vereos, Philips®). These two group (AD and HC) are compared, using a Semi-Quantitative Analysis (SQA), to two age and sex matched controls acquired with a digital PET/CT (Vereos, Philips®) or a conventional PET/CT (Biograph 6, Siemens®) camera, at group and individual levels. Moreover, individual visual interpretation of SPM T-maps was provided for the positive diagnosis of AD by 3 experienced raters. Results At group level, SQA using digital controls detected more marked hypometabolic areas in AD (+ 116 cm3 at p < 0.001 uncorrected for the voxel, corrected for the cluster) than SQA using conventional controls. At the individual level, the accuracy of SQA for discriminating AD using digital controls was higher than SQA using conventional controls (86% vs. 80%, p < 0.01, at p < 0.005 uncorrected for the voxel, corrected for the cluster), with higher sensitivity (89% vs. 78%) and similar specificity (82% vs. 82%). These results were confirmed by visual analysis (accuracies of 84% and 82% for digital and conventional controls respectively, p = 0.01). Conclusion There is an urgent need to establish specific digital PET control databases for SQA of brain 18F-FDG PET images as such databases improve the accuracy of AD diagnosis.

Moving the goalposts while scoring―the dilemma posed by new PET technologies

“…New metrics that may be more suited for advanced reconstructions, e.g., based on SUVpeak for lesional uptake [ 56 , 72 ], should also be explored in such clinical trial datasets. Besides Deauville scores and other established diagnostic cut-offs [ 65 ], such studies may address the influence on prognostic cut-offs [ 58 ] or on normal databases [ 91 , 92 ]. Using either approach, non-inferiority of the new standard regarding clinical endpoints must be well proven, before it can be used safely to guide crucial clinical decisions.…”

“…• Image noise [75][76][77][78] • Subjective image quality [73,77,[79][80][81][82][83][84] • Prediction of subjective image quality from objectified measures [85] • Lesion SUV [75,86,87] • Conspicuity of pulmonary lesions [69,77,86] • Detection rate in pulmonary lesions [88] • Lesion detectability in pelvic PSMA PET/MRI at low activity [89] • Detection rate of lymph nodes in fluorocholine PET [90] • Normal databases for brain PET in neurodegenerative disease [91,92] • Acquisition time reduction [81,84,89,93] Level 2:…”

Effect of Point Spread Function Deconvolution in Reconstruction of Brain 18F-FDG PET Images on the Diagnostic Thinking Efficacy in Alzheimer's Disease

“…However, this particular study did not perform reconstructions on the normal database which was provided by the software. This is in contrast to our own study which exclusively compared populations with the same PSF status, given that our normal database was reconstructed with or without PSF depending on the nature of the population investigated (5). It has been previously suggested that PET images should be smoothed to the same low spatial resolution to assess comparability between images (10).…”

“…We selected three groups of subjects that had undergone brain 18 F-FDG PET on a digital camera (Vereos, Philips ® ) with a similar acquisition protocol and reported reconstruction protocol derived from the guidelines ( 5 , 6 ). Briefly, brain 18 F-FDG PET scans were recorded over a 15 min one bed acquisition, 45–50 min after injection of 2 MBq/kg of 18 F-FDG.…”

“…Brain 18 F-FDG PET images were pre-processed using SPM12 (London, UK). All 18 F-FDG PET images, reconstructed with or without the PSF deconvolution, were spatially normalized with an adaptive CT template using CT attenuation corrected images and then corrected for partial volume effects ( 5 ). The pons region was used as reference for intensity normalization ( 8 ).…”