Dataset Information

Poisson noise obscures hypometabolic lesions in PET.

ABSTRACT: The technology of fluoro-deoxyglucose positron emission tomography (PET) has drastically increased our ability to visualize the metabolic process of numerous neurological diseases. The relationship between the methodological noise sources inherent to PET technology and the resulting noise in the reconstructed image is complex. In this study, we use Monte Carlo simulations to examine the effect of Poisson noise in the PET signal on the noise in reconstructed space for two pervasive reconstruction algorithms: the historical filtered back-projection (FBP) and the more modern expectation maximization (EM). We confirm previous observations that the image reconstructed with the FBP biases all intensity values toward the mean, likely due to spatial spreading of high intensity voxels. However, we demonstrate that in both algorithms the variance from high intensity voxels spreads to low intensity voxels and obliterates their signal to noise ratio. This finding has profound impacts on the clinical interpretation of hypometabolic lesions. Our results suggest that PET is relatively insensitive when it comes to detecting and quantifying changes in hypometabolic tissue. Further, the images reconstructed with EM visually match the original images more closely, but more detailed analysis reveals as much as a 40 percent decrease in the signal to noise ratio for high intensity voxels relative to the FBP. This suggests that even though the apparent spatial resolution of EM outperforms FBP, the signal to noise ratio of the intensity of each voxel may be higher in the FBP. Therefore, EM may be most appropriate for manual visualization of pathology, but FBP should be used when analyzing quantitative markers of the PET signal. This suggestion that different reconstruction algorithms should be used for quantification versus visualization represents a major paradigm shift in the analysis and interpretation of PET images.

SUBMITTER: Kerr WT

PROVIDER: S-EPMC3516894 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:In the recently revised diagnostic criteria for Alzheimer disease (AD), the National Institute on Aging and Alzheimer Association suggested that confidence in diagnosing dementia due to AD and mild cognitive impairment (MCI) due to AD could be improved by the use of certain biomarkers, such as (18)F-FDG PET evidence of hypometabolism in AD-affected brain regions. Three groups have developed automated data analysis techniques to characterize the AD-related pattern of hypometabolism in a single measurement. In this study, we sought to directly compare the ability of these three (18)F-FDG PET data analysis techniques--the PMOD Alzheimer discrimination analysis tool, the hypometabolic convergence index, and a set of meta-analytically derived regions of interest reflecting AD hypometabolism pattern (metaROI)--to distinguish moderate or mild AD dementia patients and MCI patients who subsequently converted to AD dementia from cognitively normal older adults.One hundred sixty-six (18)F-FDG PET patients from the AD Neuroimaging Initiative, 308 from the Network for Efficiency and Standardization of Dementia Diagnosis, and 176 from the European Alzheimer Disease Consortium PET study were categorized, with masking of group classification, as AD, MCI, or healthy control. For each AD-related (18)F-FDG PET index, receiver-operating-characteristic curves were used to characterize and compare subject group classifications.The 3 techniques were roughly comparable in their ability to distinguish each of the clinical groups from cognitively normal older adults with high sensitivity and specificity. Accuracy of classification (in terms of area under the curve) in each clinical group varied more as a function of dataset than by technique. All techniques were differentially sensitive to disease severity, with the classification accuracy for MCI due to AD to moderate AD varying from 0.800 to 0.949 (PMOD Alzheimer tool), from 0.774 to 0.967 (metaROI), and from 0.801 to 0.983 (hypometabolic convergence index).The 3 tested techniques have the potential to help detect AD in research and clinical settings. Additional efforts are needed to clarify their ability to address particular scientific and clinical questions. Their incremental diagnostic value over other imaging and biologic markers makes them easier to implement by other groups for these purposes.

Project description:IntroductionOne major challenge in PET radiomics is its sensitivity to noise. Low signal-to-noise ratio (SNR) affects not only the precision but also the accuracy of quantitative metrics extracted from the images resulting in noise-induced bias. This phantom study aims to identify the radiomic features that are robust to noise in terms of precision and accuracy and to explore some methods that might help to correct noise-induced bias.MethodsA phantom containing three 18F-FDG filled 3D printed inserts, reflecting heterogeneous tracer uptake and realistic tumor shapes, was used in the study. The three different phantom inserts were filled and scanned with three different tumor-to-background ratios, simulating a total of nine different tumors. From the 40-minute list-mode data, ten frames each for 5 s, 10 s, 30 s, and 120 s frame duration were reconstructed to generate images with different noise levels. Under these noise conditions, the precision and accuracy of the radiomic features were analyzed using intraclass correlation coefficient (ICC) and similarity distance metric (SDM) respectively. Based on the ICC and SDM values, the radiomic features were categorized into four groups: poor, moderate, good, and excellent precision and accuracy. A "difference image" created by subtracting two statistically equivalent replicate images was used to develop a model to correct the noise-induced bias. Several regression methods (e.g., linear, exponential, sigmoid, and power-law) were tested. The best fitting model was chosen based on Akaike information criteria.ResultsSeveral radiomic features derived from low SNR images have high repeatability, with 68% of radiomic features having ICC ≥ 0.9 for images with a frame duration of 5 s. However, most features show a systematic bias that correlates with the increase in noise level. Out of 143 features with noise-induced bias, the SDM values were improved based on a regression model (53 features to excellent and 67 to good) indicating that the noise-induced bias of these features can be, at least partially, corrected.ConclusionTo have a predictive value, radiomic features should reflect tumor characteristics and be minimally affected by noise. The present study has shown that it is possible to correct for noise-induced bias, at least in a subset of the features, using a regression model based on the local image noise estimates.

Project description:BACKGROUND:Routine PET exams are increasingly performed with reduced injected activities, leading to the use of different image reconstruction parameters than the NEMA parameters, in order to prevent from any deleterious decrease in signal-to-noise ratio (SNR) and thus, in lesion detectability. This study aimed to provide a global head-to-head comparison between digital (Vereos, Philips®) and analog (Ingenuity TF, Philips®) PET cameras of the trade-off between SNR and contrast through a wide-ranging number of reconstruction iterations, and with a further reconstruction optimization based on the SNR of small lesions. METHODS:Image quality parameters were compared between the two cameras on human and phantom images for a number of OSEM reconstruction iterations ranging from 1 to 10, the number of subsets being fixed at 10, and with the further identification of reconstruction parameters maximizing the SNR of spheres and adenopathies nearing 10?mm in diameter. These reconstructions were additionally obtained with and without time-of-flight (TOF) information (TOF and noTOF images, respectively) for further comparisons. RESULTS:On both human and phantom TOF images, the compromise between SNR and contrast was consistently more advantageous for digital than analog PET, with the difference being particularly pronounced for the lowest numbers of iterations and the smallest spheres. SNR was maximized with 1 and 2 OSEM iterations for the TOF images from digital and analog PET, respectively, whereas 4 OSEM iterations were required for the corresponding noTOF images from both cameras. On the TOF images obtained with this SNR optimization, digital PET exhibited a 37% to 44% higher SNR as compared with analog PET, depending on sphere size. These relative differences were however much lower for the noTOF images optimized for SNR (-?4 to +?18%), as well as for images reconstructed according to NEMA standards (-?4 to +?12%). CONCLUSION:SNR may be dramatically higher for digital PET than for analog PET, especially when optimized for small lesions. This superiority is mostly attributable to enhanced TOF resolution and is significantly underestimated in NEMA-based analyses.

Project description:Several investigators have shown that noise equivalent count rate (NECR) is linearly proportional to the square of image signal-to-noise ratio (SNR) when PET images are reconstructed using filtered back-projection. However, to our knowledge, none have shown a similar relationship in fully 3D ordered-subset expectation maximization (OSEM) reconstruction. This paper has two aims. The first is to investigate the NECR-SNR relationship for 3D-OSEM reconstruction using phantom studies while the second aim is to evaluate the NECR-SNR relationship using patient data.An anthropomorphic phantom was scanned on a GE Discovery-STE (DSTE) PET∕CT scanner in 3D mode with an initial activity concentration of 66.34 kBq∕cc. PET data were acquired over the lower chest∕upper abdomen region in dynamic mode. The experiment was repeated with the same activity concentration on a GE Discovery-RX (DRX) scanner. Care was taken to place the phantom at identical positions in both scanners. PET data were then reconstructed using 3D Reprojection (3D-RP) and 3D-OSEM with different reconstruction parameters and the NECR and SNR for each frame∕image were calculated. SNR(2) was then plotted versus the NECR for each scanner, reconstruction method and parameters. In addition, 40 clinical PET∕CT studies from the two scanners (20 patients∕scanner) were evaluated retrospectively. The patient studies from each scanner were further divided into two subgroups of body mass indices (BMI). Each PET study was acquired in 3D mode and reconstructed using both 3D-OSEM and 3D-RP. The NECR and SNR of the bed position covering the patient liver were calculated for each patient and averaged for each subgroup. Comparisons of the NECR and SNR between scanner types and BMIs were performed using a t-test and a p value less than 0.05 was considered significant.Phantom results showed that SNR(2) versus NECR was linear for 3D-RP reconstruction across all activity concentration on both scanners, as expected. However, when 3D-OSEM was used, this relationship was nonlinear at activity concentrations beyond the peak NECR on both scanners. On the other hand, the plot of SNR(2) versus trues count rate was linear for 3D-OSEM across all activity concentrations on both scanners independent of reconstruction parameters used. In addition, for activity concentrations <30kBq∕cc, phantom results showed a higher SNR (by 12 ± 10%; p < 0.05) and NECR for the DRX scanner compared to DSTE for 3D-RP reconstruction. However, for 3D-OSEM reconstruction, these two scanners had similar SNRs (different by 2% ± 9%; p > 0.05), despite having different NECRs. Patient studies showed a statistically significant difference in NECR as well as the SNR for 3D-RP reconstruction between the two scanners. However, no statistically significant difference was found for 3D-OSEM. A statistically significant difference in both NECR and SNR were found between the different BMI subgroups for both 3D-RP and 3D-OSEM reconstructions.For the scanners and reconstruction algorithm used in this study, our results suggest that the image SNR cannot be predicted by the NEC when using 3D-OSEM reconstruction particularly for those clinical applications requiring high activity concentration. Instead, our results suggest that image SNR varies with activity concentration and is dominated by the 3D-OSEM reconstruction algorithm and its associated parameters, while not being affected by the scanner type for the range of activity concentrations usually found in the clinic.

Dataset Information

Poisson noise obscures hypometabolic lesions in PET.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets