Use Of Point Spread Function Research Articles

Adaptive Optics Image Restoration via Regularization Priors With Gaussian Statistics

In order to compensate for any failure on the use of point spread function (blur kernel) estimation and image estimation priors, we propose a novel regularization priors scheme with adapting the parameter for image restoration involving adaptive optics (AO) images. Our scheme uses a maximum a posteriori estimation with Gaussian statistics on the image and point spread function (blur kernel). An efficient regularization prior method associated with alternating minimization method is described to obtain the optimal solution recursively. Our method is applied to synthetic and real adaptive optics images. After applying our restoration method, satisfying results are obtained. Experimental results demonstrate that our proposed model and method performs better for restoring images in terms of both subjective results and objective assessments than the current state-of-the-art restoring methods. In addition, our proposed method can be a new way to promote their performances for AO image restoration.

Optimized Feature Extraction for Radiomics Analysis of 18F-FDG PET Imaging.

Radiomics analysis of 18F-FDG PET/CT images promises well for an improved in vivo disease characterization. To date, several studies have reported significant variations in textural features due to differences in patient preparation, imaging protocols, lesion delineation, and feature extraction. Our objective was to study variations in features before a radiomics analysis of 18F-FDG PET data and to identify those feature extraction and imaging protocol parameters that minimize radiomic feature variations across PET imaging systems. Methods: A whole-body National Electrical Manufacturers Association image-quality phantom was imaged with 13 PET/CT systems at 12 different sites following local protocols. We selected 37 radiomic features related to the 4 largest spheres (17-37 mm) in the phantom. On the basis of a combined analysis of voxel size, bin size, and lesion volume changes, feature and imaging system ranks were established. A 1-way ANOVA was performed over voxel size, bin size, and lesion volume subgroups to identify the dependency and the trend change in feature variations across these parameters. Results: Feature ranking revealed that the gray-level cooccurrence matrix and shape features are the least sensitive to PET imaging system variations. Imaging system ranking illustrated that the use of point-spread function, small voxel sizes, and narrow gaussian postfiltering helped minimize feature variations. ANOVA subgroup analysis indicated that variations in each of the 37 features and for a given voxel size and bin size can be minimized. Conclusion: Our results provide guidance to selecting optimized features from 18F-FDG PET/CT studies. We were able to demonstrate that feature variations can be minimized for selected image parameters and imaging systems. These results can help imaging specialists and feature engineers in increasing the quality of future radiomics studies involving PET/CT.

Abstracts of the 33rd International Austrian Winter Symposium

K. Binzel1, A. Adelaja1, C. L. Wright1, D. Scharre2, J. Zhang1, M. V. Knopp11Wright Center of Innovation in Biomedical Imaging, Department of Radiology, The Ohio State University Wexner Medical Center, Columbus, OH, USA; 2Department of Neurology, The Ohio State University Wexner Medical Center, Columbus, OH, USACorrespondence: K. Binzel Aim: Digital photon counting allows for higher definition image reconstruction than on conventional PET systems due to increased system sensitivity, time of flight timing resolution, and count densities (1). For neurologic applications, these improvements may greatly increase the utility and accuracy of PET imaging. We validated and optimized high definition image reconstruction methodologies using the Hoffman brain phantom. Methods: A Hoffman brain phantom was filled with 52 MBq 18F-FDG and imaged on a pre-commercial release digital photon counting PET/CT (Philips Vereos). Five minute acquisitions were repeated over several runs using the dedicated brain 256 mm field of view (FOV) and the whole body 576 mm FOV. Listmode data from each acquisition was reconstructed using the high definition (HD) 2x2x2 mm voxel matrix. Reconstructions were performed with 3 iterations and a range of subsets, 21, 17, 13, and 9. Additionally, the use of system point spread function (PSF) correction and/or a Gaussian filter were enabled. Regions of interest (ROIs) were placed in 10 unique regions of the phantom for quantitative assessment. Blinded reader reviews were performed to assess image quality. Results: We found that the optimal reconstruction settings for each FOV were distinct. For the 576 mm FOV the use of PSF alone, no Gaussian filter, gave the most accurate quantitative results. The addition of the Gaussian filter resulted in underestimation of activity concentrations. The average recovery coefficients (RCs) of all ROIs were very similar among reconstructions with different numbers of subsets. The average RCs were 0.97, 0.95, 0.94, and 0.92 for the 21, 17, 13, and 9 subsets reconstructions, respectively. Blinded review conveyed that the 13 subset images were most preferable with regard to contrast and image noise. Thus the HD reconstruction with PSF only using 3 iterations and 13 subsets was optimal for the whole body FOV acquisitions. For the 256 mm FOV a Gaussian filter was used in reconstruction as the PSF alone lead to overestimation of activity concentrations. As with the whole body FOV, each subset setting gave similar quantitative results, average RCs were 1.02, 1.01, 0.98, and 0.95 for the 21, 17, 13, and 9 subset reconstructions. Image review again showed that fewer subsets were preferred, with the 9 subset reconstruction now being most ideal. For acquisitions with the dedicated brain FOV, HD reconstruction with both PSF correction and a Gaussian filter using 3 iterations with 9 subsets is optimal for imaging with FDG. Conclusion: Optimization of Neuro-PET reconstruction settings revealed that these setting must be tailored to acquisition characteristics, namely the chosen field of view. Phantom validation demonstrated that high quantitative accuracy and excellent image quality is readily achieved for neurologic imaging on next-generation digital PET systems when using either a dedicated brain field of view or the wider whole body field of view. Reference 1. Wright CL, Binzel K, Zhang J, Knopp MV. Advanced Functional Tumor Imaging and Precision Nuclear Medicine Enabled by Digital PET Technologies. Contrast Media & Molecular Imaging. 2017;2017:5260305. https://doi.org/10.1155/2017/5260305.

Does PET Reconstruction Method Affect Deauville Score in Lymphoma Patients?

When evaluating 18F-FDG PET images with the Deauville score (DS), the quantification of tumor and reference organs limits the problem of optical misinterpretation. Compared with conventional reconstruction algorithms, point-spread function (PSF) modeling increases SUVs significantly in tumors but only moderately in the liver, which could affect the DS. We investigated whether the choice of the reconstruction algorithm affects the DS and whether discordance affects the capability of 18F-FDG PET to stratify lymphoma patients. Methods: Overall, 126 patients with diffuse large B-cell lymphoma were included (56 female and 70 male; median age, 65 y; range, 20-88 y). PET data were reconstructed with the unfiltered PSF method. Additionally, a 6-mm filter was applied to PSF images to meet the requirements of the EANM Research Ltd. (EARL) harmonization program from the European Association of Nuclear Medicine (EANM) (PSFEARL). One hundred interim PET (i-PET) and 95 end-of-treatment PET (EoT-PET) studies were analyzed. SUVmax in the liver and aorta was determined using automatic volumes of interest and compared with SUVmax in the residual mass with the highest 18F-FDG uptake. Results: For i-PET, using PSF and PSFEARL, we classified patients as responders and nonresponders in 60 and 40 cases versus 63 and 37 cases, respectively. Five cases of major discordance (5.0%) occurred (i.e., changes from responder to nonresponder). For Eot-PET, patients were classified using PSF and PSFEARL as responders and nonresponders in 69 and 26 cases versus 72 and 23 cases, respectively. Three cases of major discordance (3.2%) occurred. Concordance (Cohen unweighted κ) between the PSF and the PSFEARL DS was 0.82 (95% confidence interval, 0.73-0.91) for i-PET and 0.89 (95% confidence interval, 0.81-0.96) for EoT-PET. The median follow-up periods were 28.4 and 27.4 mo for i-PET and EoT-PET, respectively. Kaplan-Meier analysis showed statistically significant differences in progression-free survival and overall survival among responders and nonresponders no matter which reconstruction was used for i-PET and EoT-PET. Conclusion: Neither DS nor risk stratification of diffuse large B-cell lymphoma patients is affected by the choice of PET reconstruction. Specifically, the use of PSF is not an issue in routine clinical processes or in multicenter trials. These findings have to be confirmed in escalation and deescalation procedures based on early i-PET.

EORTC PET response criteria are more influenced by reconstruction inconsistencies than PERCIST but both benefit from the EARL harmonization program

BackgroundThis study evaluates the consistency of PET evaluation response criteria in solid tumours (PERCIST) and European Organisation for Research and Treatment of Cancer (EORTC) classification across different reconstruction algorithms and whether aligning standardized uptake values (SUVs) to the European Association of Nuclear Medicine acquisition (EANM)/EARL standards provides more consistent response classification.Materials and methodsBaseline (PET1) and response assessment (PET2) scans in 61 patients with non-small cell lung cancer were acquired in protocols compliant with the EANM guidelines and were reconstructed with point-spread function (PSF) or PSF + time-of-flight (TOF) reconstruction for optimal tumour detection and with a standardized ordered subset expectation maximization (OSEM) reconstruction known to fulfil EANM harmonizing standards. Patients were recruited in three centres. Following reconstruction, EQ.PET, a proprietary software solution was applied to the PSF ± TOF data (PSF ± TOF.EQ) to harmonize SUVs to the EANM standards. The impact of differing reconstructions on PERCIST and EORTC classification was evaluated using standardized uptake values corrected for lean body mass (SUL).ResultsUsing OSEMPET1/OSEMPET2 (standard scenario), responders displayed a reduction of −57.5% ± 23.4 and −63.9% ± 22.4 for SULmax and SULpeak, respectively, while progressing tumours had an increase of +63.4% ± 26.5 and +60.7% ± 19.6 for SULmax and SULpeak respectively. The use of PSF ± TOF reconstruction impacted the classification of tumour response. For example, taking the OSEMPET1/PSF ± TOFPET2 scenario reduced the apparent reduction in SUL in responding tumours (−39.7% ± 31.3 and −55.5% ± 26.3 for SULmax and SULpeak, respectively) but increased the apparent increase in SUL in progressing tumours (+130.0% ± 50.7 and +91.1% ± 39.6 for SULmax and SULpeak, respectively).Consequently, variation in reconstruction methodology (PSF ± TOFPET1/OSEMPET2 or OSEM PET1/PSF ± TOFPET2) led, respectively, to 11/61 (18.0%) and 10/61 (16.4%) PERCIST classification discordances and to 17/61 (28.9%) and 19/61 (31.1%) EORTC classification discordances. An agreement was better for these scenarios with application of the propriety filter, with kappa values of 1.00 and 0.95 compared to 0.75 and 0.77 for PERCIST and kappa values of 0.93 and 0.95 compared to 0.61 and 0.55 for EORTC, respectively.ConclusionPERCIST classification is less sensitive to reconstruction algorithm-dependent variability than EORTC classification but harmonizing SULs within the EARL program is equally effective with either.

Evaluation of the Effect of Tumor Position on Standardized Uptake Value Using Time-of-Flight Reconstruction and Point Spread Function.

Objective(s):The present study was conducted to examine whether the standardized uptake value (SUV) may be affected by the spatial position of a lesion in the radial direction on positron emission tomography (PET) images, obtained via two methods based on time-of-flight (TOF) reconstruction and point spread function (PSF).Methods:A cylinder phantom with the sphere (30 mm diameter), located in the center was used in this study. Fluorine-18 fluorodeoxyglucose (18F-FDG) concentrations of 5.3 kBq/ml and 21.2 kBq/ml were used for the background in the cylinder phantom and the central sphere respectively. By the use of TOF and PSF, SUVmax and SUVmean were determined while moving the phantom in a horizontal direction (X direction) from the center of field of view (FOV: 0 mm) at 50, 100, 150 and 200 mm positions, respectively. Furthermore, we examined 41 patients (23 male, 18 female, mean age: 68±11.2 years) with lymph node tumors, who had undergone 18F-FDG PET examinations. The distance of each lymph node from FOV center was measured, based on the clinical images.Results:As the distance of a lesion from the FOV center exceeded 100 mm, the value of SUVmax, which was obtained with the cylinder phantom, was overestimated, while SUVmean by TOF and/or PSF was underestimated. Based on the clinical examinations, the average volume of interest was 8.5 cm3. Concomitant use of PSF increased SUVmax and SUVmean by 27.9% and 2.8%, respectively. However, size of VOI and distance from the FOV center did not affect SUVmax or SUVmean in clinical examinations.Conclusion:The reliability of SUV quantification by TOF and/or PSF decreased, when the tumor was located at a 100 mm distance (or farther) from the center of FOV. In clinical examinations, if the lymph node was located within 100 mm distance from the center of FOV, SUV remained stable within a constantly increasing range by use of both TOF and PSF. We conclude that, use of both TOF and PSF may be helpful.

Evaluation of bias and variance in low-count OSEM list mode reconstruction

Statistical algorithms have been widely used in PET image reconstruction. The maximum likelihood expectation maximization reconstruction has been shown to produce bias in applications where images are reconstructed from a relatively small number of counts. In this study, image bias and variability in low-count OSEM reconstruction are investigated on images reconstructed with MOLAR (motion-compensation OSEM list-mode algorithm for resolution-recovery reconstruction) platform. A human brain ([11C]AFM) and a NEMA phantom are used in the simulation and real experiments respectively, for the HRRT and Biograph mCT. Image reconstructions were repeated with different combinations of subsets and iterations. Regions of interest were defined on low-activity and high-activity regions to evaluate the bias and noise at matched effective iteration numbers (iterations × subsets). Minimal negative biases and no positive biases were found at moderate count levels and less than 5% negative bias was found using extremely low levels of counts (0.2 M NEC). At any given count level, other factors, such as subset numbers and frame-based scatter correction may introduce small biases (1–5%) in the reconstructed images. The observed bias was substantially lower than that reported in the literature, perhaps due to the use of point spread function and/or other implementation methods in MOLAR.

Benefits of Point-Spread Function and Time of Flight for PET/CT Image Quality in Relation to the Body Mass Index and Injected Dose

The PET image quality of overweight patients and patients who receive low injected doses deteriorates because of increases in statistical noise. The purpose of this study was to investigate the benefits of the point-spread function (PSF) and time-of-flight (TOF) for PET/CT image quality in such patients. The PET images were reconstructed using the baseline ordered-subsets expectation-maximization algorithm (OSEM), OSEM + PSF, OSEM + TOF, and OSEM + PSF + TOF. In the phantom study, we used a National Electrical Manufacturers Association body phantom with different radioactivity concentrations and analyzed image quality using the coefficient of variance in the background (CVphantom). In the clinical study, we retrospectively studied 39 patients who underwent clinical F-FDG PET/CT. The patients were classified into groups based on body mass index and injected dose. Image quality was evaluated using the CV in the liver (CVliver). In the phantom study, PSF and TOF improved the CVphantom, especially in low-activity models. Among all of the reconstructions, the best CVphantom was obtained with OSEM + PSF + TOF. In the clinical study, the CVliver of the low-dose group with OSEM + PSF + TOF was comparable to that of the high-dose group with conventional OSEM. Point-spread function and TOF improved PET/CT image quality for overweight patients who received a lower injected dose. Therefore, the use of PSF and TOF is suggested to maintain the image quality of such patients without extending scanning times. It is greatly beneficial to obtain sufficient image quality for larger patients, especially in delivery institutions where the injection dose cannot be easily increased.

Harmonizing SUVs in multicentre trials when using different generation PET systems: prospective validation in non-small cell lung cancer patients

PurposeWe prospectively evaluated whether a strategy using point spread function (PSF) reconstruction for both diagnostic and quantitative analysis in non-small cell lung cancer (NSCLC) patients meets the European Association of Nuclear Medicine (EANM) guidelines for harmonization of quantitative values.MethodsThe NEMA NU-2 phantom was used to determine the optimal filter to apply to PSF-reconstructed images in order to obtain recovery coefficients (RCs) fulfilling the EANM guidelines for tumour positron emission tomography (PET) imaging (PSFEANM). PET data of 52 consecutive NSCLC patients were reconstructed with unfiltered PSF reconstruction (PSFallpass), PSFEANM and with a conventional ordered subset expectation maximization (OSEM) algorithm known to meet EANM guidelines. To mimic a situation in which a patient would undergo pre- and post-therapy PET scans on different generation PET systems, standardized uptake values (SUVs) for OSEM reconstruction were compared to SUVs for PSFEANM and PSFallpass reconstruction.ResultsOverall, in 195 lesions, Bland-Altman analysis demonstrated that the mean ratio between PSFEANM and OSEM data was 1.03 [95 % confidence interval (CI) 0.94–1.12] and 1.02 (95 % CI 0.90–1.14) for SUVmax and SUVmean, respectively. No difference was noticed when analysing lesions based on their size and location or on patient body habitus and image noise. Ten patients (84 lesions) underwent two PET scans for response monitoring. Using the European Organization for Research and Treatment of Cancer (EORTC) criteria, there was an almost perfect agreement between OSEMPET1/OSEMPET2 (current standard) and OSEMPET1/PSFEANM-PET2 or PSFEANM-PET1/OSEMPET2 with kappa values of 0.95 (95 % CI 0.91–1.00) and 0.99 (95 % CI 0.96–1.00), respectively. The use of PSFallpass either for pre- or post-treatment (i.e. OSEMPET1/PSFallpass-PET2 or PSFallpass-PET1/OSEMPET2) showed considerably less agreement with kappa values of 0.75 (95 % CI 0.67–0.83) and 0.86 (95 % CI 0.78–0.94), respectively.ConclusionProtocol-optimized images and compliance with EANM guidelines allowed for a reliable pre- and post-therapy evaluation when using different generation PET systems. These data obtained in NSCLC patients could be extrapolated to other solid tumours.Electronic supplementary materialThe online version of this article (doi:10.1007/s00259-013-2391-1) contains supplementary material, which is available to authorized users.

Accurate 3D Fluorophore Localization using Phase Retrieved Pupil Functions

When near the focus, the image of a point emitter has little information about the emitter's position along the optical axis. To improve the 3D localization precision, several methods such as astigmatic imaging, double helix point spread functions, and multiple focal plane strategies have been developed to encode the axial position in one or more 2D images of the same emitter. In each of these strategies, a proper model for the 3D microscope point spread function (PSF) is essential to avoid inaccuracies in localization. Generating PSF models from phase retrieved (PR) pupil functions has the advantage that the model images are noise free, can be calculated for any focal plane, and include aberrations specific to the optical system. This representation can also be extremely compact by representing the pupil function as a set of Zernike polynomials. Here we demonstrate the use of PSFs generated from PR pupil functions for 3D single molecule localization based super-resolution imaging using a dual focal plane setup. We show the phase retrieval process, as well as the details of fast PSF calculation by implementation on graphics processing unit hardware. We show a comparison between fitting with ideal, aberration free PSF models and that generated from PR pupil functions using a variety of cellular structures including the ER, microtubules and the membrane proteins STIM1 and ORAI1.

Clinical use of the ocular point spread function for retinal image quality assessment

The purpose of this study is to demonstrate the clinical use of the point spread function (PSF) as metric for the assessment of retinal image quality in eyes with different ocular conditions. The PSF was computed from the wavefront aberrations obtained in seven eyes with different ocular conditions: emmetropia, myopia, hyperopia, post-myopic laser in situ keratomileusis (LASIK), post-hyperopic LASIK, keratoconus, Intacs® in keratoconus, radial keratotomy (RK), LASIK after RK, cataract, phacoemulsification after cataract surgery, and LASIK after astigmatic keratotomy. The Strehl intensity ratio was calculated for all the PSFs computed. The PSF was correlated with the change in the optical elements of the eye: cornea and lens. Corneal refractive surgery, keratoconus, Intacs insertion and cataract surgery induce different changes in the optics of the eye which were well reflected by the PSF and the Strehl intensity ratio. Spread of the PSF in each case was caused by optical aberrations and/or scatter. The PSF measurement and the Strehl intensity ratio calculation is useful for the qualitative assessment of the retinal image and for quantifying optical degradation imposed by different optical conditions.