If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
A novel figure of merit to investigate 68Ga PET/CT image quality based on patient weight and lesion size using Q.Clear reconstruction algorithm: A phantom study
Department of Medical Physics, IRCCS Azienda Ospedaliero-Universitaria di Bologna, 40138 Bologna, ItalyMedical Physics Specialization School, Alma Mater Studiorum, University of Bologna, 40138 Bologna, Italy
Department of Medical Physics, IRCCS Azienda Ospedaliero-Universitaria di Bologna, 40138 Bologna, ItalyMedical Physics Specialization School, Alma Mater Studiorum, University of Bologna, 40138 Bologna, Italy
The β-values of the Q.Clear algorithm affect the 68Ga PET/CT image quality.
•
The new figure of merit (CRBV), incorporating both CR and BV behaviours, identifies the optimal β-values for image quality.
•
Q.Clear algorithm improves the image quality with respect to the OSEM with few exceptions.
•
CRBV suggests different β-values for normal weight and overweight patient-like phantoms.
•
CRBV identifies different β-values according to the sphere diameter.
Abstract
Introduction
Q.Clear is a Bayesian penalised-likelihood algorithm that uses a β-value for positron emission tomography(PET)/computed tomography(CT) image reconstruction(IR). Our study proposes a novel figure of merit, named CRBV, to compare the Q.Clear performances using 68Ga PET/CT image with the ordered-subset-expectation–maximization(OSEM) algorithm and to identify the optimal β-values for these images using two phantoms mimicking normal and overweight patients.
Methods
NEMA IQ phantom with or without a ring of water-filled plastic bags (NEMAstd and NEMAow, respectively) was acquired and reconstructed with OSEM and Q.Clear at various β-values and minutes/bed position(min/bp). Contrast recovery(CR), background variability(BV) and CRBV were calculated. Highest CRBV values were used to identify optimal β-value ranges.
Results
Q.Clear with 250 ≤ β ≤ 800 improved CRBV compared to OSEM for all the investigated spheres and acquisition setups. Outside of this range, Q.Clear still outperformed OSEM with few exceptions depending on spheres diameters and phantoms(e.g.,β-value = 1600 for diameters ≤ 17 mm using the NEMAow phantom). Regarding the CRBV performance for IR optimization, for the 4 min/bp NEMAstd IR, β-values = 300 ÷ 350 allowed to simultaneously optimize all diameters(except for the 10 mm); for the NEMAow IR, β-values = 350 ÷ 500 were needed for diameters > 20 mm, while β-values = 200 ÷ 250 were selected for the remaining diameters. For the 2 min/bp, β-value = 500 was suitable for diameters > 17 mm in both NEMAstd and NEMAow IR, while for smaller diameters β-value = 200 and β-values = 250 ÷ 350 were obtained for NEMAstd and NEMAow, respectively.
Conclusion
Almost all tested β-values of Q.Clear improved the CRBV compared to OSEM. In both phantoms, simulating normal and over-weight patients, optimal β-values were found according to lesion sizes and investigated acquisition times.
68Ga labelled radiopharmaceuticals are widely used in Positron Emission Tomography (PET)/Computed Tomography (CT) acquisition for the diagnosis and for the decision of therapy regimens of several tumour types, including neuroendocrine tumours (NET) and prostate cancer (PCa). The investigation of NET tumours is achieved by binding the 68Ga radionuclide with different structural linkers and somatostatin surrogates [
]. Increased prostate-specific membrane antigen (PSMA) expression of PCa can be investigated with 68Ga-PSMA, obtained through the radiosynthesis of various PSMA ligands, such as 68Ga-PSMA-11, 68Ga-PSMA-617 and 68Ga-PSMA-I&T [
], image reconstruction algorithms should be investigated to assess the appropriate settings that may need specific adjustments for this category of patients.
In the last years, a new Bayesian penalized-likelihood (BPL) image reconstruction algorithm, named Q.Clear, has been implemented by GE Healthcare [
Optimization of image reconstruction for (90)Y selective internal radiotherapy on a lutetium yttrium orthosilicate PET/CT System Using a Bayesian penalized likelihood reconstruction algorithm.
] and determines the level of penalization of the relative differences between neighbouring pixels. This allows for full convergence of the reconstruction algorithm, overcoming the limitation of the ordered subset expectation maximization (OSEM) approach, in which the image noise increases with each iteration and the algorithm must be stopped before obtaining the full convergence [
The impact of time-of-flight, resolution recovery, and noise modelling in reconstruction algorithms in non-solid-state detectors PET/CT scanners: - multi-centric comparison of activity recovery in a 68Ge phantom.
]. With the Q.Clear algorithm low β-values will generally result in images reconstructed with high noise and sharp contrast, while high β-values will suppress image noise but could result in excessive smoothing [
Comparison of regularized reconstruction and ordered subset expectation maximization reconstruction in the diagnostics of prostate cancer using digital time-of-flight (68)Ga-PSMA-11 PET/CT imaging.
Optimization of image reconstruction for (90)Y selective internal radiotherapy on a lutetium yttrium orthosilicate PET/CT System Using a Bayesian penalized likelihood reconstruction algorithm.
] purposes. As PET-tracer, nevertheless, 68Ga exhibits different characteristics with respect to 18F, such as the emitted positron energy spectrum and the mean positron range in various tissues [
], thus requiring specific image quality evaluation and optimization strategies to retrieve the optimal scenario to perform an accurate diagnosis, as defined by international guidelines [
]. Several studies related to 68Ga-specific image acquisition and reconstruction protocols have therefore been proposed in the last years. For example, Q.Clear parameter optimization for 68Ga based on visual evaluation on patients’ cohorts was performed for various clinical settings, including PCa assessment with 68Ga-PSMA and 68Ga-citrate on various PET/CT [
Comparison of regularized reconstruction and ordered subset expectation maximization reconstruction in the diagnostics of prostate cancer using digital time-of-flight (68)Ga-PSMA-11 PET/CT imaging.
Evaluation of quantitative Ga-68 PSMA PET/CT repeatability of recurrent prostate cancer lesions using both OSEM and Bayesian penalized likelihood reconstruction algorithms.
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
]. Unfortunately, this methodology depends on the experience of one or more operators and on the investigated cohort of patients and led to a wide variety of suggested β-values, ranging from β = 500 to β = 1600.
Moreover, Standardized Uptake Value (SUV)-based semi-quantitative measurements on patients were extensively reported [
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
Quantification of FDG PET studies using standardised uptake values in multi-centre trials: effects of image reconstruction, resolution and ROI definition parameters.
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
For these reasons, a quantitative evaluation based on a phantom modelling is useful for an objective assessment of the algorithm performance. A phantom-based evaluation on the optimal β-parameter for 68Ga image reconstruction with Q.Clear algorithm has been reported on PET/CT [
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
] that can be assessed in terms of contrast recovery (CR) and background variability (BV) as defined in the NEMA NU-2018 protocol. In these studies, since there is a great variety among the image quality-based figures of merit, such as signal-to-noise ratio (SNR), signal-to-background ratio (SBR), and image noise, that cannot typically be simultaneously optimized, the results are always a trade-off between two or more measurements (e.g., Lindstrom et al. [
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
] identified the Q.Clear image reconstruction that was able to achieve the same noise level as OSEM with an increase of SNR or SBR.
Moreover, NEMA IQ phantom, although one of the most available phantom across European PET/CT /MRI centres as reported by an international survey by the PET/CT/MR Quality Control EFOMP Working group [
Impact of target-to-background ratio, target size, emission scan duration, and activity on physical figures of merit for a 3D LSO-based whole body PET/CT scanner.
] proposed a modification of the NEMA IQ phantom by adding a ring of plastic bags filled with water to assess the impact of a larger scattering fraction due to the additional material on the image quality of 3D lutetium oxyorthosilicate (LSO)-based whole-body 18F PET/CT scans.
To overcome the problems related to the variety of figures of merits in the selection of the optimal PET image reconstruction, the primary aim of this study is to propose a new figure of merit, called CRBV, as a new quantitative image reconstruction quality index. The CRBV incorporates by definition both the CR and BV behaviours versus the reconstruction parameters. This figure of merit was used to compare the Q.Clear performances with the ordered subset expectation maximization (OSEM) algorithm and as an optimization strategy to identify the optimal β-values for 68Ga PET/CT images. A NEMA IQ phantom and a modified version similarly to Matheoud et al. [
Impact of target-to-background ratio, target size, emission scan duration, and activity on physical figures of merit for a 3D LSO-based whole body PET/CT scanner.
], also called NEMA IQ phantom, was used for the technical image quality of the PET/CT image acquisitions. The NEMA IQ phantom is constituted by six fillable spheres of various diameters (37-, 28-, 22–, 17-, 13- and 10 mm) in a homogeneous background of approximately 9.8 L and a solid insert, called lung insert.
The phantom was prepared with a 68GaCl solution obtained from a 68Ge-68Ga generator by filling each sphere with an activity concentration of 111 kBq/ml surrounded by a phantom background activity concentration of 14 kBq/ml, achieving a hot spheres-to-background ratio of approximately 7.9:1, as performed by Krokos et al. [
]. Of note, this activity ratio, which is higher with respect to the procedure described in the NEMA NU-2018 protocol, was conservatively chosen to better mimic the typically higher tumour-to-background ratios that are encountered in 68Ga-PSMA [
Quantitative 3D assessment of (68)Ga-DOTATOC PET/MRI with diffusion-weighted imaging to assess imaging markers for gastroenteropancreatic neuroendocrine tumors: preliminary results.
] and to allow the comparison with pre-existing literature. A total of 143 MBq was used, similarly to the recommended injected activity in clinical acquisitions [
]. This NEMA IQ phantom standard configuration will be referred to as NEMAstd (Fig. 1a). The same phantom was surrounded by a mean layer of about 6 cm of water-filled plastic bags, to simulate the effect of additional water-like tissues in terms of scattering and attenuation in reconstructed images of overweight patients (NEMAow, similarly to the phantom setup proposed in [
Impact of target-to-background ratio, target size, emission scan duration, and activity on physical figures of merit for a 3D LSO-based whole body PET/CT scanner.
], reported in Fig. 1b). This resulted in a NEMAow volume of approximately 20 L. To estimate the patient’s BMI simulated by the NEMAstd and the NEMAow, the same methodology described by Matheoud et al. [
] was followed. In brief, the effective diameter was estimated by measuring the latero-lateral and the antero-posterior diameter of the NEMAow in the CT image. The BMI was therefore obtained through the relationship reported in [
The PET/CT images were acquired on a 3-rings Discovery MI PET/CT system (GE Healthcare, Milwaukee, USA) with the acquisition protocol adopted in clinical practice for 68Ga-DOTANOC patients. The protocol consisted in two beds acquisitions, 4 min/bed position (min/bp), 70 cm diameter field of view (FOV), 256x256 matrix size. Two acquisitions (i.e., one for the NEMAstd and one for the NEMAow setup) were performed in list-mode and the PET images were additionally reconstructed at 2 min/bp.
The PET images of both phantom configurations were reconstructed using the clinical image reconstruction protocol which is based on an OSEM algorithm with 4 iterations, 8 subsets, 6 mm gaussian filter. In addition, Q.Clear image reconstruction algorithm was investigated at various β-values (i.e., 50, 100, 150, 200, 250, 300, 350, 500, 800, 1000, and 1600). For both algorithms, time of flight (TOF) data were used.
All the CT images were acquired in helical mode with a rotation time of 0.6 s, slice thickness of 3.75 mm, pitch of 1.735, large body FOV, 100 kV, 300 mA, and reconstructed with GE Standard (STD)/AR40 algorithm. Attenuation (based on CT images), scatter and random coincidences, detector efficiency and dead time corrections were applied in all image sets.
Image quality assessment
The PET images were imported in a software for image analysis (MIM Maestro, MIM Software Inc., Cleveland, OH) for the delineation of regions of interest (ROIs) and statistics extraction. Image quality assessment was performed following the NEMA NU 2–2018 protocol. In brief, a circular ROI for each hot sphere with the same diameter of the nominal inner diameter of the sphere was drawn. The circular ROIs were drawn by the MIM software at the spatial resolution of the CT images; PET images voxels that lied on the sphere edges were therefore partially included in the ROIs according to the software interpolation algorithm. All circular ROIs were drawn in the same axial slice centred on the hot spheres (IAx,c). In IAx,c, and at about ±2 cm and ±1 cm from it, concentrical ROIs having the diameters of the six hot spheres in twelve different background regions were drawn; a total of 60 background ROIs were drawn for each sphere diameter. In Fig. 2, the ROIs placement for the image quality assessment is shown. Mean values and standard deviations of hot spheres and background ROIs, in terms of activity concentration in Bq/ml, were extracted in csv format for the data analysis.
Fig. 2Hot spheres and background ROI placement for image quality assessment of NEMAstd (upper panel) and NEMAow (lower panel) in axial (left), sagittal (centre) and coronal (right) views. The fusion between CT (grey colorbar) and PET Q.Clear with β = 1000 images (red colorbar) is reported. The background and circular ROI contours are presented with different colours based on their diameter. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Csv files with mean values and standard deviations, in terms of Bq/ml, of each hot spheres and background ROIs were imported and analysed with R software v 4.0.2 (R Core Team, Vienna).
Contrast recovery (CR) and background variability (BV), according to NEMA NU 2–2018 protocol, were calculated with an in-house developed script. Specifically, the CR was defined as:
where is the average ROI counts for a specified hot sphere, is the average counts of the background ROI relative to the same sphere, is the activity concentration in the hot spheres, and is the activity concentration in the background.
BV was defined as:
where is the standard deviation of the background ROI counts.
In the following, CRd,β and BVd,β will indicate the CR and BV obtained for the sphere of diameter d and the image reconstruction algorithm with the penalty term regulated by β, where β = 0 indicates the OSEM reconstruction.
In addition to the above-defined figures of merit, an additional image quality index, called CRBVd,β, was proposed which allows to simultaneously incorporate both CR and BV indexes by the following definition:
This function has been derived in analogy with P+ function from Agren et al. [
] which was introduced for the radiobiological modelling and describes the probability of uncomplicated tumour control as a function of the probability of tumour control and the probability of injury :
By definition, CR, BV, and CRBVd,β fall in the range [0,1] and will be expressed as percentage values (%) in the following sections and figures to facilitate reading.
CRBVd,β was computed for all OSEM- and Q.Clear-based image reconstructions in both phantom configurations. A total of 24 image sets (two phantoms’ configurations, one OSEM and 11 Q.Clear image reconstructions) were analysed. For each phantom configuration and hot sphere, the β-value that maximized the CRBV index was identified.
To assess the uncertainties of the CRBV index for the 6 spheres, we evaluated the , , values by manually repositioning the above-described ROIs 10 time using the 2 min/bp NEMAstd and NEMAow acquisitions reconstructed with Q.Clear having β = 50, which represented the worst scenarios (i.e., the mostly affected by noise) among the investigated setups and reconstructions.
The relative uncertainty on CRBV was calculated as:
Where , , , and , , were respectively obtained as the standard deviation and the mean value of 10 repeated measurements.
The CRBVd,β variation from the maximum CRBV was obtained for each sphere diameter as follows:
The ranges of β-values for which the was lower than the estimated median relative uncertainty on CRBV (e.g., X%) were also reported for each phantom setup, acquisition time and sphere size. CR and BV variations over these ranges of β-values were indicated with CRX% and BVX%.
All the analyses were conducted using R which allows to automatically import all the exported files containing the statistical information generated using the MIM software.
Results
Phantom BMI analysis
The dAP measured on the CT images for the NEMAstd and the NEMAow were 22.8 cm and 32.0 cm respectively, while the dLL were 29.4 cm and 44.5 cm respectively. These resulted in deff of 26.1 cm and 37.7 cm and BMI of 20.4 and 38.6 kg/m2 for the NEMAstd and the NEMAow, respectively. These BMI values fall within the ranges of adult normal weight and grade 2 obesity, respectively.
OSEM and Q.Clear CRBVd,β performances
Twelve image reconstruction settings were investigated for both for NEMAstd and NEMAow phantoms and for 4 min/bp and 2 min/bp acquisitions. A total of 48 phantom image-sets were analysed.
Highest CRBVd,β values in OSEM-based image reconstruction were observed for the 37 mm diameter sphere, being 72.5 % and 55.4 % in NEMAstd and the NEMAow phantom configurations with 4 min/bp, while being 70.6 % and 54.5 % in NEMAstd and the NEMAow phantom configurations with 2 min/bp, respectively. On the contrary, as expected, lowest CRBVd,β values were obtained for the 10 mm diameter sphere: 25.1 % and 12.1 % CRBVd,β values were obtained in NEMAstd and NEMAow phantom configurations with 4 min/bp, while CRBVd,β resulted in 21.8 % and 12.9 % in NEMAstd and NEMAow phantom configurations with 2 min/bp, respectively.
In Q.Clear-based image reconstructions at 4 min/bp, CRBVd,β values ranged between 19.6 % (in d = 10 mm, β = 1600 configuration) and 76.2 % (d = 37 mm, β = 500) and between 8.8 % (d = 10 mm, β = 1600) and 58.5 % (d = 37 mm, β = 500) for the NEMAstd and the NEMAow phantom setups, respectively. For the image reconstruction at 2 min/bp, CRBVd,β values ranged between 16.6 % (in d = 10 mm, β = 1600 configuration) and 73.8 % (d = 37 mm, β = 800) and between 9.8 % (d = 10 mm, β = 1600) and 57.8 % (d = 37 mm, β = 800) for the NEMAstd and the NEMAow phantom setups, respectively.
The BPL algorithm-based image reconstruction showed overall superiority with respect to OSEM-based one. For the 4 min/bp setup, at least nine and seven Q.Clear image reconstructions outperformed the OSEM ones for each of the six spheres for NEMAstd and NEMAow, respectively. More in details, Q.Clear with 250 ≤ β ≤ 800 improved CRBVd,β with respect to OSEM for all the investigated spheres and phantom acquisition setups. Outside of this range, Q.Clear still outperformed OSEM with a limited number of exceptions at the edges of the investigated range of β-values depending on sphere diameter and phantom setup. For example, for the 22 mm diameter sphere CRBVd,β values of OSEM reconstruction were higher than Q.Clear with β ≤ 100 and β = 1600 for the NEMAow phantom setup and with β = 50 for the NEMAstd. Similarly, in the 2 min/bp setup, for all spheres higher CRBVd,β were obtained for at least seven Q.Clear image reconstructions with respect to the OSEM ones for both NEMAstd and NEMAow, while OSEM resulted in higher CRBVd,β performance for β ≤ 50 for NEMAow and β ≤ 100 for NEMAstd for the 22 mm diameter sphere. Detailed results are reported in supplementary materials (Tables S1, S2, S3).
The CRBVd,β values obtained for the six spheres of the NEMAstd and the NEMAow phantom setups with various OSEM and Q.Clear-based image reconstructions parameters are reported in Fig. 3 and Fig. 4 for the 4 min/bp and 2 min/bp, respectively. CR and BV values are separately reported in supplementary materials (Table S2 and S3, Figure S1 and S2).
Fig. 3CRBVd,β values of a) NEMAstd and b) NEMAow phantom from 4 min/bp PET images with Q.Clear image reconstructions (β = 50, 100, 150, 200, 250, 300, 350, 500, 800, 1000, 1600) and OSEM image reconstruction (4 iterations, 8 subset and 6 mm gaussian filter) for hot spheres with 37 mm diameter (dot-dashed lines with star), 28 mm diameter (dotted lines with empty squares), 22 mm diameter (dashed line with cross), 17 mm diameter (dashed line with filled squares), 13 mm diameter (dashed line with filled triangles) and 10 mm diameter (continuous line with filled circle).
Fig. 4CRBVd,β values of a) NEMAstd and b) NEMAow phantom from 2 min/bp PET images with Q.Clear image reconstructions (β = 50, 100, 150, 200, 250, 300, 350, 500, 800, 1000, 1600) and OSEM image reconstruction (4 iterations, 8 subset and 6 mm gaussian filter), for hot spheres with 37 mm diameter (dot-dashed lines with star), 28 mm diameter (dotted lines with empty squares), 22 mm diameter (dashed line with cross), 17 mm diameter (dashed line with filled squares), 13 mm diameter (dashed line with filled triangles) and 10 mm diameter (continuous line with filled circle).
At fixed β-values, for both acquisition time setups CRBVd,β increased with the sphere diameter. For example, in the 4 min/bp setup, for β = 500 CRBVd,β ranged between 33.9 % and 76.2 % between 10 mm and 37 mm diameter in the NEMAstd phantom, and between 13.1 % and 58.5 % in the NEMAow phantom; in the 2 min/bp setup, for β = 500, CRBV ranged between 28.0 % and 72.3 % between 10 mm and 37 mm diameter in the NEMAstd phantom and between 15.4 % and 57.5 % in the NEMAow phantom.
At fixed diameter, CRBVd,β showed similar trends for all the spheres at increasing β-values, which resulted in a CRBVd,β increase until reaching a plateau, followed by a decrease. Steeper increases and decreases were observed far from the CRBVd,β plateau. These trends were observed for PET images reconstructed at both 4 min/bp and 2 min/bp.
Q.Clear optimization based on CRBVd,β
Maximum values of CRBVd,β were observed at various β-values according to the evaluated sphere diameters and the NEMA phantom setup. In details, for the NEMAstd phantom images reconstructed at 4 min/bp CRBVd,β values maxima were found at β = 500 for the 37 mm and 28 mm diameter spheres (76.2 % and 70.2 %, respectively), at β = 350 for the 22 mm diameter sphere (61.5 %), at β = 300 for the 17 mm diameter sphere (61.1 %), at β = 250 for the 13 mm diameter sphere (48.1 %), and at β = 150 for the 10 mm diameter sphere (40.5 %). Similarly, for the NEMAow phantom setup CRBVd,β values maxima were found at β = 500 for the 37 mm and 28 mm diameter spheres (58.5 % and 53.2 %, respectively), at β = 350 for the 22 mm diameter sphere (47.8 %), at β = 300 for the 17 mm diameter sphere (45.0 %), at β = 250 for the 13 mm diameter sphere (24.8 %), and at β = 200 for the 10 mm diameter sphere (14.2 %).
For the NEMAstd phantom images reconstructed at 2 min/bp CRBVd,β values maxima were found at β = 800 for the 37 mm and 28 mm diameter spheres (73.8 % and 68.5 %, respectively), at β = 500 for the 22 mm diameter sphere (61.6 %), at β = 350 for the 17 mm diameter sphere (59.3 %), at β = 250 for the 13 mm diameter sphere (46.1 %), and at β = 150 for the 10 mm diameter sphere (32.8 %). Similarly, for the NEMAow phantom images reconstructed at 2 min/bp CRBVd,β values maxima were found at β = 800 for the 37 mm and 28 mm diameter spheres (57.8 % and 52.1 %, respectively), at β = 500 for the 22 mm diameter sphere (46.4 %), at β = 350 for the 17 mm diameter sphere (42.0 %), at β = 300 for the 13 mm diameter sphere (24.7 %), and at β = 250 for the 10 mm diameter sphere (16.7 %).
The uncertainty analysis for the 2 min/bp NEMAstd and NEMAow phantom acquisitions reconstructed with Q.Clear with β = 50 (representing the worst cases) led to a median[range] relative uncertainty on CRBV of 1.0 %[2.2 %, 0.4 %].
For each sphere, CRBVd,β plateaus around the maximum CRBVd,β were observed at varying β values according to the (i.e., the median relative uncertainty on CRBV). For the 4 min/bp image reconstructions, largest plateaus were found for the 37 mm diameter sphere (β = 300 ÷ 1000 for NEMAstd, β = 350 ÷ 1000 for NEMAow), while smallest plateaus were observed for the 10 mm diameter sphere (β = 100 ÷ 150 for NEMAstd, β = 150 ÷ 250 for NEMAow). For the 2 min/bp image reconstructions these trends in the plateaus were even more pronounced, as largest plateaus were found for the 37 mm diameter sphere (β = 500 ÷ 1600 for NEMAstd, β = 500 ÷ 1000 for NEMAow) while smallest plateaus were observed for the 10 mm diameter sphere (β = 150 ÷ 200 for NEMAstd, β = 200 ÷ 300 for NEMAow).
Fig. 5 graphically shows the range of β-values leading to CRBVd,β within the estimated relative uncertainty (i.e., 1 %) with respect to the maximum value plateaus (i.e., image reconstructions equivalent to the optimal one) for the various sphere diameters in NEMAstd and NEMAow phantom setup for the images reconstructed at 4 and 2 min/bp.
Fig. 5Graphical representation of the ranges of β-values leading to Q.Clear image reconstructions with within the estimated relative uncertainty (i.e., 1 %) from the maximum value for each sphere. In panel a) and b) are reported the results for the images reconstructed at 4 min/bp and 2 min/bp, respectively. NEMAstd and NEMAow phantom setup are indicated with yellow and blue bars, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
For the β-values that lead to image reconstruction with within the estimated relative uncertainty of 1.0 %, the variability of CR and BV in terms of absolute differences from the CR and BV associated to the β-value that maximized the CRBV depended on phantom setup, acquisition time and sphere size. The widest range of CR was found for the 28 mm diameter sphere in the NEMAstd phantom acquired with 2 min/bp, ranging from 74.5 % (β = 350) to 71.8 % (β = 1000) with a CR of 73.8 % at β = 500 corresponding to the maximum CRBV. In the same interval of β-values, the BV ranged from 8.8 % (β = 350) to 4.9 % (β = 1000), with a BV of 7.1 % at β = 500. Similarly, the widest range of BV was found for the 10 mm diameter sphere in the NEMAow acquired with 2 min/bp, ranging from 17.1 % (β = 200) to 14.6 % (β = 300) with a BV of 14.4 % at β = 250 in correspondence of the maximum CRBV, while CR ranged from 20.1 % (β = 200) to 19.0 % (β = 300) with a CR of 19.5 % at β = 250.
Overall, in the phantoms acquired with 2 min/bp the median[range] of CR1.0% were 0.00 [−0.94, 1.98] and 0.00 [−1.21, 1.32] for NEMAstd and NEMAow, respectively, while the median(range) of BV1.0% were 0.00 [−2.10, 2.70] and 0.00 [−2.73, 2.30]. Similarly, in the phantoms acquired with 4 min/bp the median[range] of CR1.0% were 0.00 [−1.38, 1.60] and 0.00 [−1.02, 1.58] for NEMAstd and NEMAow, respectively, while the median(range) of BV1.0% were 0.00 [−2.91, 1.61] and 0.00 [−2.47, 3.70].
Discussion
Image reconstruction is a crucial step for a correct interpretation of diagnostic imaging in PET/CT and PET/MR systems. Image reconstruction algorithms impact on image quality both in terms of signal recovery and noise and should be carefully calibrated to bring out the diagnostic content of an image. This is typically related to the diagnostic task and can differ among multiple settings, including tracer (e.g., 68Ga, 18F), radiopharmaceutical carrier (e.g., 68Ga-PSMA, 68Ga-DOTA), tumour (e.g., PCa, NET), examination (e.g., baseline disease staging, local recurrence) and patient type (e.g., normal weight, overweight patients).
In the last years, several studies reported on 68Ga tracers image reconstruction optimization with a commercial BPL algorithm, named Q.Clear. For what concerns image reconstruction, Q.Clear optimization is only based on the selection of the β parameter value that regulates the penalty term in the image reconstruction algorithm. The trend of signal recovery (which can be defined as contrast recovery, mean SUV, SUV max, SUV peak, and other similar metrics) and image noise (typically expressed as background variability in low and homogeneous uptake areas) at varying β values has been well characterized [
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
]: low β values lead to high recovery coefficients at a price of noisy images, while high β values allow to obtain smoother images, despite of accurate activity quantification. The same trend is confirmed in our phantom study.
Unfortunately, a straightforward strategy for image reconstruction optimization is not available and multiple approach have been proposed. Visual scoring methodologies have been reported on real patients imaged with 68Ga-PSMA [
Comparison of regularized reconstruction and ordered subset expectation maximization reconstruction in the diagnostics of prostate cancer using digital time-of-flight (68)Ga-PSMA-11 PET/CT imaging.
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
]. Different optimal β values were selected (ranging between 500 and 1600) across several settings and evaluation criteria. Main limitations of the visual scoring approach lie in the limited number of images (i.e., image reconstruction algorithm settings) that can be evaluated by a physician and by the tendency to compare the novel image reconstruction method to the standard OSEM-based one potentially introducing bias in the subjective judgment [
Quantitative approaches have been proposed on patients (including semiquantitative metrics such as mean SUV, maximum SUV and peak SUV on lesions identified by the physician, or background variability on low and homogenous uptake areas) and phantoms (evaluating recovery coefficients, contrast recovery of high activity concentration areas, background variability of background areas with low activity concentration, or contrast-to-noise ratio, CNR) to quantify the impact of the image reconstruction algorithm. These approaches have been widely reported [
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
]. Nevertheless, although patients-based quantitative metrics have the advantage of proposing various patients or lesions in commonly occurring locations, they suffer by the lack of ground truth [
]. On the other side, in Q.Clear-based images common quantitative metrics adopted for phantom studies do not always allow for reconstruction optimization by picking up one single β value, since signal recovery and image noise have monotonic trends (i.e., descending and ascending, respectively) at increasing β-values. For the benefit of the reader, a table has been added in the Supplementary Materials (Table S4) reporting the main studies on Q.Clear optimization for 68Ga reconstructed images and a brief description of the methods and results in terms of β-value optimization. Of note, all the authors that assessed image quality using the NEMA IQ phantom adopted CR and BV or similar indexes separately to assess image quality, as reported in the Supplementary MaterialTable S4 [
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
]. Due to the monotone increase of both indexes according to the β-values increase, the identification of the best parameter for image reconstruction relied on visual assessment or direct comparison with OSEM noise. In this context, the use of a figure of merit as the CRBV function, that incorporates both the CR and BV, as proposed in our study, would allow to select an optimal β-value (or an optimal range of β-values) through its maximization. The CRBV approach has been designed similarly to the P+ function [
] and can be applied to select the best trade-off between the signal recovery and variability. Refinements of the CRBV function will be proposed in further studies by testing additional combinations of CR and BV metrics for image quality optimization.
The results of the CRBVd,β image quality index highlighted that Q.Clear with 250 ≤ β ≤ 800 was always able to outperform OSEM, while, outside of this range of β-values, Q.Clear still outperformed OSEM in most of the combinations of sphere diameter and phantom setup. In particular, OSEM outperformed Q.Clear for very high β-values for the smaller spheres (e.g., β ≥ 1000 for the 13 mm diameter sphere in the 4 min/bp acquisition of the NEMAow phantom) and for very low β-values for the larger spheres (e.g., β ≤ 250 for the 28 mm diameter sphere in the 2 min/bp acquisition of the NEMAow phantom). In addition, β = 50 showed poor CRBV performance for most of the investigated setups. More generally, the width of the β-values range that allowed Q.Clear to outperform OSEM was larger for longer acquisition time (i.e., 4 min/bp) and for the NEMAstd phantom setup. For the smaller spheres (i.e., diameter < 20 mm), the CRBV improvement in OSEM image reconstruction was mainly due to improved CR with respect to high β-values Q.Clear reconstructions, while BV was comparable. For example, for the 13 mm diameter sphere in the 4 min/bp acquisition of the NEMAstd, CR were 39.6 % and 36.9 % and BV were 3.9 % and 2.9 % for OSEM and Q.Clear β = 1600, respectively, leading to CRBV of 38.1 % and 35.8 %. For the largest spheres, conversely, OSEM CRBV resulted higher due to a lower BV with respect to very low β-values Q.Clear images while obtaining similar CR. For example, for the 37 mm diameter sphere in the 4 min/bp acquisition of the NEMAstd, CR were 75.6 % and 81.6 % and BV were 4.1 % and 13.4 % for OSEM and Q.Clear β = 100, respectively, leading to CRBV of 72.5 % and 70.6 %. These trends in terms of CR and BV performances of OSEM reconstruction when compared to very high or very low β-values Q.Clear images are in agreement with the existing literature [
In our study, the use of the CRBV function, which incorporates both CR and BV indexes, allowed to identify optimal β-value specifically for lesion size and patient type (i.e., normal weight or overweight), thus providing an added value to the individual analysis of CR and BV currently performed by most PET/CT/MRI centres [
] following the NEMA methodology. The optimization strategy highlighted that the best compromise between CR and BV could be found at lower β-values ranges for smaller spheres and higher β-value ranges for larger spheres. For the 4 min/bp image reconstructions of NEMAstd phantom, nevertheless, β-values of 300 and 350 allowed to simultaneously optimize all spheres (except for the 10 mm diameter sphere), while for NEMAow phantom such balance was not feasible; for this phantom setup, β-values of 350–500 were needed for the 22–, 28- and 37 mm diameter spheres and β-values of 200–250 were selected for the remaining spheres. In addition, for the 17-, 22–, 28- and 37-mm diameter spheres of both NEMAstd and NEMAow phantom images reconstructed at 2 min/bp images a β-value of 500 was needed, while for the spheres with a low diameter (i.e., 10- and 13 mm) a β-value of 200 and between 250 ÷ 350 were selected in NEMAstd and NEMAow phantoms, respectively. Finally, the BPL algorithm overperformed the OSEM one in most of the investigated images and spheres, regardless of the min/bp acquisition times, in agreement with previously reported studies.
Although obtained with a novel approach, the results of our optimization strategy can be compared to what has been reported by visual studies. Seo et al. [
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
] indicated an optimal β-value of 500 based on overall image quality in a small cohort of patients. Interestingly, the mean BMI was relatively high (28.9 ± 3.6 kg/m2) and their results resulted consistent with the CRBV function on the NEMAow phantom. Lindstrom et al. [
] selected β = 400 and β = 900 as preferred images on a limited number of options in terms of lesion conspicuity, image sharpness and overall image quality. On the contrary, larger β-values were indicated by other studies. For example, Baratto et al. [
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
] investigated OSEM and Q.Clear based image reconstruction on PCa with 68Ga-RM2 and 68Ga-PSMA-11 and the visual score by a panel of nuclear medicine physicians judged a relatively high Q.Clear β parameter as optimal (i.e., β = 750), while low β-values (such as 250 or 350) were reported as insufficient in diagnosis. Nevertheless, the reported lesions were of various size and 22 out of 57 had a diameter of <1 cm, a size that was not investigated in our study. Also, Zanoni [
] indicated higher β-values (β = 1600 and β from 1100 to 1600, depending on acquisition time per bed, respectively). As they pointed out, a reason can be possibly found in the fact that physicians chose to focus essentially on noise in the images: this may induce a shift towards larger β-values. In addition, Chicheportiche used a spheres-to-background ratio of about 4:1 and a total activity much lower than ours, thus not allowing a direct comparison. Modifications to our CRBV function could possibly reconcile the results of our phantom study and visual studies and better model the visual score given by the physicians.
For what concerns phantom studies, our results confirmed the already published descending trends of both CR and BV at increasing β-values, as reported on NEMAstd in various works [
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
]. In particular, our study reports the first experience using a phantom with a nominal 8:1 concentration activity ratio on a fully digital Discovery MI PET/CT system, while phantom studies with nominal concentration activity ratios of 4:1 for the same scanner model have been already reported in literature [
]. In addition, when the same phantom configuration and hot spheres-to-background ratio were adopted, different injected activities were found with respect to our setup. In particular, Krokos et al. [
] reported an activity concentration of 19.9 kBq/ml and 2.4 kBq/ml for the hot spheres and background respectively, which is about one-sixth of the total injected activity used in our study. Our choice was justified by injecting the full activity used in clinical examination (i.e., about 140 MBq), on the assumption that most of the uptake occurs in the clinically interesting areas. The high injected activity, that results in higher signal and signal-to-noise ratio, may have affected our results, and shifted the optimal β-values towards lower values; a similar trend has been observed by Chicheportiche et al. [2021] by varying the acquisition time per bed. This observation is confirmed by the higher optimal β-values that can be found by lowering the acquisition time per bed position (i.e., 2 min/bp versus 4 min/bp), thus reducing the total amount of collected signal. Similarly, although the proposed phantom can be considered to simulate overweight patients, additional layers of scattering materials to represent higher BMI patients could make the comparison with visual studies on patients fairer. This will be furtherly investigated in additional studies, as well as the impact on spheres of diameter smaller than 10 mm [
Our study proposed a new figure of merit for the evaluation of PET image quality based on a combination of CR and BV measurements on phantom images. This strategy can be used to compare the performances of Q.Clear with respect to OSEM image reconstruction algorithm and to identify the optimal β-values ranges of the Q.Clear algorithm at varying phantom setup settings (i.e., representing normal weight and over-weight patients) and hot spheres, simulating lesions uptake, dimensions (i.e., from 10 mm to 37 mm diameters). Q.Clear showed an overall superiority with respect to the OSEM for a wide range of β-values in terms of the novel CRBV image quality index; nevertheless, differences in the performance of the Q.Clear image reconstruction algorithm in phantom simulating normal weight and overweight patients were found, as well as for different spheres size. This suggests the use of different β-values for image reconstruction when analysing small (i.e., <17 mm) and large lesions uptake.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Appendix A. Supplementary data
The following are the Supplementary data to this article:
Optimization of image reconstruction for (90)Y selective internal radiotherapy on a lutetium yttrium orthosilicate PET/CT System Using a Bayesian penalized likelihood reconstruction algorithm.
The impact of time-of-flight, resolution recovery, and noise modelling in reconstruction algorithms in non-solid-state detectors PET/CT scanners: - multi-centric comparison of activity recovery in a 68Ge phantom.
Comparison of regularized reconstruction and ordered subset expectation maximization reconstruction in the diagnostics of prostate cancer using digital time-of-flight (68)Ga-PSMA-11 PET/CT imaging.
Evaluation of quantitative Ga-68 PSMA PET/CT repeatability of recurrent prostate cancer lesions using both OSEM and Bayesian penalized likelihood reconstruction algorithms.
The effect of various β values on image quality and semiquantitative measurements in 68Ga-RM2 and 68Ga-PSMA-11 PET/MRI images reconstructed with a block sequential regularized expectation maximization algorithm.
Quantitative and qualitative improvement of low-count [(68)Ga]citrate and [(90)Y]microspheres PET image reconstructions using block sequential regularized expectation maximization algorithm.
Evaluation of block-sequential regularized expectation maximization reconstruction of (68)Ga-DOTATOC, (18)F-fluoride, and (11)C-acetate whole-body examinations acquired on a digital time-of-flight PET/CT scanner.
Quantification of FDG PET studies using standardised uptake values in multi-centre trials: effects of image reconstruction, resolution and ROI definition parameters.
Impact of target-to-background ratio, target size, emission scan duration, and activity on physical figures of merit for a 3D LSO-based whole body PET/CT scanner.
Quantitative 3D assessment of (68)Ga-DOTATOC PET/MRI with diffusion-weighted imaging to assess imaging markers for gastroenteropancreatic neuroendocrine tumors: preliminary results.
03319-1&id=gr1.jpg){kind=link}
03319-1&id=gr2.jpg){kind=link}
03319-1&id=gr3.jpg){kind=link}
03319-1&id=gr4.jpg){kind=link}
03319-1&id=gr5.jpg){kind=link}