A Poisson resampling method for simulating reduced counts in nuclear medicine images

Verification of image quality improvement of low-count bone scintigraphy using deep learning

To improve image quality for low-count bone scintigraphy using deep learning and evaluate their clinical applicability. Six hundred patients (training, 500; validation, 50; evaluation, 50) were included in this study. Low-count original images (75%, 50%, 25%, 10%, and 5% counts) were generated from reference images (100% counts) using Poisson resampling. Output (DL-filtered) images were obtained after training with U-Net using reference images as teacher data. Gaussian-filtered images were generated for comparison. Peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) to the reference image were calculated to determine image quality. Artificial neural network (ANN) value, bone scan index (BSI), and number of hotspots (Hs) were computed using BONENAVI analysis to assess diagnostic performance. Accuracy of bone metastasis detection and area under the curve (AUC) were calculated. PSNR and SSIM for DL-filtered images were highest in all count percentages. BONENAVI analysis values for DL-filtered images did not differ significantly, regardless of the presence or absence of bone metastases. BONENAVI analysis values for original and Gaussian-filtered images differed significantly at ≦25% counts in patients without bone metastases. In patients with bone metastases, BSI and Hs for original and Gaussian-filtered images differed significantly at ≦10% counts, whereas ANN values did not. The accuracy of bone metastasis detection was highest for DL-filtered images in all count percentages; the AUC did not differ significantly. The deep learning method improved image quality and bone metastasis detection accuracy for low-count bone scintigraphy, suggesting its clinical applicability.

Quantitative validation of Monte Carlo SPECT simulation: application to a Mediso AnyScan GATE simulation

BACKGROUND

Monte Carlo (MC) simulations are used in nuclear medicine imaging as they provide unparalleled insight into processes that are not directly experimentally measurable, such as scatter and attenuation in an acquisition. Whilst MC is often used to provide a 'ground-truth', this is only the case if the simulation is fully validated against experimental data. This work presents a quantitative validation for a MC simulation of a single-photon emission computed tomography (SPECT) system.

METHODS

An MC simulation model of the Mediso AnyScan SCP SPECT system installed at the UK National Physical Laboratory was developed in the GATE (Geant4 Application for Tomographic Emission) toolkit. Components of the detector head and two collimator configurations were modelled according to technical specifications and physical measurements. Experimental detection efficiency measurements were collected for a range of energies, permitting an energy-dependent intrinsic camera efficiency correction function to be determined and applied to the simulation on an event-by-event basis. Experimental data were collected in a range of geometries with [Formula: see text]Tc for comparison to simulation. The procedure was then repeated with [Formula: see text]Lu to determine how the validation extended to another isotope and set of collimators.

RESULTS

The simulation's spatial resolution, sensitivity, energy spectra and the projection images were compared with experimental measurements. The simulation and experimental uncertainties were determined and propagated to all calculations, permitting the quantitative agreement between simulated and experimental SPECT acquisitions to be determined. Statistical agreement was seen in sinograms and projection images of both [Formula: see text]Tc and [Formula: see text]Lu data. Average simulated and experimental sensitivity ratios of ([Formula: see text]) were seen for emission and scatter windows of [Formula: see text]Tc, and ([Formula: see text]) and ([Formula: see text]) for the 113 and 208 keV emissions of [Formula: see text]Lu, respectively.

CONCLUSIONS

MC simulations will always be an approximation of a physical system and the level of agreement should be assessed. A validation method is presented to quantify the level of agreement between a simulation model and a physical SPECT system.

SPECT reconstruction with a trained regularizer using CT-side information: Application to 177Lu SPECT imaging

Improving low-count SPECT can shorten scans and support pre-therapy theranostic imaging for dosimetry-based treatment planning, especially with radionuclides like 177Lu known for low photon yields. Conventional methods often underperform in low-count settings, highlighting the need for trained regularization in model-based image reconstruction. This paper introduces a trained regularizer for SPECT reconstruction that leverages segmentation based on CT imaging. The regularizer incorporates CT-side information via a segmentation mask from a pre-trained network (nnUNet). In this proof-of-concept study, we used patient studies with 177Lu DOTATATE to train and tested with phantom and patient datasets, simulating pre-therapy imaging conditions. Our results show that the proposed method outperforms both standard unregularized EM algorithms and conventional regularization with CT-side information. Specifically, our method achieved marked improvements in activity quantification, noise reduction, and root mean square error. The enhanced low-count SPECT approach has promising implications for theranostic imaging, post-therapy imaging, whole body SPECT, and reducing SPECT acquisition times.

Anthropomorphic cardiac phantom for dynamic SPECT

BACKGROUND

As myocardial blood flow measurement (MBF) in SPECT systems became recently available, significant effort has been devoted to its validation. For that purpose, we have developed a cardiac phantom that is able to mimic physiological radiotracer variation in the left ventricle cavity and in the myocardium, while performing beating-like motion. The new phantom is integrated inside a standard anthropomorphic torso allowing a realistic tissue attenuation and gamma-ray scattering METHODS AND RESULTS: A mechanical cardiac phantom was integrated in a commercially available anthropomorphic torso. Using a GE Discovery 530c SPECT, measurements were performed. It was found that gamma-ray attenuation effects are significant and limit the MBF measurements to global/three-vessel resolution. Dynamic SPECT experiments were performed to validate MBF accuracy and showed mean relative error of 14%. Finally, the effect of varying radiotracer dose on the accuracy of dynamic SPECT was studied CONCLUSIONS: A dynamic cardiac phantom has been developed and successfully integrated in a standard SPECT torso. A good agreement was found between SPECT-reported MBF values and the expected results. Despite increased noise-to-signal ratio when radiotracer doses were reduced, MBF uncertainty did not increase significantly down to very low doses, thanks to the temporal integration of the activity during the measurement.

Quantification and dosimetry of small volumes including associated uncertainty estimation

BACKGROUND

Accurate quantification of radioactivity in a source of interest relies on accurate registration between SPECT and anatomical images, and appropriate correction of partial volume effects (PVEs). For small volumes, exact registration between the two imaging modalities and recovery factors used to correct for PVE are unreliable. There is currently no guidance relating to quantification or the associated uncertainty estimation for small volumes.

MATERIAL AND METHODS

A method for quantification of small sources of interest is proposed, which uses multiple oversized volumes of interest. The method was applied to three Na[131I]I activity distributions where a Na[131I]I capsule was situated within a cylindrical phantom containing either zero background, uniform background or non-uniform background and to a scenario with small lesions placed in an anthropomorphic phantom. The Na[131I]I capsule and lesions were quantified using the proposed method and compared with measurements made using two alternative quantification methods. The proposed method was also applied to assess the absorbed dose delivered to a bone metastasis following [131I]mIBG therapy for neuroblastoma including the associated uncertainty estimation.

RESULTS

The method is accurate across a range of activities and in varied radioactivity distributions. Median percentage errors using the proposed method in no background, uniform backgrounds and non-uniform backgrounds were - 0.4%, - 0.3% and 1.7% with median associated uncertainties of 1.4%, 1.4% and 1.6%, respectively. The technique is more accurate and robust when compared to currently available alternative methods.

CONCLUSIONS

The proposed method provides a reliable and accurate method for quantification of sources of interest, which are less than three times the spatial resolution of the imaging system. The method may be of use in absorbed dose calculation in cases of bone metastasis, lung metastasis or thyroid remnants.

A practical solution to estimate the sample size required for clinical prediction models generated from observational research on data

Background

Estimating the required sample size is crucial when developing and validating clinical prediction models. However, there is no consensus about how to determine the sample size in such a setting. Here, the goal was to compare available methods to define a practical solution to sample size estimation for clinical predictive models, as applied to Horizon 2020 PRIMAGE as a case study.

Methods

Three different methods (Riley’s; “rule of thumb” with 10 and 5 events per predictor) were employed to calculate the sample size required to develop predictive models to analyse the variation in sample size as a function of different parameters. Subsequently, the sample size for model validation was also estimated.

Results

To develop reliable predictive models, 1397 neuroblastoma patients are required, 1060 high-risk neuroblastoma patients and 1345 diffuse intrinsic pontine glioma (DIPG) patients. This sample size can be lowered by reducing the number of variables included in the model, by including direct measures of the outcome to be predicted and/or by increasing the follow-up period. For model validation, the estimated sample size resulted to be 326 patients for neuroblastoma, 246 for high-risk neuroblastoma, and 592 for DIPG.

Conclusions

Given the variability of the different sample sizes obtained, we recommend using methods based on epidemiological data and the nature of the results, as the results are tailored to the specific clinical problem. In addition, sample size can be reduced by lowering the number of parameter predictors, by including direct measures of the outcome of interest.

Simulating dose reduction for myocardial perfusion SPECT using a Poisson resampling method

Purpose

The purpose of this study was to determine the lowest Tl-201 dose that does not reduce the image quality of myocardial perfusion SPECT (MPS) by Poisson resampling simulation.

Methods

One hundred and twelve consecutive MPS data from patients with suspected or known coronary artery disease were collected retrospectively. Stress and rest MPS data were resampled using the Poisson method with 33%, 50%, 67%, and 100% count settings. Two nuclear medicine physicians assessed the image quality of reconstructed data visually by giving grades from - 2 to + 2. The summed stress score (SSS), summed rest score (SRS), and summed difference score (SDS) were obtained on the workstation. Image quality grades and semi-quantitative scores were then compared among these resampled images.

Results

The proportions of "adequate" image quality were 0.48, 0.75, 0.92, and 0.96 for the groups of images with 33%, 50%, 67%, and 100% data, respectively. The quality of the resampled images was significantly degraded at 50% and 33% count settings, while the image quality was not different between 67 and 100% count settings. We also found that high body mass index further decreased image quality at 33% count setting. Among the semi-quantitative parameters, SSS and SRS showed a tendency to increase with a decline in count.

Conclusion

Based on the simulation results, Tl-201 dose for MPS can be reduced to 74 MBq without significant loss of image quality. However, the SSS and SRS can be changed significantly, and it needs to be further verified under the different conditions.

Noise heterogeneity in attenuation-corrected cardiac SPECT images increases perfusion value uncertainty near the base of the heart

BACKGROUND

Dedicated cardiac SPECT cameras which employ multi-pinhole detectors have variable photon sensitivity within the camera's field-of-view such that a lower number of photon counts is typically detected from the base of the heart than from the apex. Consequently, the noise in a reconstructed image is expected to be higher at the base than at the apex of the heart.

METHODS

Patient emission images were resampled to create statistical replicates which were reconstructed with and without attenuation correction. Noise images were computed using one standard deviation of the replicated images. These were evaluated for 93 patients with normal study results, each imaged with both a dual-headed parallel-hole camera and a multi-pinhole camera. Statistics for a normal database (NDB) of images from the 93 patients were also calculated.

RESULTS

Image noise (1.7-fold) and NDB uncertainty (1.3-fold) increase significantly from the apex-to-the base of the heart in attenuation-corrected multi-pinhole SPECT images. The differences for non-attenuation-corrected images or those acquired with a parallel-hole camera were not significant.

CONCLUSIONS

For best interpretation of attenuation-corrected images acquired with multi-pinhole cameras, knowledge of NDB uncertainty gradients should be taken into consideration.

Clinical validation of time reduction strategy in continuous step-and-shoot mode during SPECT acquisition

Background

The SwiftScan solution (General Electric Healthcare) combines a new low-energy high-resolution sensitivity collimator and a tomographic step-and-shoot continuous (SSC) mode acquisition. The purpose of this study is to determine whether SSC mode can be used in clinical practice with shorter examination times, while preserving image quality and ensuring accurate semi-quantification. Twenty bone scan and 10 lung scan studies were randomly selected over a period of 2 months. Three sets of image datasets were produced: step-and-shoot (SS) acquisition, simulated 25% count reduction using the Poisson resampling method (SimSS), and SimSS continuous acquisition (SimSSC), where SimSS was summed with counts acquired during detector head rotation. Visual assessment (5-point Likert scale, 2 readers) and semi-quantitative evaluation (50 focal uptake from 10 bone studies), assessed by SUV_mean, coefficient of variation (COV), and contrast-to-noise ratio (CNR), were performed using t test and Bland-Altman analysis.

Results

Intra-reader agreement was substantial for reader 1 (k = 0.71) and for reader 2 (k = 0.61). Inter-reader agreement was substantial for SS set (k = 0.93) and moderate for SimSSC (k = 0.52). Bland-Altman analysis showed a good interchangeability of SS and SimSSC SUV values. The mean CNR between SS and SimSSC was not significantly different: 42.9 ± 43.7 [23.7–62.1] vs. 43.1 ± 46 [22.9–63.3] (p = 0.46), respectively. COV values, assessing noise level, did not deviate significantly between SS and SimSSC: 0.20 ± 0.08 [0.18–0.23] vs. 0.21 ± 0.08, [0.18–0.23] (p = 0.15), respectively, whereas a significant difference was demonstrated between SS and SimSS: 0.20 ± 0.08 [0.18–0.23] vs. 0.23 ± 0.09 [0.20–0.25] (p < 0.0001), respectively.

Conclusions

SSC mode acquisition decreases examination time by approximately 25% in bone and lung SPECT/CT studies compared to SS mode (~ 2 min per single-bed SPECT), without compromising image quality and signal quantification. This SPECT sensitivity improvement also offers the prospect of more comfortable exams, with less motion artifacts, especially in painful or dyspneic patients.

Patient-specific SPECT imaging protocols to standardize image noise

BACKGROUND

In addition to acquired photon counts, image noise depends on the image reconstruction algorithm. This work develops patient-specific activity or acquisition time protocols to standardize the average noise in a reconstructed image for different patients, cameras, and reconstruction algorithms.

METHODS

Image noise was calculated for images from 43 patients acquired on both a conventional and a multiple-pinhole cardiac SPECT camera. Functions were found to relate image noise to radiotracer activity, scan time, and body mass and were validated by normalizing the image noise in a test set of 58 patients.

RESULTS

There was a 3.6-fold difference in photon sensitivity between the two cameras but a 16-fold difference in activity-scan time was necessary to match the noise levels. Image noise doubled from 45 to 128 kg for the conventional camera (12.8 minutes) and tripled for the multiple-pinhole camera (5 minutes) for 350 MBq (9.5 mCi) 99mTc-tetrofosmin. It was 16.3% and 6.1% respectively for an average sized patient.

CONCLUSIONS

A linear scaling of activity with respect to the patient weight normalizes image noise but the scaling factors depend on the choice of camera and image reconstruction parameters. Therefore, equivalent numbers of acquired photon counts are not sufficient to guarantee equivalent image noise.

Comment on: 'A Poisson resampling method for simulating reduced counts in nuclear medicine images'

In order to be able to calculate half-count images from already acquired data, White and Lawson published their method based on Poisson resampling. They verified their method experimentally by measurements with a Co-57 flood source. In this comment their results are reproduced and confirmed by a direct numerical simulation in Matlab. Not only Poisson resampling, but also two direct redrawing methods were investigated. Redrawing methods were based on a Poisson and a Gaussian distribution. Mean, standard deviation, skewness and excess kurtosis half-count/full-count ratios were determined for all methods, and compared to the theoretical values for a Poisson distribution. Statistical parameters showed the same behavior as in the original note and showed the superiority of the Poisson resampling method. Rounding off before saving of the half count image had a severe impact on counting statistics for counts below 100. Only Poisson resampling was not affected by this, while Gaussian redrawing was less affected by it than Poisson redrawing. Poisson resampling is the method of choice, when simulating half-count (or less) images from full-count images. It simulates correctly the statistical properties, also in the case of rounding off of the images.