PET Image Denoising Using a Synergistic Multiresolution Analysis of Structural (MRI/CT) and Functional Datasets

Impact of deep learning-based multiorgan segmentation methods on patient-specific internal dosimetry in PET/CT imaging: A comparative study

PURPOSE

Accurate and fast multiorgan segmentation is essential in image-based internal dosimetry in nuclear medicine. While conventional manual PET image segmentation is widely used, it suffers from both being time-consuming as well as subject to human error. This study exploited 2D and 3D deep learning (DL) models. Key organs in the trunk of the body were segmented and then used as a reference for networks.

METHODS

The pre-trained p2p-U-Net-GAN and HighRes3D architectures were fine-tuned with PET-only images as inputs. Additionally, the HighRes3D model was alternatively trained with PET/CT images. Evaluation metrics such as sensitivity (SEN), specificity (SPC), intersection over union (IoU), and Dice scores were considered to assess the performance of the networks. The impact of DL-assisted PET image segmentation methods was further assessed using the Monte Carlo (MC)-derived S-values to be used for internal dosimetry.

RESULTS

A fair comparison with manual low-dose CT-aided segmentation of the PET images was also conducted. Although both 2D and 3D models performed well, the HighRes3D offers superior performance with Dice scores higher than 0.90. Key evaluation metrics such as SEN, SPC, and IoU vary between 0.89-0.93, 0.98-0.99, and 0.87-0.89 intervals, respectively, indicating the encouraging performance of the models. The percentage differences between the manual and DL segmentation methods in the calculated S-values varied between 0.1% and 6% with a maximum attributed to the stomach.

CONCLUSION

The findings prove while the incorporation of anatomical information provided by the CT data offers superior performance in terms of Dice score, the performance of HighRes3D remains comparable without the extra CT channel. It is concluded that both proposed DL-based methods provide automated and fast segmentation of whole-body PET/CT images with promising evaluation metrics. Between them, the HighRes3D is more pronounced by providing better performance and can therefore be the method of choice for 18F-FDG-PET image segmentation.

Virtual high-count PET image generation using a deep learning method

PURPOSE

Recently, deep learning-based methods have been established to denoise the low-count positron emission tomography (PET) images and predict their standard-count image counterparts, which could achieve reduction of injected dosage and scan time, and improve image quality for equivalent lesion detectability and clinical diagnosis. In clinical settings, the majority scans are still acquired using standard injection dose with standard scan time. In this work, we applied a 3D U-Net network to reduce the noise of standard-count PET images to obtain the virtual-high-count (VHC) PET images for identifying the potential benefits of the obtained VHC PET images.

METHODS

The training datasets, including down-sampled standard-count PET images as the network input and high-count images as the desired network output, were derived from 27 whole-body PET datasets, which were acquired using 90-min dynamic scan. The down-sampled standard-count PET images were rebinned with matched noise level of 195 clinical static PET datasets, by matching the normalized standard derivation (NSTD) inside 3D liver region of interests (ROIs). Cross-validation was performed on 27 PET datasets. Normalized mean square error (NMSE), peak signal to noise ratio (PSNR), structural similarity index (SSIM), and standard uptake value (SUV) bias of lesions were used for evaluation on standard-count and VHC PET images, with real-high-count PET image of 90 min as the gold standard. In addition, the network trained with 27 dynamic PET datasets was applied to 195 clinical static datasets to obtain VHC PET images. The NSTD and mean/max SUV of hypermetabolic lesions in standard-count and VHC PET images were evaluated. Three experienced nuclear medicine physicians evaluated the overall image quality of randomly selected 50 out of 195 patients' standard-count and VHC images and conducted 5-score ranking. A Wilcoxon signed-rank test was used to compare differences in the grading of standard-count and VHC images.

RESULTS

The cross-validation results showed that VHC PET images had improved quantitative metrics scores than the standard-count PET images. The mean/max SUVs of 35 lesions in the standard-count and true-high-count PET images did not show significantly statistical difference. Similarly, the mean/max SUVs of VHC and true-high-count PET images did not show significantly statistical difference. For the 195 clinical data, the VHC PET images had a significantly lower NSTD than the standard-count images. The mean/max SUVs of 215 hypermetabolic lesions in the VHC and standard-count images showed no statistically significant difference. In the image quality evaluation by three experienced nuclear medicine physicians, standard-count images and VHC images received scores with mean and standard deviation of 3.34±0.80 and 4.26 ± 0.72 from Physician 1, 3.02 ± 0.87 and 3.96 ± 0.73 from Physician 2, and 3.74 ± 1.10 and 4.58 ± 0.57 from Physician 3, respectively. The VHC images were consistently ranked higher than the standard-count images. The Wilcoxon signed-rank test also indicated that the image quality evaluation between standard-count and VHC images had significant difference.

CONCLUSIONS

A DL method was proposed to convert the standard-count images to the VHC images. The VHC images had reduced noise level. No significant difference in mean/max SUV to the standard-count images was observed. VHC images improved image quality for better lesion detectability and clinical diagnosis.

Unsupervised PET logan parametric image estimation using conditional deep image prior

Recently, deep learning-based denoising methods have been gradually used for PET images denoising and have shown great achievements. Among these methods, one interesting framework is conditional deep image prior (CDIP) which is an unsupervised method that does not need prior training or a large number of training pairs. In this work, we combined CDIP with Logan parametric image estimation to generate high-quality parametric images. In our method, the kinetic model is the Logan reference tissue model that can avoid arterial sampling. The neural network was utilized to represent the images of Logan slope and intercept. The patient's computed tomography (CT) image or magnetic resonance (MR) image was used as the network input to provide anatomical information. The optimization function was constructed and solved by the alternating direction method of multipliers (ADMM) algorithm. Both simulation and clinical patient datasets demonstrated that the proposed method could generate parametric images with more detailed structures. Quantification results showed that the proposed method results had higher contrast-to-noise (CNR) improvement ratios (PET/CT datasets: 62.25%±29.93%; striatum of brain PET datasets : 129.51%±32.13%, thalamus of brain PET datasets: 128.24%±31.18%) than Gaussian filtered results (PET/CT datasets: 23.33%±18.63%; striatum of brain PET datasets: 74.71%±8.71%, thalamus of brain PET datasets: 73.02%±9.34%) and nonlocal mean (NLM) denoised results (PET/CT datasets: 37.55%±26.56%; striatum of brain PET datasets: 100.89%±16.13%, thalamus of brain PET datasets: 103.59%±16.37%).

Clinical and phantom validation of a deep learning based denoising algorithm for F-18-FDG PET images from lower detection counting in comparison with the standard acquisition

Background

PET/CT image quality is directly influenced by the F-18-FDG injected activity. The higher the injected activity, the less noise in the reconstructed images but the more radioactive staff exposition. A new FDA cleared software has been introduced to obtain clinical PET images, acquired at 25% of the count statistics considering US practices. Our aim is to determine the limits of a deep learning based denoising algorithm (SubtlePET) applied to statistically reduced PET raw data from 3 different last generation PET scanners in comparison to the regular acquisition in phantom and patients, considering the European guidelines for radiotracer injection activities. Images of low and high contrasted (SBR = 2 and 5) spheres of the IEC phantom and high contrast (SBR = 5) of micro-spheres of Jaszczak phantom were acquired on 3 different PET devices. 110 patients with different pathologies were included. The data was acquired in list-mode and retrospectively reconstructed with the regular acquisition count statistic (PET100), 50% reduction in counts (PET50) and 66% reduction in counts (PET33). These count reduced images were post-processed with SubtlePET to obtain PET50 + SP and PET33 + SP images. Patient image quality was scored by 2 senior nuclear physicians. Peak-signal-to-Noise and Structural similarity metrics were computed to compare the low count images to regular acquisition (PET100).

Results

SubtlePET reliably denoised the images and maintained the SUV_max values in PET50 + SP. SubtlePET enhanced images (PET33 + SP) had slightly increased noise compared to PET100 and could lead to a potential loss of information in terms of lesion detectability. Regarding the patient datasets, the PET100 and PET50 + SP were qualitatively comparable. The SubtlePET algorithm was able to correctly recover the SUV_max values of the lesions and maintain a noise level equivalent to full-time images.

Conclusion

Based on our results, SubtlePET is adapted in clinical practice for half-time or half-dose acquisitions based on European recommended injected dose of 3 MBq/kg without diagnostic confidence loss.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40658-022-00465-z.

Content-Noise Complementary Learning for Medical Image Denoising

Medical imaging denoising faces great challenges, yet is in great demand. With its distinctive characteristics, medical imaging denoising in the image domain requires innovative deep learning strategies. In this study, we propose a simple yet effective strategy, the content-noise complementary learning (CNCL) strategy, in which two deep learning predictors are used to learn the respective content and noise of the image dataset complementarily. A medical image denoising pipeline based on the CNCL strategy is presented, and is implemented as a generative adversarial network, where various representative networks (including U-Net, DnCNN, and SRDenseNet) are investigated as the predictors. The performance of these implemented models has been validated on medical imaging datasets including CT, MR, and PET. The results show that this strategy outperforms state-of-the-art denoising algorithms in terms of visual quality and quantitative metrics, and the strategy demonstrates a robust generalization capability. These findings validate that this simple yet effective strategy demonstrates promising potential for medical image denoising tasks, which could exert a clinical impact in the future. Code is available at: https://github.com/gengmufeng/CNCL-denoising.

Adaptive rectification based adversarial network with spectrum constraint for high-quality PET image synthesis

Positron emission tomography (PET) is a typical nuclear imaging technique, which can provide crucial functional information for early brain disease diagnosis. Generally, clinically acceptable PET images are obtained by injecting a standard-dose radioactive tracer into human body, while on the other hand the cumulative radiation exposure inevitably raises concerns about potential health risks. However, reducing the tracer dose will increase the noise and artifacts of the reconstructed PET image. For the purpose of acquiring high-quality PET images while reducing radiation exposure, in this paper, we innovatively present an adaptive rectification based generative adversarial network with spectrum constraint, named AR-GAN, which uses low-dose PET (LPET) image to synthesize standard-dose PET (SPET) image of high-quality. Specifically, considering the existing differences between the synthesized SPET image by traditional GAN and the real SPET image, an adaptive rectification network (AR-Net) is devised to estimate the residual between the preliminarily predicted image and the real SPET image, based on the hypothesis that a more realistic rectified image can be obtained by incorporating both the residual and the preliminarily predicted PET image. Moreover, to address the issue of high-frequency distortions in the output image, we employ a spectral regularization term in the training optimization objective to constrain the consistency of the synthesized image and the real image in the frequency domain, which further preserves the high-frequency detailed information and improves synthesis performance. Validations on both the phantom dataset and the clinical dataset show that the proposed AR-GAN can estimate SPET images from LPET images effectively and outperform other state-of-the-art image synthesis approaches.

Two novel PET image restoration methods guided by PET-MR kernels: Application to brain imaging

PURPOSE

Post-reconstruction positron emission tomography (PET) image restoration methods that take advantage of available anatomical information can play an important role in accurate quantification of PET images. However, when using anatomical information, the resulting PET image may lose resolution in certain regions where the anatomy does not agree with the change in functional activity. In this work, this problem is addressed by using both magnetic resonance (MR) and filtered PET images to guide the denoising process.

METHODS

In this work, two novel post-reconstruction methods for restoring PET images using the subject's registered T1-weighted MR image are proposed. The first method is based on a representation of the image using basis functions extracted from T1-weighted MR and filtered PET image. The coefficients for these basis functions are estimated using a sparsity-penalized least squares objective function. The second method is a noniterative fast method that uses guided kernel filtering in combination with twicing to restore the noisy PET image. When applied after conventional PVE correction, these methods can be considered as voxel-based MR-guided partial volume effect (PVE) correction methods.

RESULTS

Using simulation analyses of [ 18 F]FDG PET images of the brain with lesions, the proposed methods are compared to other denoising methods through different figures of merit. The results show promising improvements in image quality as well as reduction in bias and variance of the lesions. We also show the application of the proposed methods on real [ 18 F]FDG data.

CONCLUSION

Two methods for restoring PET images were proposed. The methods were evaluated on simulation and real brain images. Most MR-guided PVE correction methods are only based on segmented T1-weighted images and their accuracy is very sensitive to segmentation errors, especially in regions of abnormalities and lesions. However, both proposed methods can use the T1-weighted image without segmentation. The simplicity and the very low computational cost of the second method make it suitable for clinical applications and large data studies. The proposed methods can be naturally extended to PVE correction and denoising of other functional modalities using corresponding anatomical information.

Joint solution for PET image segmentation, denoising, and partial volume correction

Segmentation, denoising, and partial volume correction (PVC) are three major processes in the quantification of uptake regions in post-reconstruction PET images. These problems are conventionally addressed by independent steps. In this study, we hypothesize that these three processes are dependent; therefore, jointly solving them can provide optimal support for quantification of the PET images. To achieve this, we utilize interactions among these processes when designing solutions for each challenge. We also demonstrate that segmentation can help in denoising and PVC by locally constraining the smoothness and correction criteria. For denoising, we adapt generalized Anscombe transformation to Gaussianize the multiplicative noise followed by a new adaptive smoothing algorithm called regional mean denoising. For PVC, we propose a volume consistency-based iterative voxel-based correction algorithm in which denoised and delineated PET images guide the correction process during each iteration precisely. For PET image segmentation, we use affinity propagation (AP)-based iterative clustering method that helps the integration of PVC and denoising algorithms into the delineation process. Qualitative and quantitative results, obtained from phantoms, clinical, and pre-clinical data, show that the proposed framework provides an improved and joint solution for segmentation, denoising, and partial volume correction.

Characterization and simulation of noise in PET images reconstructed with OSEM: Development of a method for the generation of synthetic images

INTRODUCTION

The goals of the study are to characterize imaging properties in 2D PET images reconstructed with the iterative algorithm ordered-subset expectation maximization (OSEM) and to propose a new method for the generation of synthetic images.

MATERIAL AND METHODS

The noise is analyzed in terms of its magnitude, spatial correlation, and spectral distribution through standard deviation, autocorrelation function, and noise power spectrum (NPS), respectively. Their variations with position and activity level are also analyzed. This noise analysis is based on phantom images acquired from ¹⁸F uniform distributions. Experimental recovery coefficients of hot spheres in different backgrounds are employed to study the spatial resolution of the system through point spread function (PSF). The NPS and PSF functions provide the baseline for the proposed simulation method: convolution with PSF as kernel and noise addition from NPS.

RESULTS

The noise spectral analysis shows that the main contribution is of random nature. It is also proven that attenuation correction does not alter noise texture but it modifies its magnitude. Finally, synthetic images of 2 phantoms, one of them an anatomical brain, are quantitatively compared with experimental images showing a good agreement in terms of pixel values and pixel correlations. Thus, the contrast to noise ratio for the biggest sphere in the NEMA IEC phantom is 10.7 for the synthetic image and 8.8 for the experimental image.

CONCLUSIONS

The properties of the analyzed OSEM-PET images can be described by NPS and PSF functions. Synthetic images, even anatomical ones, are successfully generated by the proposed method based on the NPS and PSF.

Deep Auto-context Convolutional Neural Networks for Standard-Dose PET Image Estimation from Low-Dose PET/MRI

Positron emission tomography (PET) is an essential technique in many clinical applications such as tumor detection and brain disorder diagnosis. In order to obtain high-quality PET images, a standard-dose radioactive tracer is needed, which inevitably causes the risk of radiation exposure damage. For reducing the patient's exposure to radiation and maintaining the high quality of PET images, in this paper, we propose a deep learning architecture to estimate the high-quality standard-dose PET (SPET) image from the combination of the low-quality low-dose PET (LPET) image and the accompanying T1-weighted acquisition from magnetic resonance imaging (MRI). Specifically, we adapt the convolutional neural network (CNN) to account for the two channel inputs of LPET and T1, and directly learn the end-to-end mapping between the inputs and the SPET output. Then, we integrate multiple CNN modules following the auto-context strategy, such that the tentatively estimated SPET of an early CNN can be iteratively refined by subsequent CNNs. Validations on real human brain PET/MRI data show that our proposed method can provide competitive estimation quality of the PET images, compared to the state-of-the-art methods. Meanwhile, our method is highly efficient to test on a new subject, e.g., spending ~2 seconds for estimating an entire SPET image in contrast to ~16 minutes by the state-of-the-art method. The results above demonstrate the potential of our method in real clinical applications.

PET image reconstruction using multi-parametric anato-functional priors

In this study, we investigate the application of multi-parametric anato-functional (MR-PET) priors for the maximum a posteriori (MAP) reconstruction of brain PET data in order to address the limitations of the conventional anatomical priors in the presence of PET-MR mismatches. In addition to partial volume correction benefits, the suitability of these priors for reconstruction of low-count PET data is also introduced and demonstrated, comparing to standard maximum-likelihood (ML) reconstruction of high-count data. The conventional local Tikhonov and total variation (TV) priors and current state-of-the-art anatomical priors including the Kaipio, non-local Tikhonov prior with Bowsher and Gaussian similarity kernels are investigated and presented in a unified framework. The Gaussian kernels are calculated using both voxel- and patch-based feature vectors. To cope with PET and MR mismatches, the Bowsher and Gaussian priors are extended to multi-parametric priors. In addition, we propose a modified joint Burg entropy prior that by definition exploits all parametric information in the MAP reconstruction of PET data. The performance of the priors was extensively evaluated using 3D simulations and two clinical brain datasets of [18F]florbetaben and [18F]FDG radiotracers. For simulations, several anato-functional mismatches were intentionally introduced between the PET and MR images, and furthermore, for the FDG clinical dataset, two PET-unique active tumours were embedded in the PET data. Our simulation results showed that the joint Burg entropy prior far outperformed the conventional anatomical priors in terms of preserving PET unique lesions, while still reconstructing functional boundaries with corresponding MR boundaries. In addition, the multi-parametric extension of the Gaussian and Bowsher priors led to enhanced preservation of edge and PET unique features and also an improved bias-variance performance. In agreement with the simulation results, the clinical results also showed that the Gaussian prior with voxel-based feature vectors, the Bowsher and the joint Burg entropy priors were the best performing priors. However, for the FDG dataset with simulated tumours, the TV and proposed priors were capable of preserving the PET-unique tumours. Finally, an important outcome was the demonstration that the MAP reconstruction of a low-count FDG PET dataset using the proposed joint entropy prior can lead to comparable image quality to a conventional ML reconstruction with up to 5 times more counts. In conclusion, multi-parametric anato-functional priors provide a solution to address the pitfalls of the conventional priors and are therefore likely to increase the diagnostic confidence in MR-guided PET image reconstructions.

Multi-Level Canonical Correlation Analysis for Standard-Dose PET Image Estimation

Positron emission tomography (PET) images are widely used in many clinical applications, such as tumor detection and brain disorder diagnosis. To obtain PET images of diagnostic quality, a sufficient amount of radioactive tracer has to be injected into a living body, which will inevitably increase the risk of radiation exposure. On the other hand, if the tracer dose is considerably reduced, the quality of the resulting images would be significantly degraded. It is of great interest to estimate a standard-dose PET (S-PET) image from a low-dose one in order to reduce the risk of radiation exposure and preserve image quality. This may be achieved through mapping both S-PET and low-dose PET data into a common space and then performing patch-based sparse representation. However, a one-size-fits-all common space built from all training patches is unlikely to be optimal for each target S-PET patch, which limits the estimation accuracy. In this paper, we propose a data-driven multi-level canonical correlation analysis scheme to solve this problem. In particular, a subset of training data that is most useful in estimating a target S-PET patch is identified in each level, and then used in the next level to update common space and improve estimation. In addition, we also use multi-modal magnetic resonance images to help improve the estimation with complementary information. Validations on phantom and real human brain data sets show that our method effectively estimates S-PET images and well preserves critical clinical quantification measures, such as standard uptake value.

Choosing Wavelet Methods, Filters, and Lengths for Functional Brain Network Construction

Wavelet methods are widely used to decompose fMRI, EEG, or MEG signals into time series representing neurophysiological activity in fixed frequency bands. Using these time series, one can estimate frequency-band specific functional connectivity between sensors or regions of interest, and thereby construct functional brain networks that can be examined from a graph theoretic perspective. Despite their common use, however, practical guidelines for the choice of wavelet method, filter, and length have remained largely undelineated. Here, we explicitly explore the effects of wavelet method (MODWT vs. DWT), wavelet filter (Daubechies Extremal Phase, Daubechies Least Asymmetric, and Coiflet families), and wavelet length (2 to 24)—each essential parameters in wavelet-based methods—on the estimated values of graph metrics and in their sensitivity to alterations in psychiatric disease. We observe that the MODWT method produces less variable estimates than the DWT method. We also observe that the length of the wavelet filter chosen has a greater impact on the estimated values of graph metrics than the type of wavelet chosen. Furthermore, wavelet length impacts the sensitivity of the method to detect differences between health and disease and tunes classification accuracy. Collectively, our results suggest that the choice of wavelet method and length significantly alters the reliability and sensitivity of these methods in estimating values of metrics drawn from graph theory. They furthermore demonstrate the importance of reporting the choices utilized in neuroimaging studies and support the utility of exploring wavelet parameters to maximize classification accuracy in the development of biomarkers of psychiatric disease and neurological disorders.

Improving (18)F-Fluoro-D-Glucose-Positron Emission Tomography/Computed Tomography Imaging in Alzheimer's Disease Studies

The goal was to improve Alzheimer's 2-deoxy-2-¹⁸F-fluoro-D-glucose (¹⁸F FDG)-positron emission tomography (PET)/computed tomography (CT) imaging through application of a novel, hybrid Fourier-wavelet windowed Fourier transform (WFT) restoration technique, in order to provide earlier and more accurate clinical results. General Electric Medical Systems downward-looking sonar PET/CT 16 slice system was used to acquire studies. Patient data were acquired according the Alzheimer's disease Neuroimaging Initiative (ADNI) protocol. Here, we implemented Fourier-wavelet regularized restoration, with a Butterworth low-pass filter, order n = 6 and a cut-off frequency f = 0.35 cycles/pixel and wavelet (Daubechies, order 2) noise suppression. The original (PET-O) and restored (PET-R) ADNI subject PET images were compared using the Alzheimer's discrimination analysis by dedicated software. Forty-two PET/CT scans were used in the study. They were performed on eleven ADNI subjects at intervals of approximately 6 months. The final clinical diagnosis was used as a gold standard. For three subjects, the final clinical diagnosis was mild cognitive impairment and those 13 PET/CT studies were not included in the final comparison, as the result was considered as inconclusive. Using the reminding 29 PET/CT studies (23 AD and 6 normal), the sensitivity and specificity of the PET-O and PET-R were calculated. The sensitivity was 0.65 and 0.96 for PET-O and PET-R, respectively, and the specificity was 0.67 and 0.50 for PET-O and PET-R. The accuracy was 0.66 and 0.86 for PET-O and PET-R, respectively. The results of the study demonstrated that the accuracy of three-dimensional brain F-18 FDG PET images was significantly improved by Fourier-wavelet restoration filtering.

Multimodal Partial-Volume Correction: Application to 18F-Fluoride PET/CT Bone Metastases Studies

(18)F-fluoride PET/CT offers the opportunity for accurate skeletal metastasis staging, compared with conventional imaging methods. (18)F-fluoride is a bone-specific tracer whose uptake depends on osteoblastic activity. Because of the resulting increase in bone mineralization and sclerosis, the osteoblastic process can also be detected morphologically in CT images. Although CT is characterized by high resolution, the potential of PET is limited by its lower spatial resolution and the resulting partial-volume effect. In this context, the synergy between PET and CT presents an opportunity to resolve this limitation using a novel multimodal approach called synergistic functional-structural resolution recovery (SFS-RR). Its performance is benchmarked against current resolution recovery technology using the point-spread function (PSF) of the scanner in the reconstruction procedure.

METHODS

The SFS-RR technique takes advantage of the multiresolution property of the wavelet transform applied to both functional and structural images to create a high-resolution PET image that exploits the structural information of CT. Although the method was originally conceived for PET/MR imaging of brain data, an ad hoc version for whole-body PET/CT is proposed here. Three phantom experiments and 2 datasets of metastatic bone (18)F-fluoride PET/CT images from primary prostate and breast cancer were used to test the algorithm performances. The SFS-RR images were compared with the manufacturer's PSF-based reconstruction using the standardized uptake value (SUV) and the metabolic volume as metrics for quantification.

RESULTS

When compared with standard PET images, the phantom experiments showed a bias reduction of 14% in activity and 1.3 cm(3) in volume estimates for PSF images and up to 20% and 2.5 cm(3) for the SFS-RR images. The SFS-RR images were characterized by a higher recovery coefficient (up to 60%) whereas noise levels remained comparable to those of standard PET. The clinical data showed an increase in the SUV estimates for SFS-RR images up to 34% for peak SUV and 50% for maximum SUV and mean SUV. Images were also characterized by sharper lesion contours and better lesion detectability.

CONCLUSION

The proposed methodology generates PET images with improved quantitative and qualitative properties. Compared with standard methods, SFS-RR provides superior lesion segmentation and quantification, which may result in more accurate tumor characterization.

Surface-based partial-volume correction for high-resolution PET

Tissue radioactivity concentrations, measured with positron emission tomography (PET) are subject to partial volume effects (PVE) due to the limited spatial resolution of the scanner. Last generation high-resolution PET cameras with a full width at half maximum (FWHM) of 2-4mm are less prone to PVEs than previous generations. Corrections for PVEs are still necessary, especially when studying small brain stem nuclei or small variations in cortical neuroreceptor concentrations which may be related to cytoarchitectonic differences. Although several partial-volume correction (PVC) algorithms exist, these are frequently based on a priori assumptions about tracer distribution or only yield corrected values of regional activity concentrations without providing PVE corrected images. We developed a new iterative deconvolution algorithm (idSURF) for PVC of PET images that aims to overcome these limitations by using two innovative techniques: 1) the incorporation of anatomic information from a cortical gray matter surface representation, extracted from magnetic resonance imaging (MRI) and 2) the use of anatomically constrained filtering to attenuate noise. PVE corrected images were generated with idSURF implemented into a non-interactive processing pipeline. idSURF was validated using simulated and clinical PET data sets and compared to a frequently used standard PVC method (Geometric Transfer Matrix: GTM). The results on simulated data sets show that idSURF consistently recovers accurate radiotracer concentrations within 1-5% of true values. Both radiotracer concentrations and non-displaceable binding potential (BPnd) values derived from clinical PET data sets with idSURF were highly correlated with those obtained with the standard PVC method (R(2) = 0.99, error = 0%-3.2%). These results suggest that idSURF is a valid and potentially clinically useful PVC method for automatic processing of large numbers of PET data sets.

Denoising PET images using singular value thresholding and Stein's unbiased risk estimate

Image denoising is an important pre-processing step for accurately quantifying functional morphology and measuring activities of the tissues using PET images. Unlike structural imaging modalities, PET images have two difficulties: (1) the Gaussian noise model does not necessarily fit into PET imaging because the exact nature of noise propagation in PET imaging is not well known, and (2) PET images are low resolution; therefore, it is challenging to denoise them while preserving structural information. To address these two difficulties, we introduce a novel methodology for denoising PET images. The proposed method uses the singular value thresholding concept and Stein's unbiased risk estimate to optimize a soft thresholding rule. Results, obtained from 40 MRI-PET images, demonstrate that the proposed algorithm is able to denoise PET images successfully, while still maintaining the quantitative information.

Optimally stabilized PET image denoising using trilateral filtering

Low-resolution and signal-dependent noise distribution in positron emission tomography (PET) images makes denoising process an inevitable step prior to qualitative and quantitative image analysis tasks. Conventional PET denoising methods either over-smooth small-sized structures due to resolution limitation or make incorrect assumptions about the noise characteristics. Therefore, clinically important quantitative information may be corrupted. To address these challenges, we introduced a novel approach to remove signal-dependent noise in the PET images where the noise distribution was considered as Poisson-Gaussian mixed. Meanwhile, the generalized Anscombe's transformation (GAT) was used to stabilize varying nature of the PET noise. Other than noise stabilization, it is also desirable for the noise removal filter to preserve the boundaries of the structures while smoothing the noisy regions. Indeed, it is important to avoid significant loss of quantitative information such as standard uptake value (SUV)-based metrics as well as metabolic lesion volume. To satisfy all these properties, we extended bilateral filtering method into trilateral filtering through multiscaling and optimal Gaussianization process. The proposed method was tested on more than 50 PET-CT images from various patients having different cancers and achieved the superior performance compared to the widely used denoising techniques in the literature.

Segmentation based denoising of PET images: an iterative approach via regional means and affinity propagation

Delineation and noise removal play a significant role in clinical quantification of PET images. Conventionally, these two tasks are considered independent, however, denoising can improve the performance of boundary delineation by enhancing SNR while preserving the structural continuity of local regions. On the other hand, we postulate that segmentation can help denoising process by constraining the smoothing criteria locally. Herein, we present a novel iterative approach for simultaneous PET image denoising and segmentation. The proposed algorithm uses generalized Anscombe transformation priori to non-local means based noise removal scheme and affinity propagation based delineation. For nonlocal means denoising, we propose a new regional means approach where we automatically and efficiently extract the appropriate subset of the image voxels by incorporating the class information from affinity propagation based segmentation. PET images after denoising are further utilized for refinement of the segmentation in an iterative manner. Qualitative and quantitative results demonstrate that the proposed framework successfully removes the noise from PET images while preserving the structures, and improves the segmentation accuracy.

Development and experimental medicine applications of PET in oncology: a historical perspective

Nearly 90 years of scientific research have led to the use of PET and the ability to forge advances in the field of oncology. In this Historial Review we outline the key developments made with this imaging technique, including the evolution of cyclotrons and scanners, together with the associated advances made in image reconstruction, presentation, analysis of data, and radiochemistry. The applications of PET to experimental medicine are summarised, and we cover how these are related to the use and development of PET, especially in the assessment of tumour biology and pharmacology. The use of PET in clinical oncology and for tissue pharmacokinetic and pharmacodynamic studies as a means of supporting drug development is detailed. The current limitations of the technology are also discussed.

Evaluation of a 3D local multiresolution algorithm for the correction of partial volume effects in positron emission tomography

PURPOSE

Partial volume effects (PVEs) are consequences of the limited spatial resolution in emission tomography leading to underestimation of uptake in tissues of size similar to the point spread function (PSF) of the scanner as well as activity spillover between adjacent structures. Among PVE correction methodologies, a voxel-wise mutual multiresolution analysis (MMA) was recently introduced. MMA is based on the extraction and transformation of high resolution details from an anatomical image (MR/CT) and their subsequent incorporation into a low-resolution PET image using wavelet decompositions. Although this method allows creating PVE corrected images, it is based on a 2D global correlation model, which may introduce artifacts in regions where no significant correlation exists between anatomical and functional details.

METHODS

A new model was designed to overcome these two issues (2D only and global correlation) using a 3D wavelet decomposition process combined with a local analysis. The algorithm was evaluated on synthetic, simulated and patient images, and its performance was compared to the original approach as well as the geometric transfer matrix (GTM) method.

RESULTS

Quantitative performance was similar to the 2D global model and GTM in correlated cases. In cases where mismatches between anatomical and functional information were present, the new model outperformed the 2D global approach, avoiding artifacts and significantly improving quality of the corrected images and their quantitative accuracy.

CONCLUSIONS

A new 3D local model was proposed for a voxel-wise PVE correction based on the original mutual multiresolution analysis approach. Its evaluation demonstrated an improved and more robust qualitative and quantitative accuracy compared to the original MMA methodology, particularly in the absence of full correlation between anatomical and functional information.

Presurgical epilepsy localization with interictal cerebral dysfunction

Localization of interictal cerebral dysfunction with 2-[(18)F]fluoro-2-D-deoxyglucose (FDG) positron emission tomography (PET) and neuropsychological examination usefully supplements electroencephalography (EEG) and brain magnetic resonance imaging (MRI) in planning epilepsy surgery. In MRI-negative mesial temporal lobe epilepsy, correlation of temporal lobe hypometabolism with extracranial ictal EEG can support resection without prior intracranial EEG monitoring. In refractory localization-related epilepsies, hypometabolic sites may supplement other data in hypothesizing likely ictal onset zones in order to intracranial electrodes for ictal recording. Prognostication of postoperative seizure freedom with FDG PET appears to have greater positive than negative predictive value. Neuropsychological evaluation is critical to evaluating the potential benefit of epilepsy surgery. Cortical deficits measured with neuropsychometry are limited in lateralizing and localizing value for determination of ictal onset sites, however. Left temporal resection risks iatrogenic verbal memory deficits and dysnomia, and neuropsychological findings are useful in predicting those at greatest risk. Prognostication of cognitive risks with resection at other sites is less satisfactory.

Global-two-stage filtering of clinical PET parametric maps: application to [(11)C]-(R)-PK11195

INTRODUCTION

In Positron Emission Tomography (PET) quantification of physiological parameters at the voxel level may result in unreliable estimates due to the high noise of voxel time activity curves. Global-Two-Stage (GTS), an estimation technique belonging to the group of "population approaches", can be used to tackle this problem. GTS was previously tested on simulated PET data and yielded substantial improvements when compared to standard estimation approaches such as Weighted NonLinear Least Squares (WNLLS) and Basis Function Method (BFM). In this work GTS performance is assessed in a clinical context using the neuroinflammation marker [(11)C]-(R)-PK11195 applied to a cohort of Huntington's disease (HD) patients with and without symptoms.

MATERIALS AND METHODS

Parametric maps of binding potential (BP(ND)) of 12 normal controls (NC), 9 symptomatic and 9 presymptomatic HD patients were generated by applying a modified reference tissue model that accounts for tracer vascular activity in both reference and target tissues (SRTMV). GTS was then applied to SRTMV maps and its performance compared with that of SRTMV. Three smoothed versions of SRTMV, obtained by filtering the original SRTMV maps with Gaussian filters of 3 mm, 5 mm and 7 mm Full Width Half Maximum (FWHM), were also included in the comparison. Since striatal degeneration is the hallmark of HD, sensitivity was assessed for all methods by computing the mean of z-scores in caudate, putamen and globus pallidus in the voxel-by-voxel statistical comparison of BP(ND) between HD and NC.

RESULTS

Application of GTS to parametric maps brought a substantial qualitative improvement to SRTMV maps to the extent that anatomical structures often became visible. In addition, most parameter estimates that were outside the physiological range with SRTMV were corrected by GTS. GTS yielded a 2.3-fold increase in sensitivity with respect to SRTMV for the symptomatic cohort (mean of striatal z-scores of 0.76 for SRTMV and 1.79 for GTS) and an even more substantial increase for the presymptomatic cohort (mean of striatal z-scores of 0.34 for SRTMV and 0.96 for GTS). The sensitivity of GTS was similar to the one obtained with a filter of 7 mm FWHM applied to the initial SRTMV maps but GTS images were not characterized by the notable loss of resolution typical of smoothed maps. GTS, additionally, does not require to change/define settings according to the tracer and level of noise, whereas the choice of the FWHM value of the Gaussian filter normally employed in the smoothing procedure is typically arbitrary.

CONCLUSIONS

GTS is a powerful and robust tool for improving the quality of parametric maps in PET. The method is particularly appealing in that it can be applied to any tracer and estimation method, provided that initial estimates of the parameter vector and of its covariance are available. Although the benefits of GTS are far from being exhaustively assessed, the significant improvements obtained both on real and simulated data suggest that it could become an important tool for dynamic PET in the future.

Functional and structural synergy for resolution recovery and partial volume correction in brain PET

PURPOSE

Positron Emission Tomography (PET) has the unique capability of measuring brain function but its clinical potential is affected by low resolution and lack of morphological detail. Here we propose and evaluate a wavelet synergistic approach that combines functional and structural information from a number of sources (CT, MRI and anatomical probabilistic atlases) for the accurate quantitative recovery of radioactivity concentration in PET images. When the method is combined with anatomical probabilistic atlases, the outcome is a functional volume corrected for partial volume effects.

METHODS

The proposed method is based on the multiresolution property of the wavelet transform. First, the target PET image and the corresponding anatomical image (CT/MRI/atlas-based segmented MRI) are decomposed into several resolution elements. Secondly, high-resolution components of the PET image are replaced, in part, with those of the anatomical image after appropriate scaling. The amount of structural input is weighted by the relative high frequency signal content of the two modalities. The method was validated on a digital Zubal phantom and clinical data to evaluate its quantitative potential.

RESULTS

Simulation studies showed the expected relationship between functional recovery and the amount of correct structural detail provided, with perfect recovery achieved when true images were used as anatomical reference. The use of T1-MRI images brought significant improvements in PET image resolution. However improvements were maximized when atlas-based segmented images as anatomical references were used; these results were replicated in clinical data sets.

CONCLUSION

The synergistic use of functional and structural data, and the incorporation of anatomical probabilistic information in particular, generates morphologically corrected PET images of exquisite quality.

Partial volume effect correction in PET using regularized iterative deconvolution with variance control based on local topology

Correcting positron emission tomography (PET) images for the partial volume effect (PVE) due to the limited resolution of PET has been a long-standing challenge. Various approaches including incorporation of the system response function in the reconstruction have been previously tested. We present a post-reconstruction PVE correction based on iterative deconvolution using a 3D maximum likelihood expectation-maximization (MLEM) algorithm. To achieve convergence we used a one step late (OSL) regularization procedure based on the assumption of local monotonic behavior of the PET signal following Alenius et al. This technique was further modified to selectively control variance depending on the local topology of the PET image. No prior 'anatomic' information is needed in this approach. An estimate of the noise properties of the image is used instead. The procedure was tested for symmetric and isotropic deconvolution functions with Gaussian shape and full width at half-maximum (FWHM) ranging from 6.31 mm to infinity. The method was applied to simulated and experimental scans of the NEMA NU 2 image quality phantom with the GE Discovery LS PET/CT scanner. The phantom contained uniform activity spheres with diameters ranging from 1 cm to 3.7 cm within uniform background. The optimal sphere activity to variance ratio was obtained when the deconvolution function was replaced by a step function few voxels wide. In this case, the deconvolution method converged in approximately 3-5 iterations for most points on both the simulated and experimental images. For the 1 cm diameter sphere, the contrast recovery improved from 12% to 36% in the simulated and from 21% to 55% in the experimental data. Recovery coefficients between 80% and 120% were obtained for all larger spheres, except for the 13 mm diameter sphere in the simulated scan (68%). No increase in variance was observed except for a few voxels neighboring strong activity gradients and inside the largest spheres. Testing the method for patient images increased the visibility of small lesions in non-uniform background and preserved the overall image quality. Regularized iterative deconvolution with variance control based on the local properties of the PET image and on estimated image noise is a promising approach for partial volume effect corrections in PET.