Positron Emission Tomography Image Denoising Research Articles

MPGAN: Multi Pareto Generative Adversarial Network for the denoising and quantitative analysis of low-dose PET images of human brain

Positron emission tomography (PET) imaging is widely used in medical imaging for analyzing neurological disorders and related brain diseases. Usually, full-dose imaging for PET ensures image quality but raises concerns about potential health risks of radiation exposure. The contradiction between reducing radiation exposure and maintaining diagnostic performance can be effectively addressed by reconstructing low-dose PET (L-PET) images to the same high-quality as full-dose (F-PET). This paper introduces the Multi Pareto Generative Adversarial Network (MPGAN) to achieve 3D end-to-end denoising for the L-PET images of human brain. MPGAN consists of two key modules: the diffused multi-round cascade generator (GDmc) and the dynamic Pareto-efficient discriminator (DPed), both of which play a zero-sum game for n(n∈1,2,3) rounds to ensure the quality of synthesized F-PET images. The Pareto-efficient dynamic discrimination process is introduced in DPed to adaptively adjust the weights of sub-discriminators for improved discrimination output. We validated the performance of MPGAN using three datasets, including two independent datasets and one mixed dataset, and compared it with 12 recent competing models. Experimental results indicate that the proposed MPGAN provides an effective solution for 3D end-to-end denoising of L-PET images of the human brain, which meets clinical standards and achieves state-of-the-art performance on commonly used metrics.

OIF-Net: An Optical Flow Registration-Based PET/MR Cross-Modal Interactive Fusion Network for Low-Count Brain PET Image Denoising.

The short frames of low-count positron emission tomography (PET) images generally cause high levels of statistical noise. Thus, improving the quality of low-count images by using image postprocessing algorithms to achieve better clinical diagnoses has attracted widespread attention in the medical imaging community. Most existing deep learning-based low-count PET image enhancement methods have achieved satisfying results, however, few of them focus on denoising low-count PET images with the magnetic resonance (MR) image modality as guidance. The prior context features contained in MR images can provide abundant and complementary information for single low-count PET image denoising, especially in ultralow-count (2.5%) cases. To this end, we propose a novel two-stream dual PET/MR cross-modal interactive fusion network with an optical flow pre-alignment module, namely, OIF-Net. Specifically, the learnable optical flow registration module enables the spatial manipulation of MR imaging inputs within the network without any extra training supervision. Registered MR images fundamentally solve the problem of feature misalignment in the multimodal fusion stage, which greatly benefits the subsequent denoising process. In addition, we design a spatial-channel feature enhancement module (SC-FEM) that considers the interactive impacts of multiple modalities and provides additional information flexibility in both the spatial and channel dimensions. Furthermore, instead of simply concatenating two extracted features from these two modalities as an intermediate fusion method, the proposed cross-modal feature fusion module (CM-FFM) adopts cross-attention at multiple feature levels and greatly improves the two modalities' feature fusion procedure. Extensive experimental assessments conducted on real clinical datasets, as well as an independent clinical testing dataset, demonstrate that the proposed OIF-Net outperforms the state-of-the-art methods.

Simultaneous Denoising of Dynamic PET Images Based on Deep Image Prior.

Parametric imaging obtained from kinetic modeling analysis of dynamic positron emission tomography (PET) data is a useful tool for quantifying tracer kinetics. However, pixel-wise time-activity curves have high noise levels which lead to poor quality of parametric images. To solve this limitation, we proposed a new image denoising method based on deep image prior (DIP). Like the original DIP method, the proposed DIP method is an unsupervised method, in which no training dataset is required. However, the difference is that our method can simultaneously denoise all dynamic PET images. Moreover, we propose a modified version of the DIP method called double DIP (DDIP), which has two DIP architectures. The additional DIP model is used to generate high-quality input data for the second DIP model. Computer simulations were performed to evaluate the performance of the proposed DIP-based methods. Our simulation results showed that the DDIP method outperformed the single DIP method. In addition, the DDIP method combined with data augmentation could generate PET parametric images with superior image quality compared to the spatiotemporal-based non-local means filtering and high constrained backprojection. Our preliminary results show that our proposed DDIP method is a novel and effective unsupervised method for simultaneously denoising dynamic PET images.

4D deep image prior: dynamic PET image denoising using an unsupervised four-dimensional branch convolutional neural network

Although convolutional neural networks (CNNs) demonstrate the superior performance in denoising positron emission tomography (PET) images, a supervised training of the CNN requires a pair of large, high-quality PET image datasets. As an unsupervised learning method, a deep image prior (DIP) has recently been proposed; it can perform denoising with only the target image. In this study, we propose an innovative procedure for the DIP approach with a four-dimensional (4D) branch CNN architecture in end-to-end training to denoise dynamic PET images. Our proposed 4D CNN architecture can be applied to end-to-end dynamic PET image denoising by introducing a feature extractor and a reconstruction branch for each time frame of the dynamic PET image. In the proposed DIP method, it is not necessary to prepare high-quality and large patient-related PET images. Instead, a subject’s own static PET image is used as additional information, dynamic PET images are treated as training labels, and denoised dynamic PET images are obtained from the CNN outputs. Both simulation with [18F]fluoro-2-deoxy-D-glucose (FDG) and preclinical data with [18F]FDG and [11C]raclopride were used to evaluate the proposed framework. The results showed that our 4D DIP framework quantitatively and qualitatively outperformed 3D DIP and other unsupervised denoising methods. The proposed 4D DIP framework thus provides a promising procedure for dynamic PET image denoising.

Dynamic PET Image Denoising With Deep Learning-Based Joint Filtering

Dynamic positron emission tomography (PET) imaging usually suffers from high statistical noise due to low counts of the short frames. This study aims to improve the image quality of the short frames by utilizing information from other modality. We develop a deep learning-based joint filtering framework for simultaneously incorporating information from longer acquisition PET frames and high-resolution magnetic resonance (MR) images into the short frames. The network inputs are noisy PET images and corresponding MR images while the outputs are linear coefficients of spatially variant linear representation model. The composite of all dynamic frames is used as training label in each sample, and it is down-sampled to 1/10th of counts as the training input. L1-norm combined with two gradient-based regularizations constitute the loss function during training. Ten realistic dynamic PET/MR phantoms based on BrainWeb are used for pre-training and eleven clinical subjects from Alzheimer's Disease Neuroimaging Initiative further for fine-tuning. Simulation results show that the proposed method can reduce the statistical noise while preserving image details and achieve quantitative enhancements compared with Gaussian, guided filter, and convolutional neural network trained with the mean squared error. The clinical results perform better than others in terms of the mean activity and standard deviation. All of the results indicate that the proposed deep learning-based joint filtering framework is of great potential for dynamic PET image denoising.

Parameter-Transferred Wasserstein Generative Adversarial Network (PT-WGAN) for Low-Dose PET Image Denoising.

Due to the widespread use of positron emission tomography (PET) in clinical practice, the potential risk of PET-associated radiation dose to patients needs to be minimized. However, with the reduction in the radiation dose, the resultant images may suffer from noise and artifacts that compromise diagnostic performance. In this paper, we propose a parameter-transferred Wasserstein generative adversarial network (PT-WGAN) for low-dose PET image denoising. The contributions of this paper are twofold: i) a PT-WGAN framework is designed to denoise low-dose PET images without compromising structural details, and ii) a task-specific initialization based on transfer learning is developed to train PT-WGAN using trainable parameters transferred from a pretrained model, which significantly improves the training efficiency of PT-WGAN. The experimental results on clinical data show that the proposed network can suppress image noise more effectively while preserving better image fidelity than recently published state-of-the-art methods. We make our code available at https://github.com/90n9-yu/PT-WGAN.

An efficient method for PET image denoising by combining multi-scale transform and non-local means

The diagnosis of dementia, particularly in the early stages is very much helpful with Positron emission tomography (PET) image processing. The most important challenges in PET image processing are noise removal and region of interests (ROIs) segmentation. Although denoising and segmentation are performed independently, but the performance of the denoising process significantly affects the performance of the segmentation process. Due to the low signals to noise ratio and low contrast, PET image denoising is a challenging task. Individual wavelet, curvelet and non-local means (NLM) based methods are not well suited to handle both isotropic (smooth details) and anisotropic (edges and curves) features due to its restricted abilities. To address these issues, the present work proposes an efficient denoising framework for reducing the noise level of brain PET images based on the combination of multi-scale transform (wavelet and curvelet) and tree clustering non-local means (TNLM). The main objective of the proposed method is to extract the isotropic features from a noisy smooth PET image using tree clustering based non-local means (TNLM). Then curvelet-based denoising is applied to the residual image to extract the anisotropic features such as edges and curves. Finally, the extracted anisotropic features are inserted back into the isotropic features to obtain an estimated denoised image. Simulated phantom and clinical PET datasets have been used in this proposed work for testing and measuring the performance in the medical applications, such as gray matter segmentation and precise tumor region identification without any interaction with other structural images like MRI or CT. The results in the experimental section show that the proposed denoising method has obtained better performance than existing wavelet, curvelet, wavelet-curvelet, non-local means (NLM) and deep learning methods based on the preservation of the edges. Qualitatively, a notable gain is achieved in the proposed denoised PET images in terms of contrast enhancement than other existing denoising methods.

Kinetics-Induced Block Matching and 5-D Transform Domain Filtering for Dynamic PET Image Denoising

Dynamic positron emission tomography (PET) scans of short-time frames are required to quantitatively estimate the uptake of PET ligands. Because such short-frame scans tend to be noisy, we propose kinetics-induced block matching and 5-D transform domain filtering (KIBM5D) specialized for dynamic PET image denoising. In the proposed algorithm, kinetics-induced block matching (KIBM) and 5-D transform domain filtering are alternately repeated in two cascading stages. In each stage of KIBM5D, all time frames are included in a patch of the KIBM to collect similar patch-wise time activity curves. These similar 4-D patches are then five-dimensionally grouped and transformed to the 5-D spectrum. In the 5-D transform domain, the 5-D spectrum is shrunk by hard thresholding and Wiener filtering in the first and second stage of KIBM5D, respectively. The sparsity of the 5-D spectrum is improved because signals of similar 4-D patches are correlated, while noises of these are uncorrelated. To evaluate the performance of KIBM5D, we used both computer simulation data of a dynamic digital brain phantom using [18F]FDG kinetics, and experimental data of a normal healthy volunteer using [11C]MeQAA, and compared the results of KIBM5D, Gaussian filter (GF), bilateral filter, nonlocal means, block matching and 4-D filtering, and 4-D Gaussian filtering. For simulation data, KIBM5D performed superiorly to the other methods in terms of the peak signal to noise ratio and structural similarity measures, in all time frames. Additionally, KIBM5D generated the best image quality not only in simulations but also with human data. Accordingly, KIBM5D enables the efficient denoising of dynamic PET images.

An efficient wavelet and curvelet-based PET image denoising technique.

Positron emission tomography (PET) image denoising is a challenging task due to the presence of noise and low spatial resolution compared with other imaging techniques such as magnetic resonance imaging (MRI) and computed tomography (CT). PET image noise can hamper further processing and analysis, such as segmentation and disease screening. The wavelet transform-based techniques have often been proposed for PET image denoising to handle isotropic (smooth details) features. The curvelet transform-based PET image denoising techniques have the ability to handle multi-scale and multi-directional properties such as edges and curves (anisotropic features) as compared with wavelet transform-based denoising techniques. The wavelet denoising method is not optimal for anisotropic features, whereas the curvelet denoising method sometimes has difficulty in handling isotropic features. In order to handle the weaknesses of individual wavelet and curvelet-based methods, the present research proposes an efficient PET image denoising technique based on the combination of wavelet and curvelet transforms, along with a new adaptive threshold selection to threshold the wavelet coefficients in each subband (except last level low pass (LL) residual). The proposed threshold utilizes the advantages of adaptive threshold taken from BayesShrink along with the neighborhood window concept. The present method was tested on both simulated phantom and clinical PET datasets. Experimental results show that our method has achieved better results than the existing methods such as VisuShrink, BayesShrink, NeighShrink, ModineighShrink, curvelet, and an existing wavelet curvelet-based method with respect to different noise measurement metrics, such as mean squared error (MSE), signal-to-noise ratio (SNR), peak signal-to-noise ratio (PSNR), and image quality index (IQI). Furthermore, notable performance is achieved in the case of medical applications such as gray matter segmentation and precise tumor region identification. Graphical Abstract Block diagram of the proposed method. (a) Steps of the proposed denoising method and (b) image information generated by each step of (a).

PET image denoising using unsupervised deep learning.

Image quality of positron emission tomography (PET) is limited by various physical degradation factors. Our study aims to perform PET image denoising by utilizing prior information from the same patient. The proposed method is based on unsupervised deep learning, where no training pairs are needed. In this method, the prior high-quality image from the patient was employed as the network input and the noisy PET image itself was treated as the training label. Constrained by the network structure and the prior image input, the network was trained to learn the intrinsic structure information from the noisy image and output a restored PET image. To validate the performance of the proposed method, a computer simulation study based on the BrainWeb phantom was first performed. A 68Ga-PRGD2 PET/CT dataset containing 10 patients and a 18F-FDG PET/MR dataset containing 30 patients were later on used for clinical data evaluation. The Gaussian, non-local mean (NLM) using CT/MR image as priors, BM4D, and Deep Decoder methods were included as reference methods. The contrast-to-noise ratio (CNR) improvements were used to rank different methods based on Wilcoxon signed-rank test. For the simulation study, contrast recovery coefficient (CRC) vs. standard deviation (STD) curves showed that the proposed method achieved the best performance regarding the bias-variance tradeoff. For the clinical PET/CT dataset, the proposed method achieved the highest CNR improvement ratio (53.35% ± 21.78%), compared with the Gaussian (12.64% ± 6.15%, P= 0.002), NLM guided by CT (24.35% ± 16.30%, P= 0.002), BM4D (38.31% ± 20.26%, P= 0.002), and Deep Decoder (41.67% ± 22.28%, P= 0.002) methods. For the clinical PET/MR dataset, the CNR improvement ratio of the proposed method achieved 46.80% ± 25.23%, higher than the Gaussian (18.16% ± 10.02%, P< 0.0001), NLM guided by MR (25.36% ± 19.48%, P< 0.0001), BM4D (37.02% ± 21.38%, P< 0.0001), and Deep Decoder (30.03% ± 20.64%, P< 0.0001) methods. Restored images for all the datasets demonstrate that the proposed method can effectively smooth out the noise while recovering image details. The proposed unsupervised deep learning framework provides excellent image restoration effects, outperforming the Gaussian, NLM methods, BM4D, and Deep Decoder methods.

Segmentation improvement through denoising of PET images with 3D-context modelling in wavelet domain

Positron emission tomography (PET) images have been incorporated into the radiotherapy process as a powerful tool to assist in the contouring of lesions, leading to the emergence of a broad spectrum of automatic segmentation schemes for PET images (PET-AS). However, not all proposed PET-AS algorithms take into consideration the previous steps of image preparation. PET image noise has been shown to be one of the most relevant affecting factors in segmentation tasks. This study demonstrates a nonlinear filtering method based on spatially adaptive wavelet shrinkage using three-dimensional context modelling that considers the correlation of each voxel with its neighbours. Using this noise reduction method, excellent edge conservation properties are obtained. To evaluate the influence in the segmentation schemes of this filter, it was compared with a set of Gaussian filters (the most conventional) and with two previously optimised edge-preserving filters. Five segmentation schemes were used (most commonly implemented in commercial software): fixed thresholding, adaptive thresholding, watershed, adaptive region growing and affinity propagation clustering. Segmentation results were evaluated using the Dice similarity coefficient and classification error. A simple metric was also included to improve the characterisation of the filters used for induced blurring evaluation, based on the measurement of the average edge width. The proposed noise reduction procedure improves the results of segmentation throughout the performed settings and was shown to be more stable in low-contrast and high-noise conditions. Thus, the capacity of the segmentation method is reinforced by the denoising plan used.

Denoising of dynamic PET images using a multi-scale transform and non-local means filter

Abstract The quantification of positron emission tomography (PET) images requires a time activity curve (TAC) to provide an accurate estimation of kinetic parameters. However, the low signals to noise ratio (SNR), the important level of noise, and the low spatial resolution of PET image make the extraction of the TAC a challenging task. In this study, we present a new method based on multi-scale and non-local means method (MNLM) to reduce noise in dynamic PET sequences of small animal heart. MNLM filter takes into account the temporal correlation between images in the dynamic measurement and benefits from the complementary properties of both the Shearlet transform and the wavelet transform to provide best reduction. The method was tested on dynamic digital mouse phantom and a preclinical rat study (n = 6). Based on a comparative study with three major algorithms reviewed on the state of the art, the data analysis proved the significance of the MNLM filter. In simulated data, the major finding of the study showed that at the highest noise level (7.68%), the model gave the best result (Chi-square = 4.06). Furthermore, it presented a notable gain in terms of PSNR and SSIM plot. In real data, the MNLM showed a better result in the computation of the contrast metric with a value of 27.04 ∓ 12.1 and the highest SNR with a value of 74.38 ∓ 9.2. This approach proved a better potential and could be considered as a valuable candidate to reduce noise in clinical system.

Denoising of PET images by context modelling using local neighbourhood correlation

Positron emission tomography (PET) images are characterised by low signal-to-noise ratio and blurred edges when compared with other image modalities. It is therefore advisable to use noise reduction methods for qualitative and quantitative analyses. Given the importance of the maximum and mean uptake values, it is necessary to avoid signal loss, which could modify the clinical significance. This paper proposes a method of non-linear image denoising for PET. It is based on spatially adaptive wavelet-shrinkage and uses context modelling, which explicitly considers the correlation between neighbouring pixels. This context modelling is able to maintain the uptake values and preserve the edges in significant regions. The algorithm is proposed as an alternative to the usual filtering that is performed after reconstruction.

Postreconstruction filtering of 3D PET images by using weighted higher-order singular value decomposition.

BackgroundPositron emission tomography (PET) always suffers from high levels of noise due to the constraints of the injected dose and acquisition time, especially in the studies of dynamic PET imaging. To improve the quality of PET image, several approaches have been introduced to suppress noise. However, traditional filters often blur the image edges, or erase small detail, or rely on multiple parameters. In order to solve such problems, nonlocal denoising methods have been adapted to denoise PET images.MethodsIn this paper, we propose to use the weighted higher-order singular value decomposition for PET image denoising. We first modeled the noise in the PET image as Poisson distribution. Then, we transformed the noise to an additive Gaussian noise by use of the anscombe root transformation. Finally, we denoised the transformed image using the proposed higher-order singular value decomposition (HOSVD)-based algorithms. The denoised results were compared with results from some general filters by performing physical phantom and mice studies.ResultsCompared to other commonly used filters, HOSVD-based denoising algorithms can preserve boundaries and quantitative accuracy better. The spatial resolution and the low activity features in PET image also can be preserved by use of HOSVD-based methods. Comparing with the standard HOSVD-based algorithm, the proposed weighted HOSVD algorithm can suppress the stair-step artifact, and the time-consumption is about half of that needed by the Wiener-augmented HOSVD algorithm.ConclusionsThe proposed weighted HOSVD denoising algorithm can suppress noise while better preserving of boundary and quantity in PET images.Electronic supplementary materialThe online version of this article (doi:10.1186/s12938-016-0221-y) contains supplementary material, which is available to authorized users.

Denoising of PET images by combining wavelets and curvelets for improved preservation of resolution and quantitation

Denoising of Positron Emission Tomography (PET) images is a challenging task due to the inherent low signal-to-noise ratio (SNR) of the acquired data. A pre-processing denoising step may facilitate and improve the results of further steps such as segmentation, quantification or textural features characterization. Different recent denoising techniques have been introduced and most state-of-the-art methods are based on filtering in the wavelet domain. However, the wavelet transform suffers from some limitations due to its non-optimal processing of edge discontinuities. More recently, a new multi scale geometric approach has been proposed, namely the curvelet transform. It extends the wavelet transform to account for directional properties in the image. In order to address the issue of resolution loss associated with standard denoising, we considered a strategy combining the complementary wavelet and curvelet transforms. We compared different figures of merit (e.g. SNR increase, noise decrease in homogeneous regions, resolution loss, and intensity bias) on simulated and clinical datasets with the proposed combined approach and the wavelet-only and curvelet-only filtering techniques. The three methods led to an increase of the SNR. Regarding the quantitative accuracy however, the wavelet and curvelet only denoising approaches led to larger biases in the intensity and the contrast than the proposed combined algorithm. This approach could become an alternative solution to filters currently used after image reconstruction in clinical systems such as the Gaussian filter.

Assessment of the Wavelet Transform for Noise Reduction in Simulated PET Images

Introduction: An efficient method of tomographic imaging in nuclear medicine is positron emission tomography (PET). Compared to SPECT, PET has the advantages of higher levels of sensitivity, spatial resolution and more accurate quantification. However, high noise levels in the image limit its diagnostic utility. Noise removal in nuclear medicine is traditionally based on Fourier decomposition of images. This method is based on frequency components, irrespective of the spatial location of the noise or signal. The wavelet transform presents a solution by providing information on the frequency content while retaining spatial information. This alleviates the shortcoming of the Fourier transform and thus, wavelet transform has been extensively used for noise reduction, edge detection and compression. Materials and Methods: In this research, we used the SimSET software to simulate PET images of the NCAT phantom. The images were acquired using 250 million counts in a 128×128 matrix. For the reference image, we acquired an image with high counts (6 billion). Then, we reconstructed these images using our own software developed in MATLAB. After image reconstruction, a 250 million counts image (noisy image) and a reference image were normalized and then root-mean-square error (RMSE) was used to compare the images. Next, we wrote and applied de-noising programs. These programs were based on using 54 different wavelets and 4 methods. De-noised images were compared with the reference image using RMSE. Results: Our results indicate that the Stationary Wavelet Transform and Global Thresholding are more efficient at noise reduction compared to the other methods that we investigated. Discussion: The wavelet transform is a useful method for de-noising of simulated PET images. Noise reduction using this transform and loss of high-frequency information are simultaneous with each other. It seems that we should attend to the mutual agreement between noise reduction and the visual quality of the image

PET kinetic analysis: wavelet denoising of dynamic PET data with application to parametric imaging

Physiological functions (e.g., cerebral blood flow, glucose metabolism, and neuroreceptor binding) can be investigated as parameters estimated by kinetic modeling using dynamic positron emission tomography (PET) images. Imaging of these physiological parameters, called parametric imaging, can locate the regional distribution of functionalities. However, the most serious technical issue affecting parametric imaging is noise in dynamic PET data. This review describes wavelet denoising of dynamic PET images for improving image quality in estimated parametric images. Wavelet denoising provides significantly improved quality directly to dynamic PET images and indirectly to estimated parametric images. The application of wavelet denoising to radio-ligand and kinetic analysis is still in the development stage, but even so, it is thought that wavelet techniques will have a substantial impact on nuclear medicine in the near future.