Search

Objective. Multi-energy CT conducted by photon-counting detector has a wide range of applications, especially in multiple contrast agents imaging. However, static multi-energy (SME) CT imaging suffers from higher statistical noise because of increased energy bins with static energy thresholds. Our team has proposed a dynamic dual-energy (DDE) CT detector model and the corresponding iterative reconstruction algorithm to solve this problem. However, rigorous and detailed analysis of the statistical noise characterization in this DDE CT was lacked. Approach. Starting from the properties of the Poisson random variable, this paper analyzes the noise characterization of the DDE CT and compares it with the SME CT. It is proved that the multi-energy CT projections and reconstruction images calculated from the proposed DDE CT algorithm have less statistical noise than that of the SME CT. Main results. Simulations and experiments verify that the expectations of the multi-energy CT projections calculated from DDE CT are the same as those of the SME projections. Still, the variance of the former is smaller. We further analyze the convergence of the iterative DDE CT algorithm through simulations and prove that the derived noise characterization can be realized under different CT imaging configurations. Significance. The low statistical noise characteristics demonstrate the value of DDE CT imaging technology., (© 2024 Institute of Physics and Engineering in Medicine. All rights, including for text and data mining, AI training, and similar technologies, are reserved.)

Objective. Optical coherence tomography (OCT) is widely used in clinical practice for its non-invasive, high-resolution imaging capabilities. However, speckle noise inherent to its low coherence principle can degrade image quality and compromise diagnostic accuracy. While deep learning methods have shown promise in reducing speckle noise, obtaining well-registered image pairs remains challenging, leading to the development of unpaired methods. Despite their potential, existing unpaired methods suffer from redundancy in network structures or interaction mechanisms. Therefore, a more streamlined method for unpaired OCT denoising is essential. Approach. In this work, we propose a novel unpaired method for OCT image denoising, referred to as noise-imitation learning (NIL). NIL comprises three primary modules: the noise extraction module, which extracts noise features by denoising noisy images; the noise imitation module, which synthesizes noisy images and generates fake clean images; and the adversarial learning module, which differentiates between real and fake clean images through adversarial training. The complexity of NIL is significantly lower than that of previous unpaired methods, utilizing only one generator and one discriminator for training. Main results. By efficiently fusing unpaired images and employing adversarial training, NIL can extract more speckle noise information to enhance denoising performance. Building on NIL, we propose an OCT image denoising pipeline, NIL-NAFNet. This pipeline achieved PSNR, SSIM, and RMSE values of 31.27 dB, 0.865, and 7.00, respectively, on the PKU37 dataset. Extensive experiments suggest that our method outperforms state-of-the-art unpaired methods both qualitatively and quantitatively. Significance. These findings indicate that the proposed NIL is a simple yet effective method for unpaired OCT speckle noise reduction. The OCT denoising pipeline based on NIL demonstrates exceptional performance and efficiency. By addressing speckle noise without requiring well-registered image pairs, this method can enhance image quality and diagnostic accuracy in clinical practice., (© 2024 Institute of Physics and Engineering in Medicine. All rights, including for text and data mining, AI training, and similar technologies, are reserved.)

Objective. Magnetic particle imaging (MPI) is an emerging tracer-based in vivo imaging technology. The use of MPI at low superparamagnetic iron oxide nanoparticle concentrations has the potential to be a promising area of clinical application due to the inherent safety for humans. However, low tracer concentrations reduce the signal-to-noise ratio of the magnetization signal, leading to severe noise artifacts in the reconstructed MPI images. Hardware improvements have high complexity, while traditional methods lack robustness to different noise levels, making it difficult to improve the quality of low concentration MPI images. Approach. Here, we propose a novel deep learning method for MPI image denoising and quality enhancing based on a sparse lightweight transformer model. The proposed residual-local transformer structure reduces model complexity to avoid overfitting, in which an information retention block facilitates feature extraction capabilities for the image details. Besides, we design a noisy concentration dataset to train our model. Then, we evaluate our method with both simulated and real MPI image data. Main results. Simulation experiment results show that our method can achieve the best performance compared with the existing deep learning methods for MPI image denoising. More importantly, our method is effectively performed on the real MPI image of samples with an Fe concentration down to 67 μ g Fe ml -1 . Significance. Our method provides great potential for obtaining high quality MPI images at low concentrations., (© 2024 Institute of Physics and Engineering in Medicine. All rights, including for text and data mining, AI training, and similar technologies, are reserved.)

Objective . Low-count positron emission tomography (PET) imaging is an efficient way to promote more widespread use of PET because of its short scan time and low injected activity. However, this often leads to low-quality PET images with clinical image reconstruction, due to high noise and blurring effects. Existing PET image restoration (IR) methods hinder their own restoration performance due to the semi-convergence property and the lack of suitable denoiser prior. Approach . To overcome these limitations, we propose a novel deep plug-and-play IR method called Deep denoiser Prior driven Relaxed Iterated Tikhonov method (DP-RI-Tikhonov). Specifically, we train a deep convolutional neural network denoiser to generate a flexible deep denoiser prior to handle high noise. Then, we plug the deep denoiser prior as a modular part into a novel iterative optimization algorithm to handle blurring effects and propose an adaptive parameter selection strategy for the iterative optimization algorithm. Main results . Simulation results show that the deep denoiser prior plays the role of reducing noise intensity, while the novel iterative optimization algorithm and adaptive parameter selection strategy can effectively eliminate the semi-convergence property. They enable DP-RI-Tikhonov to achieve an average quantitative result (normalized root mean square error, structural similarity) of (0.1364, 0.9574) at the stopping iteration, outperforming a conventional PET IR method with an average quantitative result of (0.1533, 0.9523) and a state-of-the-art deep plug-and-play IR method with an average quantitative result of (0.1404, 0.9554). Moreover, the advantage of DP-RI-Tikhonov becomes more obvious at the last iteration. Experiments on six clinical whole-body PET images further indicate that DP-RI-Tikhonov successfully reduces noise intensity and recovers fine details, recovering sharper and more uniform images than the comparison methods. Significance . DP-RI-Tikhonov's ability to reduce noise intensity and effectively eliminate the semi-convergence property overcomes the limitations of existing methods. This advancement may have substantial implications for other medical IR., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective. In helical tomotherapy, image-guided radiotherapy employs megavoltage computed tomography (MVCT) for precise targeting. However, the high voltage of megavoltage radiation introduces substantial noise, significantly compromising MVCT image clarity. This study aims to enhance MVCT image quality using a deep learning-based denoising method. Approach. We propose an unpaired MVCT denoising network using a coupled generative adversarial network framework (DeCoGAN). Our approach assumes that a universal latent code within a shared latent space can reconstruct any given pair of images. By employing an encoder, we enforce this shared-latent space constraint, facilitating the conversion of low-quality (noisy) MVCT images into high-quality (denoised) counterparts. The network learns the joint distribution of images from both domains by leveraging samples from their respective marginal distributions, enhanced by adversarial training for effective denoising. Main Results. Compared to an analytical algorithm (BM3D) and three deep learning-based methods (RED-CNN, WGAN-VGG and CycleGAN), the proposed method excels in preserving image details and enhancing human visual perception by removing most noise and retaining structural features. Quantitative analysis demonstrates that our method achieves the highest peak signal-to-noise ratio and Structural Similarity Index Measurement values, indicating superior denoising performance. Significance. The proposed DeCoGAN method shows remarkable MVCT denoising performance, making it a promising tool in the field of radiation therapy., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective. The mean squared error (MSE), also known as L loss, has been widely used as a loss function to optimize image denoising models due to its strong performance as a mean estimator of the Gaussian noise model. Recently, various low-dose computed tomography (LDCT) image denoising methods using deep learning combined with the MSE loss have been developed; however, this approach has been observed to suffer from the regression-to-the-mean problem, leading to over-smoothed edges and degradation of texture in the image. 2 To overcome this issue, we propose a stochastic function in the loss function to improve the texture of the denoised CT images, rather than relying on complicated networks or feature space losses. The proposed loss function includes the MSE loss to learn the mean distribution and the Pearson divergence loss to learn feature textures. Specifically, the Pearson divergence loss is computed in an image space to measure the distance between two intensity measures of denoised low-dose and normal-dose CT images. The evaluation of the proposed model employs a novel approach of multi-metric quantitative analysis utilizing relative texture feature distance. Approach. Our experimental results show that the proposed Pearson divergence loss leads to a significant improvement in texture compared to the conventional MSE loss and generative adversarial network (GAN), both qualitatively and quantitatively. Results. Achieving consistent texture preservation in LDCT is a challenge in conventional GAN-type methods due to adversarial aspects aimed at minimizing noise while preserving texture. By incorporating the Pearson regularizer in the loss function, we can easily achieve a balance between two conflicting properties. Consistent high-quality CT images can significantly help clinicians in diagnoses and supporting researchers in the development of AI-diagnostic models.Significance. Achieving consistent texture preservation in LDCT is a challenge in conventional GAN-type methods due to adversarial aspects aimed at minimizing noise while preserving texture. By incorporating the Pearson regularizer in the loss function, we can easily achieve a balance between two conflicting properties. Consistent high-quality CT images can significantly help clinicians in diagnoses and supporting researchers in the development of AI-diagnostic models., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective . The primary objective of this study is to address the reconstruction time challenge in magnetic particle imaging (MPI) by introducing a novel approach named SNR-peak-based frequency selection (SPFS). The focus is on improving spatial resolution without compromising reconstruction speed, thereby enhancing the clinical potential of MPI for real-time imaging. Approach . To overcome the trade-off between reconstruction time and spatial resolution in MPI, the researchers propose SPFS as an innovative frequency selection method. Unlike conventional SNR-based selection, SPFS prioritizes frequencies with signal-to-noise ratio (SNR) peaks that capture crucial system matrix information. This adaptability to varying quantities of selected frequencies enhances versatility in the reconstruction process. The study compares the spatial resolution of MPI reconstruction using both SNR-based and SPFS frequency selection methods, utilizing simulated and real device data. Main results. The research findings demonstrate that the SPFS approach substantially improves image resolution in MPI, especially when dealing with a limited number of frequency components. By focusing on SNR peaks associated with critical system matrix information, SPFS mitigates the spatial resolution degradation observed in conventional SNR-based selection methods. The study validates the effectiveness of SPFS through the assessment of MPI reconstruction spatial resolution using both simulated and real device data, highlighting its potential to address a critical limitation in the field. Significance. The introduction of SPFS represents a significant breakthrough in MPI technology. The method not only accelerates reconstruction time but also enhances spatial resolution, thus expanding the clinical potential of MPI for various applications. The improved real-time imaging capabilities of MPI, facilitated by SPFS, hold promise for advancements in drug delivery, plaque assessment, tumor treatment, cerebral perfusion evaluation, immunotherapy guidance, and in vivo cell tracking., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective . To improve intravoxel incoherent motion imaging (IVIM) magnetic resonance Imaging quality using a new image denoising technique and model-independent parameterization of the signal versus b -value curve. Approach . IVIM images were acquired for 13 head-and-neck patients prior to radiotherapy. Post-radiotherapy scans were also acquired for five of these patients. Images were denoised prior to parameter fitting using neural blind deconvolution, a method of solving the ill-posed mathematical problem of blind deconvolution using neural networks. The signal decay curve was then quantified in terms of several area under the curve (AUC) parameters. Improvements in image quality were assessed using blind image quality metrics, total variation (TV), and the correlations between parameter changes in parotid glands with radiotherapy dose levels. The validity of blur kernel predictions was assessed by the testing the method's ability to recover artificial 'pseudokernels'. AUC parameters were compared with monoexponential, biexponential, and triexponential model parameters in terms of their correlations with dose, contrast-to-noise (CNR) around parotid glands, and relative importance via principal component analysis. Main results . Image denoising improved blind image quality metrics, smoothed the signal versus b -value curve, and strengthened correlations between IVIM parameters and dose levels. Image TV was reduced and parameter CNRs generally increased following denoising. AUC parameters were more correlated with dose and had higher relative importance than exponential model parameters. Significance . IVIM parameters have high variability in the literature and perfusion-related parameters are difficult to interpret. Describing the signal versus b -value curve with model-independent parameters like the AUC and preprocessing images with denoising techniques could potentially benefit IVIM image parameterization in terms of reproducibility and functional utility., (Creative Commons Attribution license.)

Objective. Respiratory motion, cardiac motion and inherently low signal-to-noise ratio (SNR) are major limitations of in vivo cardiac diffusion tensor imaging (DTI). We propose a novel enhancement method that uses unsupervised learning based invertible wavelet scattering (IWS) to improve the quality of in vivo cardiac DTI. Approach. Our method starts by extracting nearly transformation-invariant features from multiple cardiac diffusion-weighted (DW) image acquisitions using multi-scale wavelet scattering (WS). Then, the relationship between the WS coefficients and DW images is learned through a multi-scale encoder and a decoder network. Using the trained encoder, the deep features of WS coefficients of multiple DW image acquisitions are further extracted and then fused using an average rule. Finally, using the fused WS features and trained decoder, the enhanced DW images are derived. Main result. We evaluate the performance of the proposed method by comparing it with several methods on three in vivo cardiac DTI datasets in terms of SNR, contrast to noise ratio (CNR), fractional anisotropy (FA), mean diffusivity (MD) and helix angle (HA). Comparing against the best comparison method, SNR/CNR of diastolic, gastric peristalsis influenced, and end-systolic DW images were improved by 1%/16%, 5%/6%, and 56%/30%, respectively. The approach also yielded consistent FA and MD values and more coherent helical fiber structures than the comparison methods used in this work. Significance. The ablation results verify that using the transformation-invariant and noise-robust wavelet scattering features enables us to effectively explore the useful information from the limited data, providing a potential mean to alleviate the dependence of the fusion results on the number of repeated acquisitions, which is beneficial for dealing with the issues of noise and residual motion simultaneously and therefore improving the quality ofinvivocardiac DTI. Code can be found inhttps://github.com/strawberry1996/WS-MCNN., (© 2024 Institute of Physics and Engineering in Medicine. All rights, including for text and data mining, AI training, and similar technologies, are reserved.)

Objective. Deep learning has markedly enhanced the performance of sparse-view computed tomography reconstruction. However, the dependence of these methods on supervised training using high-quality paired datasets, and the necessity for retraining under varied physical acquisition conditions, constrain their generalizability across new imaging contexts and settings. Approach. To overcome these limitations, we propose an unsupervised approach grounded in the deep image prior framework. Our approach advances beyond the conventional single noise level input by incorporating multi-level linear diffusion noise, significantly mitigating the risk of overfitting. Furthermore, we embed non-local self-similarity as a deep implicit prior within a self-attention network structure, improving the model's capability to identify and utilize repetitive patterns throughout the image. Additionally, leveraging imaging physics, gradient backpropagation is performed between the image domain and projection data space to optimize network weights. Main Results. Evaluations with both simulated and clinical cases demonstrate our method's effective zero-shot adaptability across various projection views, highlighting its robustness and flexibility. Additionally, our approach effectively eliminates noise and streak artifacts while significantly restoring intricate image details. Significance . Our method aims to overcome the limitations in current supervised deep learning-based sparse-view CT reconstruction, offering improved generalizability and adaptability without the need for extensive paired training data., (Creative Commons Attribution license.)

Objective . Positron Emission Tomography and Magnetic Resonance Imaging (PET-MRI) systems can obtain functional and anatomical scans. But PET suffers from a low signal-to-noise ratio, while MRI are time-consuming. To address time-consuming, an effective strategy involves reducing k-space data collection, albeit at the cost of lowering image quality. This study aims to leverage the inherent complementarity within PET-MRI data to enhance the image quality of PET-MRI. Approach . A novel PET-MRI joint reconstruction model, termed MC-Diffusion, is proposed in the Bayesian framework. The joint reconstruction problem is transformed into a joint regularization problem, where data fidelity terms of PET and MRI are expressed independently. The regular term, the derivative of the logarithm of the joint probability distribution of PET and MRI, employs a joint score-based diffusion model for learning. The diffusion model involves the forward diffusion process and the reverse diffusion process. The forward diffusion process adds noise to transform a complex joint data distribution into a known joint prior distribution for PET and MRI simultaneously, resembling a denoiser. The reverse diffusion process removes noise using a denoiser to revert the joint prior distribution to the original joint data distribution, effectively utilizing joint probability distribution to describe the correlations of PET and MRI for improved quality of joint reconstruction. Main results . Qualitative and quantitative improvements are observed with the MC-Diffusion model. Comparative analysis against LPLS and Joint ISAT-net on the ADNI dataset demonstrates superior performance by exploiting complementary information between PET and MRI. The MC-Diffusion model effectively enhances the quality of PET and MRI images. Significance . This study employs the MC-Diffusion model to enhance the quality of PET-MRI images by integrating the fundamental principles of PET and MRI modalities and leveraging their inherent complementarity. Furthermore, utilizing the diffusion model to learn the joint probability distribution of PET and MRI, thereby elucidating their latent correlation, facilitates a more profound comprehension of the priors obtained through deep learning, contrasting with black-box prior or artificially constructed structural similarities., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective. Sparse-view dual-energy spectral computed tomography (DECT) imaging is a challenging inverse problem. Due to the incompleteness of the collected data, the presence of streak artifacts can result in the degradation of reconstructed spectral images. The subsequent material decomposition task in DECT can further lead to the amplification of artifacts and noise. Approach. To address this problem, we propose a novel one-step inverse generation network (OIGN) for sparse-view dual-energy CT imaging, which can achieve simultaneous imaging of spectral images and materials. The entire OIGN consists of five sub-networks that form four modules, including the pre-reconstruction module, the pre-decomposition module, and the following residual filtering module and residual decomposition module. The residual feedback mechanism is introduced to synchronize the optimization of spectral CT images and materials. Main results. Numerical simulation experiments show that the OIGN has better performance on both reconstruction and material decomposition than other state-of-the-art spectral CT imaging algorithms. OIGN also demonstrates high imaging efficiency by completing two high-quality imaging tasks in just 50 seconds. Additionally, anti-noise testing is conducted to evaluate the robustness of OIGN. Significance. These findings have great potential in high-quality multi-task spectral CT imaging in clinical diagnosis., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective . Photon-counting detector (PCD) CT enables routine virtual-monoenergetic image (VMI) reconstruction. We evaluated the performance of an automatic VMI energy level (keV) selection tool on a clinical PCD-CT system in comparison to an automatic tube potential (kV) selection tool from an energy-integrating-detector (EID) CT system from the same manufacturer. Approach. Four torso-shaped phantoms (20-50 cm width) containing iodine (2, 5, and 10 mg cc -1 ) and calcium (100 mg cc -1 ) were scanned on PCD-CT and EID-CT. Dose optimization techniques, task-based VMI energy level and tube-potential selection on PCD-CT (CARE keV) and task-based tube potential selection on EID-CT (CARE kV), were enabled. CT numbers, image noise, and dose-normalized contrast-to-noise ratio (CNR d ) were compared. Main results . PCD-CT produced task-specific VMIs at 70, 65, 60, and 55 keV for non-contrast, bone, soft tissue with contrast, and vascular settings, respectively. A 120 kV tube potential was automatically selected on PCD-CT for all scans. In comparison, EID-CT used x-ray tube potentials from 80 to 150 kV based on imaging task and phantom size. PCD-CT achieved consistent dose reduction at 9%, 21% and 39% for bone, soft tissue with contrast, and vascular tasks relative to the non-contrast task, independent of phantom size. On EID-CT, dose reduction factor for contrast tasks relative to the non-contrast task ranged from a 65% decrease (vascular task, 70 kV, 20 cm phantom) to a 21% increase (soft tissue with contrast task, 150 kV, 50 cm phantom) due to size-specific tube potential adaptation. PCD-CT CNR d was equivalent to or higher than those of EID-CT for all tasks and phantom sizes, except for the vascular task with 20 cm phantom, where 70 kV EID-CT CNR d outperformed 55 keV PCD-CT images. Significance . PCD-CT produced more consistent CT numbers compared to EID-CT due to standardized VMI output, which greatly benefits standardization efforts and facilitates radiation dose reduction., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective . Deep learning reconstruction (DLR) algorithms exhibit object-dependent resolution and noise performance. Thus, traditional geometric CT phantoms cannot fully capture the clinical imaging performance of DLR. This study uses a patient-derived 3D-printed PixelPrint lung phantom to evaluate a commercial DLR algorithm across a wide range of radiation dose levels. Method . The lung phantom used in this study is based on a patient chest CT scan containing ground glass opacities and was fabricated using PixelPrint 3D-printing technology. The phantom was placed inside two different size extension rings to mimic a small- and medium-sized patient and was scanned on a conventional CT scanner at exposures between 0.5 and 20 mGy. Each scan was reconstructed using filtered back projection (FBP), iterative reconstruction, and DLR at five levels of denoising. Image noise, contrast to noise ratio (CNR), root mean squared error, structural similarity index (SSIM), and multi-scale SSIM (MS SSIM) were calculated for each image. Results. DLR demonstrated superior performance compared to FBP and iterative reconstruction for all measured metrics in both phantom sizes, with better performance for more aggressive denoising levels. DLR was estimated to reduce dose by 25%-83% in the small phantom and by 50%-83% in the medium phantom without decreasing image quality for any of the metrics measured in this study. These dose reduction estimates are more conservative compared to the estimates obtained when only considering noise and CNR. Conclusion . DLR has the capability of producing diagnostic image quality at up to 83% lower radiation dose, which can improve the clinical utility and viability of lower dose CT scans. Furthermore, the PixelPrint phantom used in this study offers an improved testing environment with more realistic tissue structures compared to traditional CT phantoms, allowing for structure-based image quality evaluation beyond noise and contrast-based assessments., (Creative Commons Attribution license.)

Deep neural networks (DNNs) have been widely applied in medical image classification and achieve remarkable classification performance. These achievements heavily depend on large-scale accurately annotated training data. However, label noise is inevitably introduced in the medical image annotation, as the labeling process heavily relies on the expertise and experience of annotators. Meanwhile, DNNs suffer from overfitting noisy labels, degrading the performance of models. Therefore, in this work, we innovatively devise a noise-robust training approach to mitigate the adverse effects of noisy labels in medical image classification. Specifically, we incorporate contrastive learning and intra-group mixup attention strategies into vanilla supervised learning. The contrastive learning for feature extractor helps to enhance visual representation of DNNs. The intra-group mixup attention module constructs groups and assigns self-attention weights for group-wise samples, and subsequently interpolates massive noisy-suppressed samples through weighted mixup operation. We conduct comparative experiments on both synthetic and real-world noisy medical datasets under various noise levels. Rigorous experiments validate that our noise-robust method with contrastive learning and mixup attention can effectively handle with label noise, and is superior to state-of-the-art methods. An ablation study also shows that both components contribute to boost model performance. The proposed method demonstrates its capability of curb label noise and has certain potential toward real-world clinic applications., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective . The textures and detailed structures in computed tomography (CT) images are highly desirable for clinical diagnosis. This study aims to expand the current body of work on textures and details preserving convolutional neural networks for low-dose CT (LDCT) image denoising task. Approach . This study proposed a novel multi-scale feature aggregation and fusion network (MFAF-net) for LDCT image denoising. Specifically, we proposed a multi-scale residual feature aggregation module to characterize multi-scale structural information in CT images, which captures regional-specific inter-scale variations using learned weights. We further proposed a cross-level feature fusion module to integrate cross-level features, which adaptively weights the contributions of features from encoder to decoder by using a spatial pyramid attention mechanism. Moreover, we proposed a self-supervised multi-level perceptual loss module to generate multi-level auxiliary perceptual supervision for recovery of salient textures and structures of tissues and lesions in CT images, which takes advantage of abundant semantic information at various levels. We introduced parameters for the perceptual loss to adaptively weight the contributions of auxiliary features of different levels and we also introduced an automatic parameter tuning strategy for these parameters. Main results . Extensive experimental studies were performed to validate the effectiveness of the proposed method. Experimental results demonstrate that the proposed method can achieve better performance on both fine textures preservation and noise suppression for CT image denoising task compared with other competitive convolutional neural network (CNN) based methods. Significance . The proposed MFAF-net takes advantage of multi-scale receptive fields, cross-level features integration and self-supervised multi-level perceptual loss, enabling more effective recovering of fine textures and detailed structures of tissues and lesions in CT images., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective. Multi-contrast magnetic resonance imaging (MC MRI) can obtain more comprehensive anatomical information of the same scanning object but requires a longer acquisition time than single-contrast MRI. To accelerate MC MRI speed, recent studies only collect partial k-space data of one modality (target contrast) to reconstruct the remaining non-sampled measurements using a deep learning-based model with the assistance of another fully sampled modality (reference contrast). However, MC MRI reconstruction mainly performs the image domain reconstruction with conventional CNN-based structures by full supervision. It ignores the prior information from reference contrast images in other sparse domains and requires fully sampled target contrast data. In addition, because of the limited receptive field, conventional CNN-based networks are difficult to build a high-quality non-local dependency. Approach. In the paper, we propose an Image-Wavelet domain ConvNeXt-based network (IWNeXt) for self-supervised MC MRI reconstruction. Firstly, INeXt and WNeXt based on ConvNeXt reconstruct undersampled target contrast data in the image domain and refine the initial reconstructed result in the wavelet domain respectively. To generate more tissue details in the refinement stage, reference contrast wavelet sub-bands are used as additional supplementary information for wavelet domain reconstruction. Then we design a novel attention ConvNeXt block for feature extraction, which can capture the non-local information of the MC image. Finally, the cross-domain consistency loss is designed for self-supervised learning. Especially, the frequency domain consistency loss deduces the non-sampled data, while the image and wavelet domain consistency loss retain more high-frequency information in the final reconstruction. Main results. Numerous experiments are conducted on the HCP dataset and the M4Raw dataset with different sampling trajectories. Compared with DuDoRNet, our model improves by 1.651 dB in the peak signal-to-noise ratio. Significance. IWNeXt is a potential cross-domain method that can enhance the accuracy of MC MRI reconstruction and reduce reliance on fully sampled target contrast images., (Creative Commons Attribution license.)

Objective. Self-supervised learning methods have been successfully applied for low-dose computed tomography (LDCT) denoising, with the advantage of not requiring labeled data. Conventional self-supervised methods operate only in the image domain, ignoring valuable priors in the sinogram domain. Recently proposed dual-domain methods address this limitation but encounter issues with blurring artifacts in the reconstructed image due to the inhomogeneous distribution of noise levels in low-dose sinograms. Approach. To tackle this challenge, this paper proposes SDBDNet, an end-to-end dual-domain self-supervised method for LDCT denoising. With the network designed based on the properties of inhomogeneous noise in low-dose sinograms and the principle of moderate sinogram-domain denoising, SDBDNet achieves effective denoising in dual domains without introducing blurring artifacts. Specifically, we split the sinogram into two subsets based on the positions of detector cells to generate paired training data with high similarity and independent noise. These sub-sinograms are then restored to their original size using 1D interpolation and learning-based correction. To achieve adaptive and moderate smoothing in the sinogram domain, we integrate Dropblock, a type of convolution layer with regularization, into SDBDNet, and set a weighted average between the denoised sinograms and their noisy counterparts, leading to a well-balanced dual-domain approach. Main results. Numerical experiments show that our method outperforms popular non-learning and self-supervised learning methods, demonstrating its effectiveness and superior performance. Significance. While introducing a novel high-performance dual-domain self-supervised LDCT denoising method, this paper also emphasizes and verifies the importance of appropriate sinogram-domain denoising in dual-domain methods, which might inspire future work., (© 2024 Institute of Physics and Engineering in Medicine.)

Objective . Contrast-free microvascular imaging is clinically valuable for the assessment of physiological status and the early diagnosis of diseases. Effective clutter filtering is essential for microvascular visualization without contrast enhancement. Singular value decomposition (SVD)-based spatiotemporal filter has been widely used to suppress clutter. However, clinical real-time imaging relies on short ensembles (dozens of frames), which limits the implementation of SVD filtering due to the large error of eigen-correlated estimations and high dependence on optimal threshold when used in such ensembles. Approach . To address the above challenges of imaging in short ensembles, two optimized filters of angular domain data are proposed in this paper: grouped angle SVD (GA-SVD) and angular-coherence-based higher-order SVD (AC-HOSVD). GA-SVD applies SVD to the concatenation of all angular data to improve clutter rejection performance in short ensembles, while AC-HOSVD applies HOSVD to the angular data tensor and utilizes angular coherence in addition to spatial and temporal features for filtering. Feasible threshold selection strategies in each feature space are provided. The clutter rejection performance of the proposed filters and SVD was evaluated with Doppler phantom and in vivo studies at different cases. Moreover, the robustness of the filters was explored under wrong singular value threshold estimation, and their computational complexity was studied. Main results . Qualitative and quantitative results indicated that GA-SVD and AC-HOSVD can effectively improve clutter rejection performance in short ensembles, especially AC-HOSVD. Notably, the proposed methods using 20 frames had similar image quality to SVD using 100 frames. In vivo studies showed that compared to SVD, GA-SVD increased the signal-to-noise-ratio (SNR) by 6.03 dB on average, and AC-HOSVD increased the SNR by 8.93 dB on average. Furthermore, AC-HOSVD remained better power Doppler image quality under non-optimal thresholds, followed by GA-SVD. Significance . The proposed filters can greatly enhance contrast-free microvascular visualization in short ensembles and have potential for different clinical translations due to the performance differences., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective . Magnetic particle imaging (MPI) shows potential for contributing to biomedical research and clinical practice. However, MPI images are effectively affected by noise in the signal as its reconstruction is an ill-posed inverse problem. Thus, effective reconstruction method is required to reduce the impact of the noise while mapping signals to MPI images. Traditional methods rely on the hand-crafted data-consistency (DC) term and regularization term based on spatial priors to achieve noise-reducing and reconstruction. While these methods alleviate the ill-posedness and reduce noise effects, they may be difficult to fully capture spatial features. Approach . In this study, we propose a deep neural network for end-to-end reconstruction (DERnet) in MPI that emulates the DC term and regularization term using the feature mapping subnetwork and post-processing subnetwork, respectively, but in a data-driven manner. By doing so, DERnet can better capture signal and spatial features without relying on hand-crafted priors and strategies, thereby effectively reducing noise interference and achieving superior reconstruction quality. Main results . Our data-driven method outperforms the state-of-the-art algorithms with an improvement of 0.9-8.8 dB in terms of peak signal-to-noise ratio under various noise levels. The result demonstrates the advantages of our approach in suppressing noise interference. Furthermore, DERnet can be employed for measured data reconstruction with improved fidelity and reduced noise. In conclusion, our proposed method offers performance benefits in reducing noise interference and enhancing reconstruction quality by effectively capturing signal and spatial features. Significance . DERnet is a promising candidate method to improve MPI reconstruction performance and facilitate its more in-depth biomedical application., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective. Various deep learning methods have recently been used for low dose CT (LDCT) denoising. Aggressive denoising may destroy the edge and fine anatomical structures of CT images. Therefore a key issue in LDCT denoising tasks is the difficulty of balancing noise/artifact suppression and edge/structure preservation. Approach. We proposed an LDCT denoising network based on the encoder-decoder structure, namely the Learnable PM diffusion coefficient and efficient attention network (PMA-Net). First, using the powerful feature modeling capability of partial differential equations, we constructed a multiple learnable edge module to generate precise edge information, incorporating the anisotropic image processing idea of Perona-Malik (PM) model into the neural network. Second, a multiscale reformative coordinate attention module was designed to extract multiscale information. Non-overlapping dilated convolution capturing abundant contextual content was combined with coordinate attention which could embed the spatial location information of important features into the channel attention map. Finally, we imposed additional constraints on the edge information using edge-enhanced multiscale perceptual loss to avoid structure loss and over-smoothing. Main results. Experiments are conducted on simulated and real datasets. The quantitative and qualitative results show that the proposed method has better performance in suppressing noise/artifacts and preserving edges/structures. Significance. This work proposes a novel edge feature extraction method that unfolds partial differential equation into neural networks, which contributes to the interpretability and clinical application value of neural network., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective. Addition of a denoising filter step in ultrasound localization microscopy (ULM) has been shown to effectively reduce the error localizations of microbubbles (MBs) and achieve resolution improvement for super-resolution ultrasound (SR-US) imaging. However, previous image-denoising methods (e.g. block-matching 3D, BM3D) requires long data processing times, making ULM only able to be processed offline. This work introduces a new way to reduce data processing time through deep learning. Approach. In this study, we propose deep learning (DL) denoising based on contrastive semi-supervised network (CS-Net). The neural network is mainly trained with simulated MBs data to extract MB signals from noise. And the performances of CS-Net denoising are evaluated in both in vitro flow phantom experiment and in vivo experiment of New Zealand rabbit tumor. Main results. For in vitro flow phantom experiment, the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of single microbubble image are 26.91 dB and 4.01 dB, repectively. For in vivo animal experiment , the SNR and CNR were 12.29 dB and 6.06 dB. In addition, single microvessel of 24 μ m and two microvessels separated by 46 μ m could be clearly displayed. Most importantly,, the CS-Net denoising speeds for in vitro and in vivo experiments were 0.041 s frame -1 and 0.062 s frame -1 , respectively. Significance. DL denoising based on CS-Net can improve the resolution of SR-US as well as reducing denoising time, thereby making further contributions to the clinical real-time imaging of ULM., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective. Unsupervised learning-based methods have been proven to be an effective way to improve the image quality of positron emission tomography (PET) images when a large dataset is not available. However, when the gap between the input image and the target PET image is large, direct unsupervised learning can be challenging and easily lead to reduced lesion detectability. We aim to develop a new unsupervised learning method to improve lesion detectability in patient studies. Approach. We applied the deep progressive learning strategy to bridge the gap between the input image and the target image. The one-step unsupervised learning is decomposed into two unsupervised learning steps. The input image of the first network is an anatomical image and the input image of the second network is a PET image with a low noise level. The output of the first network is also used as the prior image to generate the target image of the second network by iterative reconstruction method. Results. The performance of the proposed method was evaluated through the phantom and patient studies and compared with non-deep learning, supervised learning and unsupervised learning methods. The results showed that the proposed method was superior to non-deep learning and unsupervised methods, and was comparable to the supervised method. Significance. A progressive unsupervised learning method was proposed, which can improve image noise performance and lesion detectability., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective. Here, we propose a dynamic residual Kaczmarz (DRK) method as an improved reconstruction method for magnetic particle imaging (MPI) to achieve a better reconstruction quality from high-noise signals. Approach. Based on the Kaczmarz (KZ) method, we introduced a residual vector to select parts of the low-noise equations for reconstruction. In each iteration, a low-noise subset was formulated based on the residual vector. Thus, the reconstruction converged to an accurate result with less noise. Main Results. To evaluate the performance of the proposed method, it was compared with classical Kaczmarz-type methods and state-of-the-art regularization models. The numerical simulation results demonstrate that the DRK method can achieve better reconstruction quality than all other comparison methods at similar noise levels. It can acquire a signal-to-background ratio (SBR) that is five times higher than that of classical Kaczmarz-type methods at a 5 dB noise level. Furthermore, the DRK method can acquire up to 0.7 structural similarity (SSIM) indicators at a 5 dB noise level when combined with the non-negative fused Least absolute shrinkage and selection operator (LASSO) regularization model. In addition, a real experiment based on the OpenMPI data set validated that the proposed DRK method can be applied to real data and perform well. Significance. The experimental results demonstrate that the proposed DRK method can significantly improve the reconstruction quality of MPI when the signals contain high noise. It has the potential to be applied to MPI instruments that contain high signal noise, such as human-sized MPI instruments. It is beneficial for expanding the biomedical applications of MPI technology., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective. The goal of this study is to develop a robust semi-weakly supervised learning strategy for vessel segmentation in laser speckle contrast imaging (LSCI), addressing the challenges associated with the low signal-to-noise ratio, small vessel size, and irregular vascular aberration in diseased regions, while improving the performance and robustness of the segmentation method. Approach. For the training dataset, the healthy vascular images denoted as normal-vessel samples were manually labeled, while the diseased LSCI images involving tumor or embolism were denoted as abnormal-vessel samples and annotated as pseudo labels by the traditional semantic segmentation methods. In the training phase, the pseudo labels were constantly updated to improve the segmentation accuracy based on DeepLabv3+. Objective evaluation was conducted on the normal-vessel test set, while subjective evaluation was performed on the abnormal-vessel test set. Main results. The proposed method achieved an IOU of 0.8671, a Dice of 0.9288, and a mean relative percentage difference (mRPD) with supervised learning of 0.5% in the objective evaluation. In the subjective evaluation, our method significantly outperformed other methods in main vessel segmentation, tiny vessel segmentation, and blood vessel connection. Additionally, our method exhibited robustness when abnormal-vessel style noise was added to normal-vessel samples using a style translation network. Significance. The proposed semi-weakly supervised learning strategy demonstrates high efficiency and excellent robustness for vascular segmentation in LSCI, providing a potential tool for assessing the morphological and structural features of vessels in clinical applications., (© 2023 Institute of Physics and Engineering in Medicine.)

Objective. A phased or a curvilinear array produces ultrasound (US) images with a sector field of view, which inherently exhibits spatially-varying image resolution with inferior quality in the far zone and towards the two sides azimuthally. Sector US images with improved spatial resolutions are favorable for accurate quantitative analysis of large and dynamic organs, such as the heart. Therefore, this study aims to translate US images with spatially-varying resolution to ones with less spatially-varying resolution. CycleGAN has been a prominent choice for unpaired medical image translation; however, it neither guarantees structural consistency nor preserves backscattering patterns between input and generated images for unpaired US images. Approach. To circumvent this limitation, we propose a constrained CycleGAN (CCycleGAN), which directly performs US image generation with unpaired images acquired by different ultrasound array probes. In addition to conventional adversarial and cycle-consistency losses of CycleGAN, CCycleGAN introduces an identical loss and a correlation coefficient loss based on intrinsic US backscattered signal properties to constrain structural consistency and backscattering patterns, respectively. Instead of post-processed B-mode images, CCycleGAN uses envelope data directly obtained from beamformed radio-frequency signals without any other nonlinear postprocessing. Main results . In vitro phantom results demonstrate that CCycleGAN successfully generates images with improved spatial resolution as well as higher peak signal-to-noise ratio and structural similarity compared with benchmarks. Significance . CCycleGAN-generated US images of the in vivo human beating heart further facilitate higher quality heart wall motion estimation than benchmarks-generated ones, particularly in deep regions. The codes are available athttps://github.com/xfsun99/CCycleGAN-TF2., (Creative Commons Attribution license.)

Introduction: Cardiac [ 18 F]FDG-PET is widely used for viability testing in patients with chronic ischemic heart disease. Guidelines recommend injection of 200-350 MBq [ 18 F]FDG, however, a reduction of radiation exposure has become increasingly important, but might come at the cost of reduced diagnostic accuracy due to the increased noise in the images. We aimed to explore the use of a common deep learning (DL) network for noise reduction in low-dose PET images, and to validate its accuracy using the clinical quantitative metrics used to determine cardiac viability in patients with ischemic heart disease., Methods: We included 168 patients imaged with cardiac [ 18 F]FDG-PET/CT. We simulated a reduced dose by keeping counts at thresholds 1% and 10%. 3D U-net with five blocks was trained to de-noise full PET volumes (128 × 128 × 111). The low-dose and de-noised images were compared in Corridor4DM to the full-dose PET images. We used the default segmentation of the left ventricle to extract the quantitative metrics end-diastolic volume (EDV), end-systolic volume (ESV), and left ventricular ejection fraction (LVEF) from the gated images, and FDG defect extent from the static images., Results: Our de-noising models were able to recover the PET signal for both the static and gated images in either dose-reduction. For the 1% low-dose images, the error is most pronounced for EDV and ESV, where the average underestimation is 25%. No bias was observed using the proposed DL de-noising method. De-noising minimized the outliers found for the 1% and 10% low-dose measurements of LVEF and extent. Accuracy of differential diagnosis based on LVEF threshold was highly improved after de-noising., Conclusion: A significant dose reduction can be achieved for cardiac [ 18 F]FDG images used for viability testing in patients with ischemic heart disease without significant loss of diagnostic accuracy when using our DL model for noise reduction. Both 1% and 10% dose reductions are possible with clinically quantitative metrics comparable to that obtained with a full dose.

The quality of reconstructed dynamic PET images, as well as the statistical reliability of the estimated pharmacokinetic parameters is often compromised by high levels of statistical noise, particularly at the voxel level. Many denoising strategies have been proposed, both in the temporal and spatial domain, which substantially improve the signal to noise ratio of the reconstructed dynamic images. However, although most filtering approaches are fairly successful in reducing the spatio-temporal inter-voxel variability, they may also average out or completely eradicate the critically important temporal signature of a transient neurotransmitter activation response that may be present in a non-steady state dynamic PET study. In this work, we explore an approach towards temporal denoising of non-steady state dynamic PET images using an artificial neural network, which was trained to identify the temporal profile of a time-activity curve, while preserving any potential activation response. We evaluated the performance of a feed-forward perceptron neural network to improve the signal to noise ratio of dynamic [ 11 C]raclopride activation studies and compared it with the widely used highly constrained back projection (HYPR) filter. Results on both simulated Geant4 Application for Tomographic Emission data of a realistic rat brain phantom and experimental animal data of a freely moving animal study showed that the proposed neural network can efficiently improve the noise characteristics of dynamic data in the temporal domain, while it can lead to a more reliable estimation of voxel-wise activation response in target region. In addition, improvements in signal-to-noise ratio achieved by denoising the dynamic data using the proposed neural network led to improved accuracy and precision of the estimated model parameters of the lp-ntPET model, compared to the HYPR filter. The performance of the proposed denoising approach strongly depends on the amount of noise in the dynamic PET data, with higher noise leading to substantially higher variability in the estimated parameters of the activation response. Overall, the feed-forward network led to a similar performance as the HYPR filter in terms of spatial denoising, but led to notable improvements in terms of temporal denoising, which in turn improved the estimation activation parameters.

Many deep learning (DL)-based image restoration methods for low-dose CT (LDCT) problems directly employ the end-to-end networks on low-dose training data without considering dose differences. However, the radiation dose difference has a great impact on the ultimate results, and lower doses increase the difficulty of restoration. Moreover, there is increasing demand to design and estimate acceptable scanning doses for patients in clinical practice, necessitating dose-aware networks embedded with adaptive dose estimation. In this paper, we consider these dose differences of input LDCT images and propose an adaptive dose-aware network. First, considering a large dose distribution range for simulation convenience, we coarsely define five dose levels in advance as lowest, lower, mild, higher and highest radiation dose levels. Instead of directly building the end-to-end mapping function between LDCT images and high-dose CT counterparts, the dose level is primarily estimated in the first stage. In the second stage, the adaptively learned low-dose level is used to guide the image restoration process as the pattern of prior information through the channel feature transform. We conduct experiments on a simulated dataset based on original high dose parts of American Association of Physicists in Medicine challenge datasets from the Mayo Clinic. Ablation studies validate the effectiveness of the dose-level estimation, and the experimental results show that our method is superior to several other DL-based methods. Specifically, our method provides obviously better performance in terms of the peak signal-to-noise ratio and visual quality reflected in subjective scores. Due to the dual-stage process, our method may suffer limitations under more parameters and coarse dose-level definitions, and thus, further improvements in clinical practical applications with different CT equipment vendors are planned in future work.

Purpose: The main goal of this work is to improve the previously proposed patient-specific pixel-based dual-energy (PP-DE) algorithm by developing an adaptive anti-correlated noise reduction (ACNR) method, resulting in reduced image noise., Methods: Theoretical models of contrast-to-noise (CNR) and signal-to-noise (SNR) ratio were developed as functions of weighting factors for DE bone ω Bn or soft tissue ω ST cancellation. These analytical expressions describe CNR and SNR properties of dual-energy (DE) images, obtained with both simple log subtraction (SLS) and ACNR algorithms, and allow for a direct comparison between experimental and theoretical results. The theoretical models demonstrate the importance of ACNR weighting factor (ω A ) optimization leading to the maximization of the SNR of the final image. A step phantom was constructed, which consisted of overlapping slabs of solid water (0-30 cm) and bone-mimicking material (0-6 cm), resulting in a total of 7 × 7 regions. High-energy (HE) and low-energy (LE) images were acquired at 140 kVp and 60 kVp with a clinical ExacTrac imaging system. The CNR and SNR were obtained for the DE images as functions of ω Bn,ST and noise reduction weighting factor ω A for different combinations of thicknesses. Weighting factors for bone cancellation were optimized for each region of interest (ROI) by finding zeros of CNR function for DE images between soft tissue only and soft tissue plus bone regions (and vice versa for soft tissue cancellation). The weighting factor for the ACNR algorithm ω A was then optimized by maximizing the SNR function for each ROI. HE and LE images for an anthropomorphic Rando phantom were obtained with the same acquisition parameters as for the step phantom. DE images for bone only and soft tissue only were obtained with three algorithms: SLS and PP-DE with conventional ACNR (uniform ω A ), and PP-DE with adaptive ACNR (region-varying ω A ). Weighting factor maps for PP-DE and adaptive ACNR methods were obtained for Rando phantom geometry (which was determined from its CT scans) by interpolation (or extrapolation) of weighting factors for the step phantom. CNR values were calculated for different regions., Results: The CNR and SNR characteristics as functions of material cancellation and noise reduction weighting factors were obtained from theoretical models and experimental data from the step phantom. This showed a good qualitative validation of the models. For the ANCR algorithm, both the theory and experiment demonstrated that the material cancellation weighting factors (ω Bn,ST ) can be optimized independently of the noise cancellation weighting factors (ω A ), which can be optimized by maximizing SNR. For each ROI (with different overlapping bone and soft tissue thicknesses) the weighting factors ω Bn,ST were determined as well as corresponding optimal weighting factors ω A for noise reduction. For the Rando phantom, CNR values for regions representing different anatomical structures (ribs, spine, and tumor) were evaluated. It was shown that the proposed adaptive ACNR further improves image quality, compared to the conventional ACNR algorithm. The improvement is maximized for regions with bones (ribs or spine), where the largest attenuation is observed., Conclusion: The ACNR weighting factors are dependent on the material thicknesses due to varying beam attenuation leading to different levels of quantum noise. This was shown with the derived theoretical expressions of the CNR and SNR functions and was validated by experimental data. The adaptive ANCR DE algorithm was developed, which allows for an increase in image quality by spatially varying weighting factors for noise reduction. This algorithm complements the previously developed PP-DE algorithm to obtain better quality DE images.

Reducing radiation dose of x-ray computed tomography (CT) and thereby decreasing the potential risk to patients are desirable in CT imaging. Deep neural network (DNN) has been proposed to reduce noise in low-dose CT (LdCT) images and showed promising results. However, most existing DNN-based methods require training a neural network using high-quality CT images as the reference. Lack of high-quality reference data has therefore been the bottleneck in the current DNN-based methods. Recently, a noise-to-noise (Noise2Noise) training method was proposed to train a denoising neural network with only noisy images. It has also been applied to LdCT data in both the count domain and image domain. However, the method still requires a separately acquired independent noisy reference image for supervising the training procedure. To address this limitation, we propose a novel method to generate both training inputs and training labels from the existing CT scans, which does not require any additional high-dose CT images or repeated scans. Therefore, existing large noisy dataset can be fully exploited for training a denoising neural network. Our experimental results show that the trained networks can reduce noise in existing CT image and hence improve the image quality for clinical diagnosis.

Objective . Ultrafast power Doppler (UPD) is an ultrasound method that can image blood flow at several thousands of frames per second. In particular, the high number of data provided by UPD enables the use of singular value decomposition (SVD) as a clutter filter for suppressing tissue signal. Notably, is has been demonstrated in various applications that SVD filtering increases significantly the sensitivity of UPD to microvascular flows. However, UPD is subjected to significant depth-dependent electronic noise and an optimal denoising approach is still being sought. Approach . In this study, we propose a new denoising method for UPD imaging: the Coherence Factor Mask (CFM). This filter is first based on filtering the ultrasound time-delayed data using SVD in the channel domain to remove clutter signal. Then, a spatiotemporal coherence mask that exploits coherence information between channels for identifying noisy pixels is computed. The mask is finally applied to beamformed images to decrease electronic noise before forming the power Doppler image. We describe theoretically how to filter channel data using a single SVD. Then, we evaluate the efficiency of the CFM filter for denoising in vitro and in vivo images and compare its performances with standard UPD and with three existing denoising approaches. Main results . The CFM filter gives gains in signal-to-noise ratio and contrast-to-noise ratio of up to 22 dB and 20 dB, respectively, compared to standard UPD and globally outperforms existing methods for reducing electronic noise. Furthermore, the CFM filter has the advantage over existing approaches of being adaptive and highly efficient while not requiring a cut-off for discriminating noise and blood signals nor for determining an optimal coherence lag. Significance . The CFM filter has the potential to help establish UPD as a powerful modality for imaging microvascular flows., (© 2023 Institute of Physics and Engineering in Medicine.)

Partly due to the use of exhaustive-annotated data, deep networks have achieved impressive performance on medical image segmentation. Medical imaging data paired with noisy annotation are, however, ubiquitous, but little is known about the effect of noisy annotation on deep learning based medical image segmentation. We studied the effect of noisy annotation in the context of mandible segmentation from CT images. First, 202 images of head and neck cancer patients were collected from our clinical database, where the organs-at-risk were annotated by one of twelve planning dosimetrists. The mandibles were roughly annotated as the planning avoiding structure. Then, mandible labels were checked and corrected by a head and neck specialist to get the reference standard. At last, by varying the ratios of noisy labels in the training set, deep networks were trained and tested for mandible segmentation. The trained models were further tested on other two public datasets. Experimental results indicated that the network trained with noisy labels had worse segmentation than that trained with reference standard, and in general, fewer noisy labels led to better performance. When using 20% or less noisy cases for training, no significant difference was found on the segmentation results between the models trained by noisy or reference annotation. Cross-dataset validation results verified that the models trained with noisy data achieved competitive performance to that trained with reference standard. This study suggests that the involved network is robust to noisy annotation to some extent in mandible segmentation from CT images. It also highlights the importance of labeling quality in deep learning. In the future work, extra attention should be paid to how to utilize a small number of reference standard samples to improve the performance of deep learning with noisy annotation.

Cone-beam x-ray luminescence computed tomography (CB-XLCT) has been proposed as a promising imaging tool, which enables three-dimensional imaging of the distribution of nanophosphors (NPs) in small animals. However, the reconstruction performance is usually unsatisfactory in terms of spatial resolution due to the ill-posedness of the CB-XLCT inverse problem. To alleviate this problem and to achieve high spatial resolution, a reconstruction method consisting of inner and outer iterations based on a restarted strategy is proposed. In this method, the primal-dual Newton conjugate gradient method (pdNCG) is adopted in the inner iterations to get fast reconstruction, which is used for resetting the permission region and increasing the convergence speed of the outer iteration. To assess the performance of the method, both numerical simulation and physical phantom experiments were conducted with a CB-XLCT system. The results demonstrate that compared with conventional reconstruction methods, the proposed re-pdNCG method can accurately and efficiently resolve the adjacent NPs with the least relative error.

The clinical value of multiple b-value diffusion-weighted (DW) magnetic resonance imaging (MRI) has been shown in many studies. However, DW-MRI often suffers from low signal-to-noise ratio, especially at high b-values. To address this limitation, we present an image denoising method based on the concept of deep image prior (DIP). In this method, high-quality prior images obtained from the same patient were used as the network input, and all noisy DW images were used as the network output. Our aim is to denoise all b-value DW images simultaneously. By using early stopping, we expect the DIP-based model to learn the content of images instead of the noise. The performance of the proposed DIP method was evaluated using both simulated and real DW-MRI data. We simulated a digital phantom and generated noise-free DW-MRI data according to the intravoxel incoherent motion model. Different levels of Rician noise were then simulated. The proposed DIP method was compared with the image denoising method using local principal component analysis (LPCA). The simulation results show that the proposed DIP method outperforms the LPCA method in terms of mean-squared error and parameter estimation. The results of real DW-MRI data show that the proposed DIP method can improve the quality of IVIM parametric images. DIP is a feasible method for denoising multiple b-value DW-MRI data.

Accelerated MRI involves undersampling k-space, creating unwanted artifacts when reconstructing the data. While the strategy of incoherent k-space acquisition is proven for techniques such as compressed sensing, it may not be optimal for all techniques. This study compares the use of coherent low-resolution (coherent-LR) and incoherent undersampling phase-encoding for real-time 3D CNN image reconstruction. Data were acquired with our 3 T Philips Achieva system. A retrospective analysis was performed on six non-small cell lung cancer patients who received dynamic acquisitions consisting of 650 free breathing images using a bSSFP sequence. We retrospectively undersampled the data by 5x and 10x acceleration using the two phase-encoding schemes. A quantitative analysis was conducted evaluating the tumor segmentations from the CNN reconstructed data using the Dice coefficient (DC) and centroid displacement. The reconstruction noise was evaluated using the structural similarity index (SSIM). Furthermore, we qualitatively investigated the CNN reconstruction using prospectively undersampled data, where the fully sampled training data set is acquired separately from the accelerated undersampled data. The patient averaged DC, centroid displacement, and SSIM for the tumor segmentation at 5x and 10x was superior using coherent low-resolution undersampling. Furthermore, the patient-specific CNN can be trained in under 6 h and the reconstruction time was 54 ms per image. Both the incoherent and coherent-LR prospective CNN reconstructions yielded qualitatively acceptable images; however, the coherent-LR reconstruction appeared superior to the incoherent reconstruction. We have demonstrated that coherent-LR undersampling for real-time CNN image reconstruction performs quantitatively better for the retrospective case of lung tumor segmentation, and qualitatively better for the prospective case. The tumor segmentation mean DC increased for all six patients at 5x acceleration and the temporal (dynamic) variance of the segmentation was reduced. The reconstruction speed achieved for our current implementation was 54 ms, providing an acceptable frame rate for real-time on-the-fly MR imaging.

Spectral and grating-based differential phase-contrast (DPC) x-ray imaging are two emerging technologies that offer additional information compared with conventional attenuation-based x-ray imaging. In the case of spectral imaging, energy-resolved measurements allow the generation of material-specific images by exploiting differences in the energy-dependent attenuation. DPC imaging uses the phase shift that an x-ray wave exhibits when traversing an object as contrast generation mechanism. Recently, we have investigated the combination of these two imaging techniques (spectral DPC imaging) and demonstrated potential advantages compared with spectral imaging. In this work, we present a noise analysis framework that allows the prediction of (co-) variances and noise power spectra for all three imaging methods. Moreover, the optimum acquisition parameters for a particular imaging task can be determined. We use this framework for a performance comparison of all three imaging methods. The comparison is focused on (projected) electron density images since they can be calculated with all three imaging methods. Our study shows that spectral DPC imaging enables the calculation of electron density images with strongly reduced noise levels compared with the other two imaging methods for a large range of clinically relevant pixel sizes. In contrast to conventional DPC imaging, there are no long-range noise correlations for spectral DPC imaging. This means that excessive low frequency noise can be avoided. We confirm the analytical predictions by numerical simulations.

We investigated whether a combination of k-space undersampling and variable density averaging enhances image quality for low-SNR MRI acquisitions. We implemented 3D Cartesian k-space prospective undersampling with a variable number of averages for each k-space line. The performance of this compressed sensing with variable-density averaging (CS-VDA) method was evaluated in retrospective analysis of fully sampled phantom MRI measurements, as well as for prospectively accelerated in vivo 3D brain and knee MRI scans. Both phantom and in vivo results showed that acquisitions using the CS-VDA approach resulted in better image quality as compared to full sampling of k-space in the same scan time. Specifically, CS-VDA with a higher number of averages in the center of k-space resulted in the best image quality, apparent from increased anatomical detail with preserved soft-tissue contrast. This novel approach will facilitate improved image quality of inherently low SNR data, such as those with high-resolution or specific contrast-weightings with low SNR efficiency.

The standard imaging setup in x-ray fluorescence computed tomography detects the fluorescence emission at a right angle with respect to the axis of the excitation beam. In this paper we have studied how the detection angle affects the signal-to-noise ratio (S/N), which is a major factor influencing the low-contrast sensitivity of the imaging system. This is done for an imaging setup using a collimated detector and a pencil beam of excitation x-rays. An ideal detection process is simulated for a generalized imaging case with gold/platinum tracers and experimental measurements are performed using a diagnostic x-ray tube. For monochromatic excitation, the results indicate that order-of-magnitude improvements of the S/N can be achieved by optimizing the detection angle. The maximal S/N, when exciting with an energy just above the K-edge, is achieved for large detection angles, i.e. with the detector close to the source. The improvements also transfer to polychromatic excitation sources and the experimental results show up to four-fold improvements of the S/N when changing the detection angle from 90° to 150°. Also, the changes of the S/N behavior when switching the fluorescent tracer is briefly demonstrated. These results suggest that the choice of detection angle should be taken seriously in the design of future XFCT imaging systems.

Objective. Deep neural network (DNN) based methods have shown promising performances for low-dose computed tomography (LDCT) imaging. However, most of the DNN-based methods are trained on simulated labeled datasets, and the low-dose simulation algorithms are usually designed based on simple statistical models which deviate from the real clinical scenarios, which could lead to issues of overfitting, instability and poor robustness. To address these issues, in this work, we present a structure-preserved meta-learning uniting network (shorten as 'SMU-Net') to suppress noise-induced artifacts and preserve structure details in the unlabeled LDCT imaging task in real scenarios. Approach. Specifically, the presented SMU-Net contains two networks, i.e., teacher network and student network. The teacher network is trained on simulated labeled dataset and then helps the student network train with the unlabeled LDCT images via the meta-learning strategy. The student network is trained on real LDCT dataset with the pseudo-labels generated by the teacher network. Moreover, the student network adopts the Co-teaching strategy to improve the robustness of the presented SMU-Net. Main results. We validate the proposed SMU-Net method on three public datasets and one real low-dose dataset. The visual image results indicate that the proposed SMU-Net has superior performance on reducing noise-induced artifacts and preserving structure details. And the quantitative results exhibit that the presented SMU-Net method generally obtains the highest signal-to-noise ratio (PSNR), the highest structural similarity index measurement (SSIM), and the lowest root-mean-square error (RMSE) values or the lowest natural image quality evaluator (NIQE) scores. Significance. We propose a meta learning strategy to obtain high-quality CT images in the LDCT imaging task, which is designed to take advantage of unlabeled CT images to promote the reconstruction performance in the LDCT environments., (© 2022 Institute of Physics and Engineering in Medicine.)

Objective. To improve positron emission tomography (PET) image quality, we aim to generate images of quality comparable to standard scan duration images using short scan duration (1/8 and 1/16 standard scan duration) inputs and assess the generated standard scan duration images quantitative and qualitatively. Also, the effect of training dataset properties (i.e. body mass index (BMI)) on the performance of the model(s) will be explored. Approach. Whole-body PET scans of 42 patients (41 18 F-FDG and one 68 Ga-PSMA) scanned with standard radiotracer dosage were included in this study. One 18 F-FDG patient data was set aside and the remaining 40 patients were split into four subsets of 10 patients with different mean patient BMI. Multiple copies of a developed cycle-GAN network were trained on each subset to predict standard scan images using 1/8 and 1/16 short duration scans. Also, the models' performance was tested on a patient scanned with the 68 Ga-PSMA radiotracer. Quantitative performance was tested using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and normalized root mean squared error (NRMSE) metrics, and two nuclear medicine specialists analyzed images qualitatively. Main results. The developed cycle-GAN model improved the PSNR, SSIM, and NRMSE of the 1/8 and 1/16 short scan duration inputs both 18 F-FDG and 68 Ga-PSMA radiotracers. Although, quantitatively PSNR, SSIM, and NRMSE of the 1/16 scan duration level were improved more than 1/8 counterparts, however, the later were qualitatively more appealing. SUV mean and SUV max of the generated images were also indicative of the improvements. The cycle-GAN model was much more capable in terms of image quality improvements and speed than the NLM denoising method. All results proved statistically significant using the paired-sample T-Test statistical test ( p -value < 0.05). Significance. Our suggested approach based on cycle-GAN could improve image quality of the 1/8 and 1/16 short scan-duration inputs through noise reduction both quantitively (PSNR, SSIM, NRMSE, SUVmean, and SUVmax) and qualitatively (contrast, noise, and diagnostic capability) to the level comparable to the standard scan-duration counterparts. The cycle-GAN model(s) had a similar performance on the 68 Ga-PSMA to the 18 F-FDG images and could improve the images qualitatively and quantitatively but requires more extensive study. Overall, images predicted from 1/8 short scan-duration inputs had the upper hand compared with 1/16 short scan-duration inputs., (© 2022 Institute of Physics and Engineering in Medicine.)

Objective. Cone-Beam CT (CBCT) often results in severe image artifacts and inaccurate HU values, meaning poor quality CBCT images cannot be directly applied to dose calculation in radiotherapy. To overcome this, we propose a cycle-residual connection with a dilated convolution-consistent generative adversarial network (Cycle-RCDC-GAN). Approach. The cycle-consistent generative adversarial network (Cycle-GAN) was modified using a dilated convolution with different expansion rates to extract richer semantic features from input images. Thirty pelvic patients were used to investigate the effect of synthetic CT (sCT) from CBCT, and 55 head and neck patients were used to explore the generalizability of the model. Three generalizability experiments were performed and compared: the pelvis trained model was applied to the head and neck; the head and neck trained model was applied to the pelvis, and the two datasets were trained together. Main results. The mean absolute error (MAE), the root mean square error (RMSE), peak signal to noise ratio (PSNR), the structural similarity index (SSIM), and spatial nonuniformity (SNU) assessed the quality of the sCT generated from CBCT. Compared with CBCT images, the MAE improved from 28.81 to 18.48, RMSE from 85.66 to 69.50, SNU from 0.34 to 0.30, and PSNR from 31.61 to 33.07, while SSIM improved from 0.981 to 0.989. The sCT objective indicators of Cycle-RCDC-GAN were better than Cycle-GAN's. The objective metrics for generalizability were also better than Cycle-GAN's. Significance. Cycle-RCDC-GAN enhances CBCT image quality and has better generalizability than Cycle-GAN, which further promotes the application of CBCT in radiotherapy., (© 2022 Institute of Physics and Engineering in Medicine.)

Objective. Medical image registration aims to find the deformation field that can align two images in a spatial position. A medical image registration method based on U-Net architecture has been proposed currently. However, U-Net architecture has few training parameters, which leads to weak learning ability, and it ignores the adverse effects of image noise on the registration accuracy. The article aims at addressing the problem of weak network learning ability and the adverse effects of noisy images on registration. Approach. Here we propose a novel unsupervised 3D brain image registration framework, which introduces the residual unit and singular value decomposition (SVD) denoising layer on the U-Net architecture. Residual unit solves the problem of network degradation, that is, registration accuracy becomes saturated and then degrades rapidly with the increase in network depth. SVD denoising layer uses the estimated model order for SVD-based low-rank image reconstruction. we use Akaike information criterion to estimate the appropriate model order, which is used to remove noise components. We use the exponential linear unit (ELU) as the activation function, which is more robust to noise than other peers. Main results. The proposed method is evaluated on the publicly available brain MRI datasets: Mindboggle101 and LPBA40. Experimental results demonstrate our method outperforms several state-of-the-art methods for the metric of Dice Score. The mean number of folding voxels and registration time are comparable to state-of-the-art methods. Significance. This study shows that Deep Residual-SVD Network can improve registration accuracy. This study also demonstrate that the residual unit can enhance the learning ability of the network, the SVD denoising layer can denoise effectively, and the ELU is more robust to noise., (© 2022 Institute of Physics and Engineering in Medicine.)

PET images often suffer poor signal-to-noise ratio (SNR). Our objective is to improve the SNR of PET images using a deep neural network (DNN) model and MRI images without requiring any higher SNR PET images in training. Our proposed DNN model consists of three modified U-Nets (3U-net). The PET training input data and targets were reconstructed using filtered-backprojection (FBP) and maximum likelihood expectation maximization (MLEM), respectively. FBP reconstruction was used because of its computational efficiency so that the trained network not only removes noise, but also accelerates image reconstruction. Digital brain phantoms downloaded from BrainWeb were used to evaluate the proposed method. Poisson noise was added into sinogram data to simulate a 6 min brain PET scan. Attenuation effect was included and corrected before the image reconstruction. Extra Poisson noise was introduced to the training inputs to improve the network denoising capability. Three independent experiments were conducted to examine the reproducibility. A lesion was inserted into testing data to evaluate the impact of mismatched MRI information using the contrast-to-noise ratio (CNR). The negative impact on noise reduction was also studied when miscoregistration between PET and MRI images occurs. Compared with 1U-net trained with only PET images, training with PET/MRI decreased the mean squared error (MSE) by 31.3% and 34.0% for 1U-net and 3U-net, respectively. The MSE reduction is equivalent to an increase in the count level by 2.5 folds and 2.9 folds for 1U-net and 3U-net, respectively. Compared with the MLEM images, the lesion CNR was improved 2.7 folds and 1.4 folds for 1U-net and 3U-net, respectively. The results show that the proposed method could improve the PET SNR without having higher SNR PET images.

Despite the utility of tumour characterisation using quantitative parameter maps from multi-b-value diffusion-weighted MRI (DWI), clinicians often prefer the use of the image with highest diffusion-weighting (b-value), for instance for defining regions of interest (ROIs). However, these images are typically degraded by noise, as they do not utilize the information from the full acquisition. We present a principal component analysis (PCA) approach for model-free denoising of DWI data. PCA-denoising was compared to synthetic MRI, where a diffusion model is fitted for each voxel and a denoised image at a given b-value is generated from the model fit. A quantitative comparison of systematic and random errors was performed on data simulated using several diffusion models (mono-exponential, bi-exponential, stretched-exponential and kurtosis). A qualitative visual comparison was also performed for in vivo images in six healthy volunteers and three pancreatic cancer patients. In simulations, the reduction in random errors from PCA-denoising was substantial (up to 55%) and similar to synthetic MRI (up to 53%). Model-based synthetic MRI denoising resulted in substantial (up to 29% of signal) systematic errors, whereas PCA-denoising was able to denoise without introducing systematic errors (less than 2%). In vivo, the signal-to-noise ratio (SNR) and sharpness of PCA-denoised images were superior to synthetic MRI, resulting in clearer tumour boundaries. In the presence of motion, PCA-denoising did not cause image blurring, unlike image averaging or synthetic MRI. Multi-b-value MRI can be denoised model-free with our PCA-denoising strategy that reduces noise to a level similar to synthetic MRI, but without introducing systematic errors associated with the synthetic MRI method.

The noise power spectrum (NPS) of a digital x-ray imaging device is usually estimated from the average of periodograms of regions of interest (ROIs) in images obtained with uniform radiation fields. In order to mitigate low frequency trends, present in the images and not arising from stochastic processes, detrending methods are applied to the images before being Fourier transformed. The most common of these methods subtracts a second-order polynomial fit from the image. In this work, it is shown that the characteristics of low frequency trends can deviate from the quadratic dependence on spatial coordinates. This results in large residual trends that give rise to important correlations in the detrended images and produce an inaccurate rise of the NPS calculations at low frequencies. A new detrending method of uniform images is presented. The method operates in the subbands of a wavelet transform, removing the low frequency contents of the uniform image. To do this, the approximation subband of the highest level of the wavelet transform is cancelled. The effect on the NPS calculations for three digital detectors is shown and the importance of the parameters of the wavelet transform is discussed. The main result states that the performance of the new method improves those of two polynomial detrending methods commonly used and is close to the performance of the subtraction of uniform exposure images method. Finally, guidelines for the implementation of the procedure, like the number of levels in the wavelet decomposition, are provided. As the number of levels in the wavelet transform increases, the removal of trends is restricted to lower frequencies. The selection of the number of levels should be guided by the shape of the autocorrelation function of the detrended image, which has to resemble the shape expected from the propagation of noise through the imaging chain.

Cone-beam CT (CBCT) systems commonly incorporate a flat-panel detector (FPD) with multiple-gain readout capability to reduce electronic noise and extend dynamic range. In this work, we report a penalized weighted least-squares (PWLS) method for CBCT image reconstruction with a system model that includes the electronic noise characteristics of FPDs, including systems with dynamic-gain or dual-gain (DG) readout in which the electronic noise is spatially varying. Statistical weights in PWLS were modified to account for the contribution of the electronic noise (algorithm denoted [Formula: see text]), and the method was combined with a certainty-based approach that improves the homogeneity of spatial resolution (algorithm denoted [Formula: see text]). The methods were tested in phantom studies designed to stress DG readout characteristics and translated to a clinical study for CBCT of patients with head traumas. The [Formula: see text] method demonstrated superior noise-resolution tradeoffs compared to filtered back-projection (FBP) and conventional PWLS. For example, with spatial resolution (edge-spread function width) matched at 0.65 mm, [Formula: see text] reduced variance by 28%-39% and 15%-25% compared to FBP and PWLS, respectively. The [Formula: see text] method achieved more homogeneous spatial resolution than [Formula: see text] while maintaining similar variance reduction. These findings were confirmed in clinical studies, which showed ~20% variance reduction in peripheral regions of the brain, potentially improving visual image quality in detection of epidural and/or subdural intracranial hemorrhage. The results are consistent with the general notion that incorporating a more accurate system model improves performance in optimization-based statistical CBCT reconstruction-in this case, a more accurate model for (spatially varying) electronic noise to improve detectability of low-contrast lesions.

We present a formalism for two-dimensional (2D) noise reconstruction in proton computed tomography (pCT). This is necessary for the application of fluence modulated pCT (FMpCT) since it permits image noise prescription and the corresponding proton fuence optimization. We aimed at extending previously published formalisms to account for the impact of multiple Coulomb scattering (MCS) on projection noise, and the use of filtered back projection (FBP) reconstruction along curved paths with distance driven binning (DDB). 2D noise reconstruction for a beam of protons with parallel initial momentum vectors, and for projections binned both at the rear tracker and with DDB, was established. Monte Carlo (MC) simulations of pCT scans of a water cylinder were employed to generate pCT projections and to calculate their noise for use in 2D noise reconstruction. These were compared to results from an analytical model accounting for MCS for rear tracker binning as well as against the previously published central pixel model which ignores MCS. Image noise reconstructed with the formalism for rear tracker binning and DDB were compared to MC results using annular regions of interest (ROIs). Agreement better than 8% was obtained between the noise of projections calculated with MC simulation and our model. Noise from annular ROIs agreed with our noise reconstructions for rear tracker binning and DDB. The central pixel model ignoring MCS underestimated projection and thus image noise by up to 40% towards the object's edge. The use of DDB decreased the image noise towards the object's edge when compared to rear tracker binning and yielded more uniform noise throughout the image. MCS should not be neglected when predicting image noise for pixels away from the center of an object in a pCT scan due to the increasing influence of the gradient of the object's hull closer to the edges.

Sensitivity is an important performance characteristic of positron emission tomography (PET) systems. Improved sensitivity can be used to reduce injected dose, reduce scan time, or improve the signal-to-noise ratio and temporal resolution for dynamic studies. One way to improve the sensitivity of PET scanners is to reduce the gaps between detector modules. In this paper, a new signal processing method, named the shared-photodetector readout method, is proposed and evaluated. In this method, the signals generated in nearest neighbor photodetectors adjacent to the detector module of interest, were used to help identify the interaction location in the detector module of interest. Using this method, scintillator array-based detector modules with almost 100% packing fraction can be built, and the edge crystals can be clearly resolved, even when the crystals are small compared to the photodetector size. To evaluate this signal processing concept in one dimension, a detector block with four dual-ended readout detector modules, was designed. The detector block consisted of eight 4  ×  4 arrays of SensL MicroFJ-30035 SiPMs coupled to both ends of a 14  ×  56 array of 0.9  ×  0.9  ×  20 mm 3 LYSO elements with a pitch size of 0.96 mm and a length of 20 mm. Performance in terms of energy resolution, flood histogram, timing resolution and depth-of-interaction resolution obtained using the shared-photodetector readout method were compared to those obtained using a conventional readout method. The results show that better over-all performance was achieved using the shared-photodetector readout method, especially at the edges and corners of the array.

Objective. The radiation dose of cerebral perfusion computed tomography (CPCT) imaging can be reduced by lowering the milliampere-second or kilovoltage peak. However, dose reduction can decrease image quality due to excessive x-ray quanta fluctuation and reduced detector signal relative to system electronic noise, thereby influencing the accuracy of hemodynamic parameters for patients with acute stroke. Existing low-dose CPCT denoising methods, which mainly focus on specific temporal and spatial prior knowledge in low-dose CPCT images, not take the noise distribution characteristics of low-dose CPCT images into consideration. In practice, the noise of low-dose CPCT images can be much more complicated. This study first investigates the noise properties in low-dose CPCT images and proposes a perfusion deconvolution model based on the noise properties. Approach. To characterize the noise distribution in CPCT images properly, we analyze noise properties in low-dose CPCT images and find that the intra-frame noise distribution may vary in the different areas and the inter-frame noise also may vary in low-dose CPCT images. Thus, we attempt the first-ever effort to model CPCT noise with a non-independent and identical distribution (i.i.d.) mixture-of-Gaussians (MoG) model for noise assumption. Furthermore, we integrate the noise modeling strategy into a perfusion deconvolution model and present a novel perfusion deconvolution method by using self-relative structural similarity information and MoG model (named as SR-MoG) to estimate the hemodynamic parameters accurately. In the presented SR-MoG method, the self-relative structural similarity information is obtained from preprocessed low-dose CPCT images. Main results. The results show that the presented SR-MoG method can achieve promising gains over the existing deconvolution approaches. In particular, the average root-mean-square error (RMSE) of cerebral blood flow (CBF), cerebral blood volume, and mean transit time was improved by 40.3%, 69.1%, and 40.8% in the digital phantom study, and the average RMSE of CBF can be improved by 81.0% in the clinical data study, compared with tensor total variation regularization deconvolution method. Significance. The presented SR-MoG method can estimate high-accuracy hemodynamic parameters andachieve promising gains over the existing deconvolution approaches., (© 2022 Institute of Physics and Engineering in Medicine.)