LEARN: Learned Experts' Assessment-Based Reconstruction Network for Sparse-Data CT

CAIR: Combining integrated attention with iterative optimization learning for sparse-view CT reconstruction

Sparse-view CT is an efficient way for low dose scanning but degrades image quality. Inspired by the successful use of non-local attention in natural image denoising and compression artifact removal, we proposed a network combining integrated attention and iterative optimization learning for sparse-view CT reconstruction (CAIR). Specifically, we first unrolled the proximal gradient descent into a deep network and added an enhanced initializer between the gradient term and the approximation term. It can enhance the information flow between different layers, fully preserve the image details, and improve the network convergence speed. Secondly, the integrated attention module was introduced into the reconstruction process as a regularization term. It adaptively fuses the local and non-local features of the image which are used to reconstruct the complex texture and repetitive details of the image, respectively. Note that we innovatively designed a one-shot iteration strategy to simplify the network structure and reduce the reconstruction time while maintaining image quality. Experiments showed that the proposed method is very robust and outperforms state-of-the-art methods in terms of both quantitative and qualitative, greatly improving the preservation of structures and the removal of artifacts.

Sinogram upsampling using Primal-Dual UNet for undersampled CT and radial MRI reconstruction

Computed tomography (CT) and magnetic resonance imaging (MRI) are two widely used clinical imaging modalities for non-invasive diagnosis. However, both of these modalities come with certain problems. CT uses harmful ionising radiation, and MRI suffers from slow acquisition speed. Both problems can be tackled by undersampling, such as sparse sampling. However, such undersampled data leads to lower resolution and introduces artefacts. Several techniques, including deep learning based methods, have been proposed to reconstruct such data. However, the undersampled reconstruction problem for these two modalities was always considered as two different problems and tackled separately by different research works. This paper proposes a unified solution for both sparse CT and undersampled radial MRI reconstruction, achieved by applying Fourier transform-based pre-processing on the radial MRI and then finally reconstructing both modalities using sinogram upsampling combined with filtered back-projection. The Primal-Dual network is a deep learning based method for reconstructing sparsely-sampled CT data. This paper introduces Primal-Dual UNet, which improves the Primal-Dual network in terms of accuracy and reconstruction speed. The proposed method resulted in an average SSIM of 0.932±0.021 while performing sparse CT reconstruction for fan-beam geometry with a sparsity level of 16, achieving a statistically significant improvement over the previous model, which resulted in 0.919±0.016. Furthermore, the proposed model resulted in 0.903±0.019 and 0.957±0.023 average SSIM while reconstructing undersampled brain and abdominal MRI data with an acceleration factor of 16, respectively - statistically significant improvements over the original model, which resulted in 0.867±0.025 and 0.949±0.025. Finally, this paper shows that the proposed network not only improves the overall image quality, but also improves the image quality for the regions-of-interest: liver, kidneys, and spleen; as well as generalises better than the baselines in presence the of a needle.

Low-Dose CT Image Synthesis for Domain Adaptation Imaging Using a Generative Adversarial Network With Noise Encoding Transfer Learning

Deep learning (DL) based image processing methods have been successfully applied to low-dose x-ray images based on the assumption that the feature distribution of the training data is consistent with that of the test data. However, low-dose computed tomography (LDCT) images from different commercial scanners may contain different amounts and types of image noise, violating this assumption. Moreover, in the application of DL based image processing methods to LDCT, the feature distributions of LDCT images from simulation and clinical CT examination can be quite different. Therefore, the network models trained with simulated image data or LDCT images from one specific scanner may not work well for another CT scanner and image processing task. To solve such domain adaptation problem, in this study, a novel generative adversarial network (GAN) with noise encoding transfer learning (NETL), or GAN-NETL, is proposed to generate a paired dataset with a different noise style. Specifically, we proposed a method to perform noise encoding operator and incorporate it into the generator to extract a noise style. Meanwhile, with a transfer learning (TL) approach, the image noise encoding operator transformed the noise type of the source domain to that of the target domain for realistic noise generation. One public and two private datasets are used to evaluate the proposed method. Experiment results demonstrated the feasibility and effectiveness of our proposed GAN-NETL model in LDCT image synthesis. In addition, we conduct additional image denoising study using the synthesized clinical LDCT data, which verified the merit of the proposed synthesis in improving the performance of the DL based LDCT processing method.

Need for objective task-based evaluation of deep learning-based denoising methods: A study in the context of myocardial perfusion SPECT

BACKGROUND

Artificial intelligence-based methods have generated substantial interest in nuclear medicine. An area of significant interest has been the use of deep-learning (DL)-based approaches for denoising images acquired with lower doses, shorter acquisition times, or both. Objective evaluation of these approaches is essential for clinical application.

PURPOSE

DL-based approaches for denoising nuclear-medicine images have typically been evaluated using fidelity-based figures of merit (FoMs) such as root mean squared error (RMSE) and structural similarity index measure (SSIM). However, these images are acquired for clinical tasks and thus should be evaluated based on their performance in these tasks. Our objectives were to: (1) investigate whether evaluation with these FoMs is consistent with objective clinical-task-based evaluation; (2) provide a theoretical analysis for determining the impact of denoising on signal-detection tasks; and (3) demonstrate the utility of virtual imaging trials (VITs) to evaluate DL-based methods.

METHODS

A VIT to evaluate a DL-based method for denoising myocardial perfusion SPECT (MPS) images was conducted. To conduct this evaluation study, we followed the recently published best practices for the evaluation of AI algorithms for nuclear medicine (the RELAINCE guidelines). An anthropomorphic patient population modeling clinically relevant variability was simulated. Projection data for this patient population at normal and low-dose count levels (20%, 15%, 10%, 5%) were generated using well-validated Monte Carlo-based simulations. The images were reconstructed using a 3-D ordered-subsets expectation maximization-based approach. Next, the low-dose images were denoised using a commonly used convolutional neural network-based approach. The impact of DL-based denoising was evaluated using both fidelity-based FoMs and area under the receiver operating characteristic curve (AUC), which quantified performance on the clinical task of detecting perfusion defects in MPS images as obtained using a model observer with anthropomorphic channels. We then provide a mathematical treatment to probe the impact of post-processing operations on signal-detection tasks and use this treatment to analyze the findings of this study.

RESULTS

Based on fidelity-based FoMs, denoising using the considered DL-based method led to significantly superior performance. However, based on ROC analysis, denoising did not improve, and in fact, often degraded detection-task performance. This discordance between fidelity-based FoMs and task-based evaluation was observed at all the low-dose levels and for different cardiac-defect types. Our theoretical analysis revealed that the major reason for this degraded performance was that the denoising method reduced the difference in the means of the reconstructed images and of the channel operator-extracted feature vectors between the defect-absent and defect-present cases.

CONCLUSIONS

The results show the discrepancy between the evaluation of DL-based methods with fidelity-based metrics versus the evaluation on clinical tasks. This motivates the need for objective task-based evaluation of DL-based denoising approaches. Further, this study shows how VITs provide a mechanism to conduct such evaluations computationally, in a time and resource-efficient setting, and avoid risks such as radiation dose to the patient. Finally, our theoretical treatment reveals insights into the reasons for the limited performance of the denoising approach and may be used to probe the effect of other post-processing operations on signal-detection tasks.

Noise-resilient deep learning for integrated circuit tomography

X-ray tomography is a non-destructive imaging technique that reveals the interior of an object from its projections at different angles. Under sparse-view and low-photon sampling, regularization priors are required to retrieve a high-fidelity reconstruction. Recently, deep learning has been used in X-ray tomography. The prior learned from training data replaces the general-purpose priors in iterative algorithms, achieving high-quality reconstructions with a neural network. Previous studies typically assume the noise statistics of test data are acquired a priori from training data, leaving the network susceptible to a change in the noise characteristics under practical imaging conditions. In this work, we propose a noise-resilient deep-reconstruction algorithm and apply it to integrated circuit tomography. By training the network with regularized reconstructions from a conventional algorithm, the learned prior shows strong noise resilience without the need for additional training with noisy examples, and allows us to obtain acceptable reconstructions with fewer photons in test data. The advantages of our framework may further enable low-photon tomographic imaging where long acquisition times limit the ability to acquire a large training set.

MDST: multi-domain sparse-view CT reconstruction based on convolution and swin transformer

Objective.Sparse-view computed tomography (SVCT), which can reduce the radiation doses administered to patients and hasten data acquisition, has become an area of particular interest to researchers. Most existing deep learning-based image reconstruction methods are based on convolutional neural networks (CNNs). Due to the locality of convolution and continuous sampling operations, existing approaches cannot fully model global context feature dependencies, which makes the CNN-based approaches less efficient in modeling the computed tomography (CT) images with various structural information.Approach.To overcome the above challenges, this paper develops a novel multi-domain optimization network based on convolution and swin transformer (MDST). MDST uses swin transformer block as the main building block in both projection (residual) domain and image (residual) domain sub-networks, which models global and local features of the projections and reconstructed images. MDST consists of two modules for initial reconstruction and residual-assisted reconstruction, respectively. The sparse sinogram is first expanded in the initial reconstruction module with a projection domain sub-network. Then, the sparse-view artifacts are effectively suppressed by an image domain sub-network. Finally, the residual assisted reconstruction module to correct the inconsistency of the initial reconstruction, further preserving image details.Main results. Extensive experiments on CT lymph node datasets and real walnut datasets show that MDST can effectively alleviate the loss of fine details caused by information attenuation and improve the reconstruction quality of medical images.Significance.MDST network is robust and can effectively reconstruct images with different noise level projections. Different from the current prevalent CNN-based networks, MDST uses transformer as the main backbone, which proves the potential of transformer in SVCT reconstruction.

MLF-IOSC: Multi-Level Fusion Network With Independent Operation Search Cell for Low-Dose CT Denoising

Computed tomography (CT) is widely used in clinical medicine, and low-dose CT (LDCT) has become popular to reduce potential patient harm during CT acquisition. However, LDCT aggravates the problem of noise and artifacts in CT images, increasing diagnosis difficulty. Through deep learning, denoising CT images by artificial neural network has aroused great interest for medical imaging and has been hugely successful. We propose a framework to achieve excellent LDCT noise reduction using independent operation search cells, inspired by neural architecture search, and introduce the Laplacian to further improve image quality. Employing patch-based training, the proposed method can effectively eliminate CT image noise while retaining the original structures and details, hence significantly improving diagnosis efficiency and promoting LDCT clinical applications.

EPRI sparse reconstruction method based on deep learning

Electron paramagnetic resonance imaging (EPRI) is an advanced tumor oxygen concentration imaging method. Now, the bottleneck problem of EPRI is that the scanning time is too long. Sparse reconstruction is an effective and fast imaging method, which means reconstructing images from sparse-view projections. However, the EPRI images sparsely reconstructed by the classic filtered back projection (FBP) algorithm often contain severe streak artifacts, which affect subsequent image processing. In this work, we propose a feature pyramid attention-based, residual, dense, deep convolutional network (FRD-Net) to suppress the streak artifacts in the FBP-reconstructed images. This network combines residual connection, attention mechanism, dense connections and introduces perceptual loss. The EPRI image with streak artifacts is used as the input of the network and the output-label is the corresponding high-quality image densely reconstructed by the FBP algorithm. After training, the FRD-Net gets the capability of suppressing streak artifacts. The real data reconstruction experiments show that the FRD-Net can better improve the sparse reconstruction accuracy, compared with three existing representative deep networks.

Physics-/Model-Based and Data-Driven Methods for Low-Dose Computed Tomography: A survey

Since 2016, deep learning (DL) has advanced tomographic imaging with remarkable successes, especially in low-dose computed tomography (LDCT) imaging. Despite being driven by big data, the LDCT denoising and pure end-to-end reconstruction networks often suffer from the black box nature and major issues such as instabilities, which is a major barrier to apply deep learning methods in low-dose CT applications. An emerging trend is to integrate imaging physics and model into deep networks, enabling a hybridization of physics/model-based and data-driven elements. In this paper, we systematically review the physics/model-based data-driven methods for LDCT, summarize the loss functions and training strategies, evaluate the performance of different methods, and discuss relevant issues and future directions.

Transfer learning framework for low-dose CT reconstruction based on marginal distribution adaptation in multiscale

BACKGROUND

With the increasing use of computed tomography (CT) in clinical practice, limiting CT radiation exposure to reduce potential cancer risks has become one of the important directions of medical imaging research. As the dose decreases, the reconstructed CT image will be severely degraded by projection noise.

PURPOSE

As an important method of image processing, supervised deep learning has been widely used in the restoration of low-dose CT (LDCT) in recent years. However, the normal-dose CT (NDCT) corresponding to a specific LDCT (it is regarded as the label of the LDCT, which is necessary for supervised learning) is very difficult to obtain so that the application of supervised learning methods in LDCT reconstruction is limited. It is necessary to construct a unsupervised deep learning framework for LDCT reconstruction that does not depend on paired LDCT-NDCT datasets.

METHODS

We presented an unsupervised learning framework for the transferring from the identity mapping to the low-dose reconstruction task, called marginal distribution adaptation in multiscale (MDAM). For NDCTs as source domain data, MDAM is an identity map with two parts: firstly, it establishes a dimensionality reduction mapping, which can obtain the same feature distribution from NDCTs and LDCTs; and then NDCTs is retrieved by reconstructing the image overview and details from the low-dimensional features. For the purpose of the feature transfer between source domain and target domain (LDCTs), we introduce the multiscale feature extraction in the MDAM, and then eliminate differences in probability distributions of these multiscale features between NDCTs and LDCTs through wavelet decomposition and domain adaptation learning.

RESULTS

Image quality evaluation metrics and subjective quality scores show that, as an unsupervised method, the performance of the MDAM approaches or even surpasses some state-of-the-art supervised methods. Especially, MDAM has been favorably evaluated in terms of noise suppression, structural preservation, and lesion detection.

CONCLUSIONS

We demonstrated that, the MDAM framework can reconstruct corresponding NDCTs from LDCTs with high accuracy, and without relying on any labeles. Moreover, it is more suitable for clinical application compared with supervised learning methods.

M3NAS: Multi-Scale and Multi-Level Memory-Efficient Neural Architecture Search for Low-Dose CT Denoising

Lowering the radiation dose in computed tomography (CT) can greatly reduce the potential risk to public health. However, the reconstructed images from dose-reduced CT or low-dose CT (LDCT) suffer from severe noise which compromises the subsequent diagnosis and analysis. Recently, convolutional neural networks have achieved promising results in removing noise from LDCT images. The network architectures that are used are either handcrafted or built on top of conventional networks such as ResNet and U-Net. Recent advances in neural network architecture search (NAS) have shown that the network architecture has a dramatic effect on the model performance. This indicates that current network architectures for LDCT may be suboptimal. Therefore, in this paper, we make the first attempt to apply NAS to LDCT and propose a multi-scale and multi-level memory-efficient NAS for LDCT denoising, termed M3NAS. On the one hand, the proposed M3NAS fuses features extracted by different scale cells to capture multi-scale image structural details. On the other hand, the proposed M3NAS can search a hybrid cell- and network-level structure for better performance. In addition, M3NAS can effectively reduce the number of model parameters and increase the speed of inference. Extensive experimental results on two different datasets demonstrate that the proposed M3NAS can achieve better performance and fewer parameters than several state-of-the-art methods. In addition, we also validate the effectiveness of the multi-scale and multi-level architecture for LDCT denoising, and present further analysis for different configurations of super-net.

Deep image and feature prior algorithm based on U-ConformerNet structure

PURPOSE

The reconstruction performance of the deep image prior (DIP) approach is limited by the conventional convolutional layer structure and it is difficult to enhance its potential. In order to improve the quality of image reconstruction and suppress artifacts, we propose a DIP algorithm with better performance, and verify its superiority in the latest case.

METHODS

We construct a new U-ConformerNet structure as the DIP algorithm's network, replacing the traditional convolutional layer-based U-net structure, and introduce the 'lpips' deep network based feature distance regularization method. Our algorithm can switch between supervised and unsupervised modes at will to meet different needs.

RESULTS

The reconstruction was performed on the low dose CT dataset (LoDoPaB). Our algorithm attained a PSNR of more than 35 dB under unsupervised conditions, and the PSNR under the supervised condition is greater than 36 dB. Both of which are better than the performance of the DIP-TV. Furthermore, the accuracy of this method is positively connected with the quality of the a priori image with the help of deep networks. In terms of noise eradication and artifact suppression, the DIP algorithm with U-ConformerNet structure outperforms the standard DIP method based on convolutional structure.

CONCLUSIONS

It is known by experimental verification that, in unsupervised mode, the algorithm improves the output PSNR by at least 2-3 dB when compared to the DIP-TV algorithm (proposed in 2020). In supervised mode, our algorithm approaches that of the state-of-the-art end-to-end deep learning algorithms.

A total variation prior unrolling approach for computed tomography reconstruction

BACKGROUND

With the rapid development of deep learning technology, deep neural networks can effectively enhance the performance of computed tomography (CT) reconstructions. One kind of commonly used method to construct CT reconstruction networks is to unroll the conventional iterative reconstruction (IR) methods to convolutional neural networks (CNNs). However, most unrolling methods primarily unroll the fidelity term of IR methods to CNNs, without unrolling the prior terms. The prior terms are always directly replaced by neural networks.

PURPOSE

In conventional IR methods, the prior terms play a vital role in improving the visual quality of reconstructed images. Unrolling the hand-crafted prior terms to CNNs may provide a more specialized unrolling approach to further improve the performance of CT reconstruction. In this work, a primal-dual network (PD-Net) was proposed by unrolling both the data fidelity term and the total variation (TV) prior term, which effectively preserves the image edges and textures in the reconstructed images.

METHODS

By further deriving the Chambolle-Pock (CP) algorithm instance for CT reconstruction, we discovered that the TV prior updates the reconstructed images with its divergences in each iteration of the solution process. Based on this discovery, CNNs were applied to yield the divergences of the feature maps for the reconstructed image generated in each iteration. Additionally, a loss function was applied to the predicted divergences of the reconstructed image to guarantee that the CNNs' results were the divergences of the corresponding feature maps in the iteration. In this manner, the proposed CNNs seem to play the same roles in the PD-Net as the TV prior in the IR methods. Thus, the TV prior in the CP algorithm instance can be directly unrolled to CNNs.

RESULTS

The datasets from the Low-Dose CT Image and Projection Data and the Piglet dataset were employed to assess the effectiveness of our proposed PD-Net. Compared with conventional CT reconstruction methods, our proposed method effectively preserves the structural and textural information in reference to ground truth.

CONCLUSIONS

The experimental results show that our proposed PD-Net framework is feasible for the implementation of CT reconstruction tasks. Owing to the promising results yielded by our proposed neural network, this study is intended to inspire further development of unrolling approaches by enabling the direct unrolling of hand-crafted prior terms to CNNs.

Deep-Learning-Based Few-Angle Cardiac SPECT Reconstruction Using Transformer

Convolutional neural networks (CNNs) have been extremely successful in various medical imaging tasks. However, because the size of the convolutional kernel used in a CNN is much smaller than the image size, CNN has a strong spatial inductive bias and lacks a global understanding of the input images. Vision Transformer, a recently emerged network structure in computer vision, can potentially overcome the limitations of CNNs for image-reconstruction tasks. In this work, we proposed a slice-by-slice Transformer network (SSTrans-3D) to reconstruct cardiac SPECT images from 3D few-angle data. To be specific, the network reconstructs the whole 3D volume using a slice-by-slice scheme. By doing so, SSTrans-3D alleviates the memory burden required by 3D reconstructions using Transformer. The network can still obtain a global understanding of the image volume with the Transformer attention blocks. Lastly, already reconstructed slices are used as the input to the network so that SSTrans-3D can potentially obtain more informative features from these slices. Validated on porcine, phantom, and human studies acquired using a GE dedicated cardiac SPECT scanner, the proposed method produced images with clearer heart cavity, higher cardiac defect contrast, and more accurate quantitative measurements on the testing data as compared with a deep U-net.

Domain-adaptive denoising network for low-dose CT via noise estimation and transfer learning

BACKGROUND

In recent years, low-dose computed tomography (LDCT) has played an important role in the diagnosis CT to reduce the potential adverse effects of X-ray radiation on patients, while maintaining the same diagnostic image quality.

PURPOSE

Deep learning (DL)-based methods have played an increasingly important role in the field of LDCT imaging. However, its performance is highly dependent on the consistency of feature distributions between training data and test data. Due to patient's breathing movements during data acquisition, the paired LDCT and normal dose CT images are difficult to obtain from realistic imaging scenarios. Moreover, LDCT images from simulation or clinical CT examination often have different feature distributions due to the pollution by different amounts and types of image noises. If a network model trained with a simulated dataset is used to directly test clinical patients' LDCT data, its denoising performance may be degraded. Based on this, we propose a novel domain-adaptive denoising network (DADN) via noise estimation and transfer learning to resolve the out-of-distribution problem in LDCT imaging.

METHODS

To overcome the previous adaptation issue, a novel network model consisting of a reconstruction network and a noise estimation network was designed. The noise estimation network based on a double branch structure is used for image noise extraction and adaptation. Meanwhile, the U-Net-based reconstruction network uses several spatially adaptive normalization modules to fuse multi-scale noise input. Moreover, to facilitate the adaptation of the proposed DADN network to new imaging scenarios, we set a two-stage network training plan. In the first stage, the public simulated dataset is used for training. In the second transfer training stage, we will continue to fine-tune the network model with a torso phantom dataset, while some parameters are frozen. The main reason using the two-stage training scheme is based on the fact that the feature distribution of image content from the public dataset is complex and diverse, whereas the feature distribution of noise pattern from the torso phantom dataset is closer to realistic imaging scenarios.

RESULTS

In an evaluation study, the trained DADN model is applied to both the public and clinical patient LDCT datasets. Through the comparison of visual inspection and quantitative results, it is shown that the proposed DADN network model can perform well in terms of noise and artifact suppression, while effectively preserving image contrast and details.

CONCLUSIONS

In this paper, we have proposed a new DL network to overcome the domain adaptation problem in LDCT image denoising. Moreover, the results demonstrate the feasibility and effectiveness of the application of our proposed DADN network model as a new DL-based LDCT image denoising method.

TIME-Net: Transformer-Integrated Multi-Encoder Network for limited-angle artifact removal in dual-energy CBCT

Dual-energy cone-beam computed tomography (DE-CBCT) is a promising imaging technique with foreseeable clinical applications. DE-CBCT images acquired with two different spectra can provide material-specific information. Meanwhile, the anatomical consistency and energy-domain correlation result in significant information redundancy, which could be exploited to improve image quality. In this context, this paper develops the Transformer-Integrated Multi-Encoder Network (TIME-Net) for DE-CBCT to remove the limited-angle artifacts. TIME-Net comprises three encoders (image encoder, prior encoder, and transformer encoder), two decoders (low- and high-energy decoders), and one feature fusion module. Three encoders extract various features for image restoration. The feature fusion module compresses these features into more compact shared features and feeds them to the decoders. Two decoders perform differential learning for DE-CBCT images. By design, TIME-Net could obtain high-quality DE-CBCT images using two complementary quarter-scans, holding great potential to reduce radiation dose and shorten the acquisition time. Qualitative and quantitative analyses based on simulated data and real rat data have demonstrated the promising performance of TIME-Net in artifact removal, subtle structure restoration, and reconstruction accuracy preservation. Two clinical applications, virtual non-contrast (VNC) imaging and iodine quantification, have proved the potential utility of the DE-CBCT images provided by TIME-Net.

X-ray CT image denoising with MINF: A modularized iterative network framework for data from multiple dose levels

In clinical applications, multi-dose scan protocols will cause the noise levels of computed tomography (CT) images to fluctuate widely. The popular low-dose CT (LDCT) denoising network outputs denoised images through an end-to-end mapping between an LDCT image and its corresponding ground truth. The limitation of this method is that the reduced noise level of the image may not meet the diagnostic needs of doctors. To establish a denoising model adapted to the multi-noise levels robustness, we proposed a novel and efficient modularized iterative network framework (MINF) to learn the feature of the original LDCT and the outputs of the previous modules, which can be reused in each following module. The proposed network can achieve the goal of gradual denoising, outputting clinical images with different denoising levels, and providing the reviewing physicians with increased confidence in their diagnosis. Moreover, a multi-scale convolutional neural network (MCNN) module is designed to extract as much feature information as possible during the network's training. Extensive experiments on public and private clinical datasets were carried out, and comparisons with several state-of-the-art methods show that the proposed method can achieve satisfactory results for noise suppression of LDCT images. In further comparisons with modularized adaptive processing neural network (MAP-NN), the proposed network shows superior step-by-step or gradual denoising performance. Considering the high quality of gradual denoising results, the proposed method can obtain satisfactory performance in terms of image contrast and detail protection as the level of denoising increases, which shows its potential to be suitable for a multi-dose levels denoising task.

Structure-preserved meta-learning uniting network for improving low-dose CT quality

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.

Deep learning tomographic reconstruction through hierarchical decomposition of domain transforms

Deep learning (DL) has shown unprecedented performance for many image analysis and image enhancement tasks. Yet, solving large-scale inverse problems like tomographic reconstruction remains challenging for DL. These problems involve non-local and space-variant integral transforms between the input and output domains, for which no efficient neural network models are readily available. A prior attempt to solve tomographic reconstruction problems with supervised learning relied on a brute-force fully connected network and only allowed reconstruction with a 128⁴ system matrix size. This cannot practically scale to realistic data sizes such as 512⁴ and 512⁶ for three-dimensional datasets. Here we present a novel framework to solve such problems with DL by casting the original problem as a continuum of intermediate representations between the input and output domains. The original problem is broken down into a sequence of simpler transformations that can be well mapped onto an efficient hierarchical network architecture, with exponentially fewer parameters than a fully connected network would need. We applied the approach to computed tomography (CT) image reconstruction for a 512⁴ system matrix size. This work introduces a new kind of data-driven DL solver for full-size CT reconstruction without relying on the structure of direct (analytical) or iterative (numerical) inversion techniques. This work presents a feasibility demonstration of full-scale learnt reconstruction, whereas more developments will be needed to demonstrate superiority relative to traditional reconstruction approaches. The proposed approach is also extendable to other imaging problems such as emission and magnetic resonance reconstruction. More broadly, hierarchical DL opens the door to a new class of solvers for general inverse problems, which could potentially lead to improved signal-to-noise ratio, spatial resolution and computational efficiency in various areas.

Superiorization-inspired unrolled SART algorithm with U-Net generated perturbations for sparse-view and limited-angle CT reconstruction

Objective.Unrolled algorithms are a promising approach for reconstruction of CT images in challenging scenarios, such as low-dose, sparse-view and limited-angle imaging. In an unrolled algorithm, a fixed number of iterations of a reconstruction method are unrolled into multiple layers of a neural network, and interspersed with trainable layers. The entire network is then trained end-to-end in a supervised fashion, to learn an appropriate regularizer from training data. In this paper we propose a novel unrolled algorithm, and compare its performance with several other approaches on sparse-view and limited-angle CT.Approach.The proposed algorithm is inspired by the superiorization methodology, an optimization heuristic in which iterates of a feasibility-seeking method are perturbed between iterations, typically using descent directions of a model-based penalty function. Our algorithm instead uses a modified U-net architecture to introduce the perturbations, allowing a network to learn beneficial perturbations to the image at various stages of the reconstruction, based on the training data.Main Results.In several numerical experiments modeling sparse-view and limited angle CT scenarios, the algorithm provides excellent results. In particular, it outperforms several competing unrolled methods in limited-angle scenarios, while providing comparable or better performance on sparse-view scenarios.Significance.This work represents a first step towards exploiting the power of deep learning within the superiorization methodology. Additionally, it studies the effect of network architecture on the performance of unrolled methods, as well as the effectiveness of the unrolled approach on both limited-angle CT, where previous studies have primarily focused on the sparse-view and low-dose cases.

RAD-UNet: a Residual, Attention-Based, Dense UNet for CT Sparse Reconstruction

To suppress the streak artifacts in images reconstructed from sparse-view projections in computed tomography (CT), a residual, attention-based, dense UNet (RAD-UNet) deep network is proposed to achieve accurate sparse reconstruction. The filtered back projection (FBP) algorithm is used to reconstruct the CT image with streak artifacts from sparse-view projections. Then, the image is processed by the RAD-UNet to suppress streak artifacts and obtain high-quality CT image. Those images with streak artifacts are used as the input of the RAD-UNet, and the output-label images are the corresponding high-quality images. Through training via the large-scale training data, the RAD-UNet can obtain the capability of suppressing streak artifacts. This network combines residual connection, attention mechanism, dense connection and perceptual loss. This network can improve the nonlinear fitting capability and the performance of suppressing streak artifacts. The experimental results show that the RAD-UNet can improve the reconstruction accuracy compared with three existing representative deep networks. It may not only suppress streak artifacts but also better preserve image details. The proposed networks may be readily applied to other image processing tasks including image denoising, image deblurring, and image super-resolution.

Noise Characteristics Modeled Unsupervised Network for Robust CT Image Reconstruction

Deep learning (DL)-based methods show great potential in computed tomography (CT) imaging field. The DL-based reconstruction methods are usually evaluated on the training and testing datasets which are obtained from the same distribution, i.e., the same CT scan protocol (i.e., the region setting, kVp, mAs, etc.). In this work, we focus on analyzing the robustness of the DL-based methods against protocol-specific distribution shifts (i.e., the training and testing datasets are from different region settings, different kVp settings, or different mAs settings, respectively). The results show that the DL-based reconstruction methods are sensitive to the protocol-specific perturbations which can be attributed to the noise distribution shift between the training and testing datasets. Based on these findings, we presented a low-dose CT reconstruction method using an unsupervised strategy with the consideration of noise distribution to address the issue of protocol-specific perturbations. Specifically, unpaired sinogram data is enrolled into the network training, which represents unique information for specific imaging protocol, and a Gaussian mixture model (GMM) is introduced to characterize the noise distribution in CT images. It can be termed as GMM based unsupervised CT reconstruction network (GMM-unNet) method. Moreover, an expectation-maximization algorithm is designed to optimize the presented GMM-unNet method. Extensive experiments are performed on three datasets from different scan protocols, which demonstrate that the presented GMM-unNet method outperforms the competing methods both qualitatively and quantitatively.

A streak artifact reduction algorithm in sparse-view CT using a self-supervised neural representation

PURPOSE

Sparse-view computed tomography (CT) has been attracting attention for its reduced radiation dose and scanning time. However, analytical image reconstruction methods suffer from streak artifacts due to insufficient projection views. Recently, various deep learning-based methods have been developed to solve this ill-posed inverse problem. Despite their promising results, they are easily overfitted to the training data, showing limited generalizability to unseen systems and patients. In this work, we propose a novel streak artifact reduction algorithm that provides a system- and patient-specific solution.

METHODS

Motivated by the fact that streak artifacts are deterministic errors, we regenerate the same artifacts from a prior CT image under the same system geometry. This prior image need not be perfect but should contain patient-specific information and be consistent with full-view projection data for accurate regeneration of the artifacts. To this end, we use a coordinate-based neural representation that often causes image blur but can greatly suppress the streak artifacts while having multiview consistency. By employing techniques in neural radiance fields originally proposed for scene representations, the neural representation is optimized to the measured sparse-view projection data via self-supervised learning. Then, we subtract the regenerated artifacts from the analytically reconstructed original image to obtain the final corrected image.

RESULTS

To validate the proposed method, we used simulated data of extended cardiac-torso phantoms and the 2016 NIH-AAPM-Mayo Clinic Low-Dose CT Grand Challenge and experimental data of physical pediatric and head phantoms. The performance of the proposed method was compared with a total variation-based iterative reconstruction method, naive application of the neural representation, and a convolutional neural network-based method. In visual inspection, it was observed that the small anatomical features were best preserved by the proposed method. The proposed method also achieved the best scores in the visual information fidelity, modulation transfer function, and lung nodule segmentation.

CONCLUSIONS

The results on both simulated and experimental data suggest that the proposed method can effectively reduce the streak artifacts while preserving small anatomical structures that are easily blurred or replaced with misleading features by the existing methods. Since the proposed method does not require any additional training datasets, it would be useful in clinical practice where the large datasets cannot be collected.

Physics-informed neural network for phase imaging based on transport of intensity equation

Non-interferometric quantitative phase imaging based on Transport of Intensity Equation (TIE) has been widely used in bio-medical imaging. However, analytic TIE phase retrieval is prone to low-spatial frequency noise amplification, which is caused by the illposedness of inversion at the origin of the spectrum. There are also retrieval ambiguities resulting from the lack of sensitivity to the curl component of the Poynting vector occurring with strong absorption. Here, we establish a physics-informed neural network (PINN) to address these issues, by integrating the forward and inverse physics models into a cascaded deep neural network. We demonstrate that the proposed PINN is efficiently trained using a small set of sample data, enabling the conversion of noise-corrupted 2-shot TIE phase retrievals to high quality phase images under partially coherent LED illumination. The efficacy of the proposed approach is demonstrated by both simulation using a standard image database and experiment using human buccal epitehlial cells. In particular, high image quality (SSIM = 0.919) is achieved experimentally using a reduced size of labeled data (140 image pairs). We discuss the robustness of the proposed approach against insufficient training data, and demonstrate that the parallel architecture of PINN is efficient for transfer learning.

Learned regularization for image reconstruction in sparse-view photoacoustic tomography

Constrained data acquisitions, such as sparse view measurements, are sometimes used in photoacoustic computed tomography (PACT) to accelerate data acquisition. However, it is challenging to reconstruct high-quality images under such scenarios. Iterative image reconstruction with regularization is a typical choice to solve this problem but it suffers from image artifacts. In this paper, we present a learned regularization method to suppress image artifacts in model-based iterative reconstruction in sparse view PACT. A lightweight dual-path network is designed to learn regularization features from both the data and the image domains. The network is trained and tested on both simulation and in vivo datasets and compared with other methods such as Tikhonov regularization, total variation regularization, and a U-Net based post-processing approach. Results show that although the learned regularization network possesses a size of only 0.15% of a U-Net, it outperforms other methods and converges after as few as five iterations, which takes less than one-third of the time of conventional methods. Moreover, the proposed reconstruction method incorporates the physical model of photoacoustic imaging and explores structural information from training datasets. The integration of deep learning with a physical model can potentially achieve improved imaging performance in practice.

A dual-domain neural network based on sinogram synthesis for sparse-view CT reconstruction

OBJECTIVE

The dual-domain deep learning-based reconstruction techniques have enjoyed many successful applications in the field of medical image reconstruction. Applying the analytical reconstruction based operator to transfer the data from the projection domain to the image domain, the dual-domain techniques may suffer from the insufficient suppression or removal of streak artifacts in areas with the missing view data, when addressing the sparse-view reconstruction problems. In this work, to overcome this problem, an intelligent sinogram synthesis based back-projection network (iSSBP-Net) was proposed for sparse-view computed tomography (CT) reconstruction. In the iSSBP-Net method, a convolutional neural network (CNN) was involved in the dual-domain method to inpaint the missing view data in the sinogram before CT reconstruction.

METHODS

The proposed iSSBP-Net method fused a sinogram synthesis sub-network (SS-Net), a sinogram filter sub-network (SF-Net), a back-projection layer, and a post-CNN into an end-to-end network. Firstly, to inpaint the missing view data, the SS-Net employed a CNN to synthesize the full-view sinogram in the projection domain. Secondly, to improve the visual quality of the sparse-view CT images, the synthesized sinogram was filtered by a CNN. Thirdly, the filtered sinogram was brought into the image domain through the back-projection layer. Finally, to yield images of high visual sensitivity, the post-CNN was applied to restore the desired images from the outputs of the back-projection layer.

RESULTS

The numerical experiments demonstrate that the proposed iSSBP-Net is superior to all competing algorithms under different scanning condintions for sparse-view CT reconstruction. Compared to the competing algorithms, the proposed iSSBP-Net method improved the peak signal-to-noise ratio of the reconstructed images about 1.21 dB, 0.26 dB, 0.01 dB, and 0.37 dB under the scanning conditions of 360, 180, 90, and 60 views, respectively.

CONCLUSION

The promising reconstruction results indicate that involving the SS-Net in the dual-domain method is could be an effective manner to suppress or remove the streak artifacts in sparse-view CT images. Due to the promising results reconstructed by the iSSBP-Net method, this study is intended to inspire the further development of sparse-view CT reconstruction by involving a SS-Net in the dual-domain method.

3Din vivodose verification in prostate proton therapy with deep learning-based proton-acoustic imaging

Dose delivery uncertainty is a major concern in proton therapy, adversely affecting the treatment precision and outcome. Recently, a promising technique, proton-acoustic (PA) imaging, has been developed to provide real-timein vivo3D dose verification. However, its dosimetry accuracy is limited due to the limited-angle view of the ultrasound transducer. In this study, we developed a deep learning-based method to address the limited-view issue in the PA reconstruction. A deep cascaded convolutional neural network (DC-CNN) was proposed to reconstruct 3D high-quality radiation-induced pressures using PA signals detected by a matrix array, and then derive precise 3D dosimetry from pressures for dose verification in proton therapy. To validate its performance, we collected 81 prostate cancer patients' proton therapy treatment plans. Dose was calculated using the commercial software RayStation and was normalized to the maximum dose. The PA simulation was performed using the open-source k-wave package. A matrix ultrasound array with 64 × 64 sensors and 500 kHz central frequency was simulated near the perineum to acquire radiofrequency (RF) signals during dose delivery. For realistic acoustic simulations, tissue heterogeneity and attenuation were considered, and Gaussian white noise was added to the acquired RF signals. The proposed DC-CNN was trained on 204 samples from 69 patients and tested on 26 samples from 12 other patients. Predicted 3D pressures and dose maps were compared against the ground truth qualitatively and quantitatively using root-mean-squared-error (RMSE), gamma-index (GI), and dice coefficient of isodose lines. Results demonstrated that the proposed method considerably improved the limited-view PA image quality, reconstructing pressures with clear and accurate structures and deriving doses with a high agreement with the ground truth. Quantitatively, the pressure accuracy achieved an RMSE of 0.061, and the dose accuracy achieved an RMSE of 0.044, GI (3%/3 mm) of 93.71%, and 90%-isodose line dice of 0.922. The proposed method demonstrates the feasibility of achieving high-quality quantitative 3D dosimetry in PA imaging using a matrix array, which potentially enables the online 3D dose verification for prostate proton therapy.

Iterator-Net: sinogram-based CT image reconstruction

Image reconstruction is extremely important for computed tomography (CT) imaging, so it is significant to be continuously improved. The unfolding dynamics method combines a deep learning model with a traditional iterative algorithm. It is interpretable and has a fast reconstruction speed, but the essence of the algorithm is to replace the approximation operator in the optimization objective with a learning operator in the form of a convolutional neural network. In this paper, we firstly design a new iterator network (iNet), which is based on the universal approximation theorem and tries to simulate the functional relationship between the former and the latter in the maximum-likelihood expectation maximization (MLEM) algorithm. To evaluate the effectiveness of the method, we conduct experiments on a CT dataset, and the results show that our iNet method improves the quality of reconstructed images.

A review on the application of deep learning for CT reconstruction, bone segmentation and surgical planning in oral and maxillofacial surgery

Computer-assisted surgery (CAS) allows clinicians to personalize treatments and surgical interventions and has therefore become an increasingly popular treatment modality in maxillofacial surgery. The current maxillofacial CAS consists of three main steps: (1) CT image reconstruction, (2) bone segmentation, and (3) surgical planning. However, each of these three steps can introduce errors that can heavily affect the treatment outcome. As a consequence, tedious and time-consuming manual post-processing is often necessary to ensure that each step is performed adequately. One way to overcome this issue is by developing and implementing neural networks (NNs) within the maxillofacial CAS workflow. These learning algorithms can be trained to perform specific tasks without the need for explicitly defined rules. In recent years, an extremely large number of novel NN approaches have been proposed for a wide variety of applications, which makes it a difficult task to keep up with all relevant developments. This study therefore aimed to summarize and review all relevant NN approaches applied for CT image reconstruction, bone segmentation, and surgical planning. After full text screening, 76 publications were identified: 32 focusing on CT image reconstruction, 33 focusing on bone segmentation and 11 focusing on surgical planning. Generally, convolutional NNs were most widely used in the identified studies, although the multilayer perceptron was most commonly applied in surgical planning tasks. Moreover, the drawbacks of current approaches and promising research avenues are discussed.

Generation of Digital Brain Phantom for Machine Learning Application of Dopamine Transporter Radionuclide Imaging

While machine learning (ML) methods may significantly improve image quality for SPECT imaging for the diagnosis and monitoring of Parkinson's disease (PD), they require a large amount of data for training. It is often difficult to collect a large population of patient data to support the ML research, and the ground truth of lesion is also unknown. This paper leverages a generative adversarial network (GAN) to generate digital brain phantoms for training ML-based PD SPECT algorithms. A total of 594 PET 3D brain models from 155 patients (113 male and 42 female) were reviewed and 1597 2D slices containing the full or a portion of the striatum were selected. Corresponding attenuation maps were also generated based on these images. The data were then used to develop a GAN for generating 2D brain phantoms, where each phantom consisted of a radioactivity image and the corresponding attenuation map. Statistical methods including histogram, Fréchet distance, and structural similarity were used to evaluate the generator based on 10,000 generated phantoms. When the generated phantoms and training dataset were both passed to the discriminator, similar normal distributions were obtained, which indicated the discriminator was unable to distinguish the generated phantoms from the training datasets. The generated digital phantoms can be used for 2D SPECT simulation and serve as the ground truth to develop ML-based reconstruction algorithms. The cumulated experience from this work also laid the foundation for building a 3D GAN for the same application.

Dielectric Breast Phantoms by Generative Adversarial Network

In order to conduct the research of machine-learning (ML) based microwave breast imaging (MBI), a large number of digital dielectric breast phantoms that can be used as training data (ground truth) are required but are difficult to be achieved from practice. Although a few dielectric breast phantoms have been developed for research purpose, the number and the diversity are limited and is far inadequate to develop a robust ML algorithm for MBI. This paper presents a neural network method to generate 2D virtual breast phantoms that are similar to the real ones, which can be used to develop ML-based MBI in the future. The generated phantoms are similar but are different from those used in training. Each phantom consists of several images with each representing the distribution of a dielectric parameter in the breast map. Statistical analysis was performed over 10,000 generated phantoms to investigate the performance of the generative network. With the generative network, one may generate unlimited number of breast images with more variations, so the ML-based MBI will be more ready to deploy.

Dual-domain sparse-view CT reconstruction with Transformers

PURPOSE

Computed Tomography (CT) has been widely used in the medical field. Sparse-view CT is an effective and feasible method to reduce the radiation dose. However, the conventional filtered back projection (FBP) algorithm will suffer from severe artifacts in sparse-view CT. Iterative reconstruction algorithms have been adopted to remove artifacts, but they are time-consuming due to repeated projection and back projection and may cause blocky effects. To overcome the difficulty in sparse-view CT, we proposed a dual-domain sparse-view CT algorithm CT Transformer (CTTR) and paid attention to sinogram information.

METHODS

CTTR treats sinograms as sentences and enhances reconstructed images with sinogram's characteristics. We qualitatively evaluate the CTTR, an iterative method TVM-POCS, a convolutional neural network based method FBPConvNet in terms of a reduction in artifacts and a preservation of details. Besides, we also quantitatively evaluate these methods in terms of RMSE, PSNR and SSIM.

RESULTS

We evaluate our method on the Lung Image Database Consortium image collection with different numbers of projection views and noise levels. Experiment studies show that, compared with other methods, CTTR can reduce more artifacts and preserve more details on various scenarios. Specifically, CTTR improves the FBPConvNet performance of PSNR by 0.76dB with 30 projections.

CONCLUSIONS

The performance of our proposed CTTR is better than the method based on CNN in the case of extremely sparse views both on visual results and quantitative evaluation. Our proposed method provides a new idea for the application of Transformers to CT image processing.

DIOR: Deep Iterative Optimization-Based Residual-Learning for Limited-Angle CT Reconstruction

Limited-angle CT is a challenging problem in real applications. Incomplete projection data will lead to severe artifacts and distortions in reconstruction images. To tackle this problem, we propose a novel reconstruction framework termed Deep Iterative Optimization-based Residual-learning (DIOR) for limited-angle CT. Instead of directly deploying the regularization term on image space, the DIOR combines iterative optimization and deep learning based on the residual domain, significantly improving the convergence property and generalization ability. Specifically, the asymmetric convolutional modules are adopted to strengthen the feature extraction capacity in smooth regions for deep priors. Besides, in our DIOR method, the information contained in low-frequency and high-frequency components is also evaluated by perceptual loss to improve the performance in tissue preservation. Both simulated and clinical datasets are performed to validate the performance of DIOR. Compared with existing competitive algorithms, quantitative and qualitative results show that the proposed method brings a promising improvement in artifact removal, detail restoration and edge preservation.

CCN-CL: A content-noise complementary network with contrastive learning for low-dose computed tomography denoising

In recent years, low-dose computed tomography (LDCT) has played an increasingly important role in the diagnosis CT to reduce the potential adverse effects of x-ray radiation on patients while maintaining the same diagnostic image quality. Current deep learning-based denoising methods applied to LDCT imaging only use normal dose CT (NDCT) images as positive examples to guide the denoising process. Recent studies on contrastive learning have proved that the original images as negative examples can also be helpful for network learning. Therefore, this paper proposes a novel content-noise complementary network with contrastive learning for an LDCT denoising task. First, to better train our proposed network, a contrastive learning loss, taking the NDCT image as a positive example and the original LDCT image as a negative example to guide the network learning is added. Furthermore, we also design a network structure that combines content-noise complementary learning strategy, attention mechanism, and deformable convolution for better network performance. In an evaluation study, we compare the performance of our designed network with some of the state-of-the-art methods in the 2016 NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge dataset. The quantitative and qualitative evaluation results demonstrate the feasibility and effectiveness of applying our proposed CCN-CL network model as a new deep learning-based LDCT denoising method.

Deep Learning-Based Image Reconstruction for Different Medical Imaging Modalities

Image reconstruction in magnetic resonance imaging (MRI) and computed tomography (CT) is a mathematical process that generates images at many different angles around the patient. Image reconstruction has a fundamental impact on image quality. In recent years, the literature has focused on deep learning and its applications in medical imaging, particularly image reconstruction. Due to the performance of deep learning models in a wide variety of vision applications, a considerable amount of work has recently been carried out using image reconstruction in medical images. MRI and CT appear as the ultimate scientifically appropriate imaging mode for identifying and diagnosing different diseases in this ascension age of technology. This study demonstrates a number of deep learning image reconstruction approaches and a comprehensive review of the most widely used different databases. We also give the challenges and promising future directions for medical image reconstruction.

Multi-domain integrative Swin transformer network for sparse-view tomographic reconstruction

Decreasing projection views to a lower X-ray radiation dose usually leads to severe streak artifacts. To improve image quality from sparse-view data, a multi-domain integrative Swin transformer network (MIST-net) was developed and is reported in this article. First, MIST-net incorporated lavish domain features from data, residual data, image, and residual image using flexible network architectures, where a residual data and residual image sub-network was considered as a data consistency module to eliminate interpolation and reconstruction errors. Second, a trainable edge enhancement filter was incorporated to detect and protect image edges. Third, a high-quality reconstruction Swin transformer (i.e., Recformer) was designed to capture image global features. The experimental results on numerical and real cardiac clinical datasets with 48 views demonstrated that our proposed MIST-net provided better image quality with more small features and sharp edges than other competitors.

Calibration by differentiation - Self-supervised calibration for X-ray microscopy using a differentiable cone-beam reconstruction operator

High-resolution X-ray microscopy (XRM) is gaining interest for biological investigations of extremely small-scale structures. XRM imaging of bones in living mice could provide new insights into the emergence and treatment of osteoporosis by observing osteocyte lacunae, which are holes in the bone of few micrometers in size. Imaging living animals at that resolution, however, is extremely challenging and requires very sophisticated data processing converting the raw XRM detector output into reconstructed images. This paper presents an open-source, differentiable reconstruction pipeline for XRM data which analytically computes the final image from the raw measurements. In contrast to most proprietary reconstruction software, it offers the user full control over each processing step and, additionally, makes the entire pipeline deep learning compatible by ensuring differentiability. This allows fitting trainable modules both before and after the actual reconstruction step in a purely data-driven way using the gradient-based optimizers of common deep learning frameworks. The value of such differentiability is demonstrated by calibrating the parameters of a simple cupping correction module operating on the raw projection images using only a self-supervisory quality metric based on the reconstructed volume and no further calibration measurements. The retrospective calibration directly improves image quality as it avoids cupping artifacts and decreases the difference in gray values between outer and inner bone by 68% to 94%. Furthermore, it makes the reconstruction process entirely independent of the XRM manufacturer and paves the way to explore modern deep learning reconstruction methods for arbitrary XRM and, potentially, other flat-panel CT systems. This exemplifies how differentiable reconstruction can be leveraged in the context of XRM and, hence, is an important step toward the goal of reducing the resolution limit of in-vivo bone imaging to the single micrometer domain. This article is protected by copyright. All rights reserved.

The use of deep learning methods in low-dose computed tomography image reconstruction: a systematic review

Conventional reconstruction techniques, such as filtered back projection (FBP) and iterative reconstruction (IR), which have been utilised widely in the image reconstruction process of computed tomography (CT) are not suitable in the case of low-dose CT applications, because of the unsatisfying quality of the reconstructed image and inefficient reconstruction time. Therefore, as the demand for CT radiation dose reduction continues to increase, the use of artificial intelligence (AI) in image reconstruction has become a trend that attracts more and more attention. This systematic review examined various deep learning methods to determine their characteristics, availability, intended use and expected outputs concerning low-dose CT image reconstruction. Utilising the methodology of Kitchenham and Charter, we performed a systematic search of the literature from 2016 to 2021 in Springer, Science Direct, arXiv, PubMed, ACM, IEEE, and Scopus. This review showed that algorithms using deep learning technology are superior to traditional IR methods in noise suppression, artifact reduction and structure preservation, in terms of improving the image quality of low-dose reconstructed images. In conclusion, we provided an overview of the use of deep learning approaches in low-dose CT image reconstruction together with their benefits, limitations, and opportunities for improvement.

Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks

Due to lack of the kernel awareness, some popular deep image reconstruction networks are unstable. To address this problem, here we introduce the bounded relative error norm (BREN) property, which is a special case of the Lipschitz continuity. Then, we perform a convergence study consisting of two parts: (1) a heuristic analysis on the convergence of the analytic compressed iterative deep (ACID) scheme (with the simplification that the CS module achieves a perfect sparsification), and (2) a mathematically denser analysis (with the two approximations: [1] AT is viewed as an inverse A- 1 in the perspective of an iterative reconstruction procedure and [2] a pseudo-inverse is used for a total variation operator H). Also, we present adversarial attack algorithms to perturb the selected reconstruction networks respectively and, more importantly, to attack the ACID workflow as a whole. Finally, we show the numerical convergence of the ACID iteration in terms of the Lipschitz constant and the local stability against noise.

Stabilizing deep tomographic reconstruction: Part A. Hybrid framework and experimental results

A recent PNAS paper reveals that several popular deep reconstruction networks are unstable. Specifically, three kinds of instabilities were reported: (1) strong image artefacts from tiny perturbations, (2) small features missed in a deeply reconstructed image, and (3) decreased imaging performance with increased input data. Here, we propose an analytic compressed iterative deep (ACID) framework to address this challenge. ACID synergizes a deep network trained on big data, kernel awareness from compressed sensing (CS)-inspired processing, and iterative refinement to minimize the data residual relative to real measurement. Our study demonstrates that the ACID reconstruction is accurate, is stable, and sheds light on the converging mechanism of the ACID iteration under a bounded relative error norm assumption. ACID not only stabilizes an unstable deep reconstruction network but also is resilient against adversarial attacks to the whole ACID workflow, being superior to classic sparsity-regularized reconstruction and eliminating the three kinds of instabilities.

Understanding the Research Landscape of Deep Learning in Biomedical Science: Scientometric Analysis

BACKGROUND

Advances in biomedical research using deep learning techniques have generated a large volume of related literature. However, there is a lack of scientometric studies that provide a bird's-eye view of them. This absence has led to a partial and fragmented understanding of the field and its progress.

OBJECTIVE

This study aimed to gain a quantitative and qualitative understanding of the scientific domain by analyzing diverse bibliographic entities that represent the research landscape from multiple perspectives and levels of granularity.

METHODS

We searched and retrieved 978 deep learning studies in biomedicine from the PubMed database. A scientometric analysis was performed by analyzing the metadata, content of influential works, and cited references.

RESULTS

In the process, we identified the current leading fields, major research topics and techniques, knowledge diffusion, and research collaboration. There was a predominant focus on applying deep learning, especially convolutional neural networks, to radiology and medical imaging, whereas a few studies focused on protein or genome analysis. Radiology and medical imaging also appeared to be the most significant knowledge sources and an important field in knowledge diffusion, followed by computer science and electrical engineering. A coauthorship analysis revealed various collaborations among engineering-oriented and biomedicine-oriented clusters of disciplines.

CONCLUSIONS

This study investigated the landscape of deep learning research in biomedicine and confirmed its interdisciplinary nature. Although it has been successful, we believe that there is a need for diverse applications in certain areas to further boost the contributions of deep learning in addressing biomedical research problems. We expect the results of this study to help researchers and communities better align their present and future work.

Training low dose CT denoising network without high quality reference data

Objective. Currently, the field of low-dose CT (LDCT) denoising is dominated by supervised learning based methods, which need perfectly registered pairs of LDCT and its corresponding clean reference image (normal-dose CT). However, training without clean labels is more practically feasible and significant, since it is clinically impossible to acquire a large amount of these paired samples. In this paper, a self-supervised denoising method is proposed for LDCT imaging. Approach. The proposed method does not require any clean images. In addition, the perceptual loss is used to achieve data consistency in feature domain during the denoising process. Attention blocks used in decoding phase can help further improve the image quality. Main results. In the experiments, we validate the effectiveness of our proposed self-supervised framework and compare our method with several state-of-the-art supervised and unsupervised methods. The results show that our proposed model achieves competitive performance in both qualitative and quantitative aspects to other methods. Significance. Our framework can be directly applied to most denoising scenarios without collecting pairs of training data, which is more flexible for real clinical scenario.

Two-stage deep learning network-based few-view image reconstruction for parallel-beam projection tomography

BACKGROUND

Projection tomography (PT) is a very important and valuable method for fast volumetric imaging with isotropic spatial resolution. Sparse-view or limited-angle reconstruction-based PT can greatly reduce data acquisition time, lower radiation doses, and simplify sample fixation modes. However, few techniques can currently achieve image reconstruction based on few-view projection data, which is especially important for in vivo PT in living organisms.

METHODS

A 2-stage deep learning network (TSDLN)-based framework was proposed for parallel-beam PT reconstructions using few-view projections. The framework is composed of a reconstruction network (R-net) and a correction network (C-net). The R-net is a generative adversarial network (GAN) used to complete image information with direct back-projection (BP) of a sparse signal, bringing the reconstructed image close to reconstruction results obtained from fully projected data. The C-net is a U-net array that denoises the compensation result to obtain a high-quality reconstructed image.

RESULTS

The accuracy and feasibility of the proposed TSDLN-based framework in few-view projection-based reconstruction were first evaluated with simulations, using images from the DeepLesion public dataset. The framework exhibited better reconstruction performance than traditional analytic reconstruction algorithms and iterative algorithms, especially in cases using sparse-view projection images. For example, with as few as two projections, the TSDLN-based framework reconstructed high-quality images very close to the original image, with structural similarities greater than 0.8. By using previously acquired optical PT (OPT) data in the TSDLN-based framework trained on computed tomography (CT) data, we further exemplified the migration capabilities of the TSDLN-based framework. The results showed that when the number of projections was reduced to 5, the contours and distribution information of the samples in question could still be seen in the reconstructed images.

CONCLUSIONS

The simulations and experimental results showed that the TSDLN-based framework has strong reconstruction abilities using few-view projection images, and has great potential in the application of in vivo PT.

ADMM-SVNet: An ADMM-Based Sparse-View CT Reconstruction Network

In clinical medical applications, sparse-view computed tomography (CT) imaging is an effective method for reducing radiation doses. The iterative reconstruction method is usually adopted for sparse-view CT. In the process of optimizing the iterative model, the approach of directly solving the quadratic penalty function of the objective function can be expected to perform poorly. Compared with the direct solution method, the alternating direction method of multipliers (ADMM) algorithm can avoid the ill-posed problem associated with the quadratic penalty function. However, the regularization items, sparsity transform, and parameters in the traditional ADMM iterative model need to be manually adjusted. In this paper, we propose a data-driven ADMM reconstruction method that can automatically optimize the above terms that are difficult to choose within an iterative framework. The main contribution of this paper is that a modified U-net represents the sparse transformation, and the prior information and related parameters are automatically trained by the network. Based on a comparison with other state-of-the-art reconstruction algorithms, the qualitative and quantitative results show the effectiveness of our method for sparse-view CT image reconstruction. The experimental results show that the proposed method performs well in streak artifact elimination and detail structure preservation. The proposed network can deal with a wide range of noise levels and has exceptional performance in low-dose reconstruction tasks.

ISP-Net: Fusing features to predict ischemic stroke infarct core on CT perfusion maps

BACKGROUND

Acute ischemic stroke is one of the leading death causes. Delineating stoke infarct core in medical images plays a critical role in optimal stroke treatment selection. However, accurate estimation of infarct core still remains challenging because of 1) the large shape and location variation of infarct cores; 2) the complex relationships between perfusion parameters and final tissue outcome.

METHODS

We develop an encoder-decoder based semantic model, i.e., Ischemic Stroke Prediction Network (ISP-Net), to predict infarct core after thrombolysis treatment on CT perfusion (CTP) maps. Features of native CTP, CBF (Cerebral Blood Flow), CBV (Cerebral Blood Volume), MTT (Mean Transit Time), Tmax are generated and fused with five-path convolutions for comprehensive analysis. A multi-scale atrous convolution (MSAC) block is firstly put forward as the enriched high-level feature extractor in ISP-Net to improve prediction accuracy. A retrospective dataset which is collected from multiple stroke centers is used to evaluate the performance of ISP-Net. The gold standard infarct cores are delineated on the follow-up scans, i.e., non-contrast CT (NCCT) or MRI diffusion-weighted image (DWI).

RESULTS

In clinical dataset cross-validation, we achieve mean Dice Similarity Coefficient (DSC) of 0.801, precision of 81.3%, sensitivity of 79.5%, specificity of 99.5%, Area Under Curve (AUC) of 0.721. Our approach yields better outcomes than several advanced deep learning methods, i.e., Deeplab V3, U-Net++, CE-Net, X-Net and Non-local U-Net, demonstrating the promising performance in infarct core prediction. No significant difference of the prediction error is shown for the patients with follow-up NCCT and follow-up DWI (P >0.05).

CONCLUSION

This study provides an approach for fast and accurate stroke infarct core estimation. We anticipate the prediction results of ISP-Net could offer assistance to the physicians in the thrombolysis or thrombectomy therapy selection.

Impact of the training loss in deep learning based CT reconstruction of bone microarchitecture

PURPOSE

Computed tomography (CT) is a technique of choice to image bone structure at different scales. Methods to enhance the quality of degraded reconstructions obtained from low-dose CT data have shown impressive results recently, especially in the realm of supervised deep learning. As the choice of the loss function affects the reconstruction quality, it is necessary to focus on the way neural networks evaluate the correspondence between predicted and target images during the training stage. This is even more true in the case of bone microarchitecture imaging at high spatial resolution where both the quantitative analysis of Bone Mineral Density (BMD) and bone microstructure are essential for assessing diseases such as osteoporosis. Our aim is thus to evaluate the quality of reconstruction on key metrics for diagnosis depending on the loss function that has been used for training the neural network.

METHODS

We compare and analyze volumes that are reconstructed with neural networks trained with pixelwise, structural and adversarial loss functions or with a combination of them. We perform realistic simulations of various low-dose acquisitions of bone microarchitecture. Our comparative study is performed with metrics that have an interest regarding the diagnosis of bone diseases. We therefore focus on bone-specific metrics such as BV/TV, resolution, connectivity assessed with the Euler number and quantitative analysis of BMD to evaluate the quality of reconstruction obtained with networks trained with the different loss functions.

RESULTS

We find that using L₁ norm as the pixelwise loss is the best choice compared to L₂ or no pixelwise loss since it improves resolution without deteriorating other metrics. VGG perceptual loss, especially when combined with an adversarial loss, allows to better retrieve topological and morphological parameters of bone microarchitecture compared to SSIM. This however leads to a decreased resolution performance. The adversarial loss enchances the reconstruction performance in terms of BMD distribution accuracy.

CONCLUSIONS

In order to retrieve the quantitative and structural characteristics of bone microarchitecture that are essential for post-reconstruction diagnosis, our results suggest to use L₁ norm as part of the loss function. Then, trade-offs should be made depending on the application: VGG perceptual loss improves accuracy in terms of connectivity at the cost of a deteriorated resolution, and adversarial losses help better retrieve BMD distribution while significantly increasing the training time. This article is protected by copyright. All rights reserved.