Image Reconstruction Performance Research Articles

Serial Decoders-Based Auto-Encoders for Image Reconstruction

Auto-encoders are composed of coding and decoding units; hence, they hold an inherent potential of being used for high-performance data compression and signal-compressed sensing. The main disadvantages of current auto-encoders comprise the following aspects: the research objective is not to achieve lossless data reconstruction but efficient feature representation; the evaluation of data recovery performance is neglected; it is difficult to achieve lossless data reconstruction using pure auto-encoders, even with pure deep learning. This paper aims at performing image reconstruction using auto-encoders, employs cascade decoders-based auto-encoders, perfects the performance of image reconstruction, approaches gradually lossless image recovery, and provides a solid theoretical and applicational basis for auto-encoders-based image compression and compressed sensing. The proposed serial decoders-based auto-encoders include the architectures of multi-level decoders and their related progressive optimization sub-problems. The cascade decoders consist of general decoders, residual decoders, adversarial decoders, and their combinations. The effectiveness of residual cascade decoders for image reconstruction is proven in mathematics. Progressive training can efficiently enhance the quality, stability, and variation of image reconstruction. It has been shown by the experimental results that the proposed auto-encoders outperform classical auto-encoders in the performance of image reconstruction.

Fourier Imager Network (FIN): A deep neural network for hologram reconstruction with superior external generalization

Deep learning-based image reconstruction methods have achieved remarkable success in phase recovery and holographic imaging. However, the generalization of their image reconstruction performance to new types of samples never seen by the network remains a challenge. Here we introduce a deep learning framework, termed Fourier Imager Network (FIN), that can perform end-to-end phase recovery and image reconstruction from raw holograms of new types of samples, exhibiting unprecedented success in external generalization. FIN architecture is based on spatial Fourier transform modules that process the spatial frequencies of its inputs using learnable filters and a global receptive field. Compared with existing convolutional deep neural networks used for hologram reconstruction, FIN exhibits superior generalization to new types of samples, while also being much faster in its image inference speed, completing the hologram reconstruction task in ~0.04 s per 1 mm2 of the sample area. We experimentally validated the performance of FIN by training it using human lung tissue samples and blindly testing it on human prostate, salivary gland tissue and Pap smear samples, proving its superior external generalization and image reconstruction speed. Beyond holographic microscopy and quantitative phase imaging, FIN and the underlying neural network architecture might open up various new opportunities to design broadly generalizable deep learning models in computational imaging and machine vision fields.

Short-exposure Image Reconstruction with The Power Spectrum Extended (PSE) Method

We analyze, in the framework of high angular resolution imaging, a novel image reconstruction method denoted as PSE (which stands for power spectrum extended). It works in the Fourier space, combining the information from both the average power spectrum of the images and a phase estimation from an ad-hoc shift-and-add process. PSE allows to perform image reconstruction up to the diffraction limit of the telescope from a series of short-exposure frames, with a refined lucky-imaging selection process. The method is well adapted to partially corrected adaptive-optics images, in particular in case of low Strehl corrections, and/or small diameter telescopes. In this paper we analyze the PSE technique by means of Monte-Carlo simulations and compare it with the ISFAS lucky-imaging method. Comparative performances were investigated using three metrics: Strehl ratio for reconstructed point-like sources, intensity ratio for binary stars, and least-square distance between images for a simulated artificial satellite. We found that PSE provides an improvement of a factor ∼2 over ISFAS on the Strehl ratio in the case of faint point sources. It seems also to give better images reconstruction on some kinds of extended objects (planets or binary stars with small magnitude difference). PSE has also the advantage to be very fast and well adapted to real-time image reconstruction.

A Dynamic Convolution Kernel Generation Method Based on Regularized Pattern for Image Super-Resolution.

Image super-resolution aims to reconstruct a high-resolution image from its low-resolution counterparts. Conventional image super-resolution approaches share the same spatial convolution kernel for the whole image in the upscaling modules, which neglect the specificity of content information in different positions of the image. In view of this, this paper proposes a regularized pattern method to represent spatially variant structural features in an image and further exploits a dynamic convolution kernel generation method to match the regularized pattern and improve image reconstruction performance. To be more specific, first, the proposed approach extracts features from low-resolution images using a self-organizing feature mapping network to construct regularized patterns (RP), which describe different contents at different locations. Second, the meta-learning mechanism based on the regularized pattern predicts the weights of the convolution kernels that match the regularized pattern for each different location; therefore, it generates different upscaling functions for images with different content. Extensive experiments are conducted using the benchmark datasets Set5, Set14, B100, Urban100, and Manga109 to demonstrate that the proposed approach outperforms the state-of-the-art super-resolution approaches in terms of both PSNR and SSIM performance.

Frequency-selective signal enhancement by a passive dual coil resonator for magnetic particle imaging

Objective. Magnetic particle imaging (MPI) visualizes the spatial distribution of magnetic nanoparticles. MPI already provides excellent temporal and good spatial resolution, however, to achieve translation into clinics, further advances in the fields of sensitivity, image reconstruction and tracer performance are needed. In this work, we propose a novel concept to enhance the MPI signal and image resolution by a purely passive receive coil insert for a preclinical MPI system. Approach. The passive dual coil resonator (pDCR) provides frequency-selective signal enhancement. This is enabled by the adaptable resonance frequency of the pDCR network, which is galvanically isolated from the MPI system and composed of two coaxial solenoids connected via a capacitor. The pDCR aims to enhance frequency components related to high mixing orders, which are crucial to achieve high spatial resolution. Main Results. In this study, system matrix measurements and image acquisitions of a resolution phantom are carried out to evaluate the performance of the pDCR compared to the integrated receive unit of the preclinical MPI and a dedicated rat-sized receive coil. Frequency-selective signal increase and spatial resolution enhancement are demonstrated. Significance. Common dedicated receive coils come along with noise-matched receive networks, which makes them costly and difficult to reproduce. The presented pDCR is a purely passive coil insert that gets along without any additional receive electronics. Therefore, it is cost-efficient, easy-to-handle and adaptable to other MPI scanners and potentially other applications providing the basis for a new breed of passive MPI receiver systems.

Monte Carlo characterization of PETALO, a full-body liquid xenon-based PET detector

New detector approaches in Positron Emission Tomography imaging will play an important role in reducing costs, lowering administered radiation doses, and improving overall performance. PETALO employs liquid xenon as the active scintillating medium and UV-sensitive silicon photomultipliers for scintillation readout. The scintillation time in liquid xenon is fast enough to register time-of-flight information for each detected coincidence, and sufficient scintillation is produced with low enough fluctuations to obtain good energy resolution. The present simulation study examines a full-body-sized PETALO detector and evaluates its potential performance in PET image reconstruction.

Improvement of performance and sensitivity of 2D and 3D image reconstruction in EIT using EFG forward model

This paper presents a pure element-free Galerkin method (EFGM) forward model for image reconstruction in 2D and 3D electrical impedance tomography (EIT) using an adaptive current injection method. In EIT systems with the adapting current injection method, both static and dynamic images can be reconstructed; however, determination of electrode contact impedances in the complete electrode model is difficult and the Gap model is used. In this paper, in the EIT forward problem a weak form functional based on the Gap model and a pure EFGM approach are developed, and in the EIT inverse problem, Jacobian matrix is computed by the EFGM, and a fast integration technique is introduced to calculate the entries of the Jacobian matrix within an adequate computation time. The influence of increasing the density of nodes at and near the electrodes with steep electric potential gradients on the accuracy of FEM and EFGM forward solutions is investigated, and the performance of the image reconstruction algorithm with the proposed fast integration technique is examined. The numerical results reveal that the proposed EFGM forward model with the fast integration technique has an efficient performance both in terms of mean relative imaging errors and computational time.

3D k-space reflectance fluorescence tomography via deep learning.

We report on the potential to perform image reconstruction in 3D k-space reflectance fluorescence tomography (FT) using deep learning (DL). Herein, we adopt a modified AUTOMAP architecture and develop a training methodology that leverages an open-source Monte-Carlo-based simulator to generate a large dataset. Using an enhanced EMNIST (EEMNIST) dataset as an embedded contrast function allows us to train the network efficiently. The optical strategy utilizes k-space illumination in a reflectance configuration to probe tissue in the mesoscopic regime with high sensitivity and resolution. The proposed DL model training and validation is performed with both in silico data and a phantom experiment. Overall, our results indicate that the approach can correctly reconstruct both single and multiple fluorescent embedding(s) in a 3D volume. Furthermore, the presented technique is shown to outperform the traditional approaches [least-squares (LSQ) and total-variation minimization (TVAL)], especially at higher depths. We, therefore, expect the proposed computational technique to have future implications in preclinical studies.

Image Denoising Using a Compressive Sensing Approach Based on Regularization Constraints.

In remote sensing applications and medical imaging, one of the key points is the acquisition, real-time preprocessing and storage of information. Due to the large amount of information present in the form of images or videos, compression of these data is necessary. Compressed sensing is an efficient technique to meet this challenge. It consists in acquiring a signal, assuming that it can have a sparse representation, by using a minimum number of nonadaptive linear measurements. After this compressed sensing process, a reconstruction of the original signal must be performed at the receiver. Reconstruction techniques are often unable to preserve the texture of the image and tend to smooth out its details. To overcome this problem, we propose, in this work, a compressed sensing reconstruction method that combines the total variation regularization and the non-local self-similarity constraint. The optimization of this method is performed by using an augmented Lagrangian that avoids the difficult problem of nonlinearity and nondifferentiability of the regularization terms. The proposed algorithm, called denoising-compressed sensing by regularization (DCSR) terms, will not only perform image reconstruction but also denoising. To evaluate the performance of the proposed algorithm, we compare its performance with state-of-the-art methods, such as Nesterov’s algorithm, group-based sparse representation and wavelet-based methods, in terms of denoising and preservation of edges, texture and image details, as well as from the point of view of computational complexity. Our approach permits a gain up to 25% in terms of denoising efficiency and visual quality using two metrics: peak signal-to-noise ratio (PSNR) and structural similarity (SSIM).

Super-resolution reconstruction of digital rock CT images based on residual attention mechanism

Computer tomography technology is widely used in geological exploration because it is a nondestructive and three-dimensional imaging method that can be integrated with computer simulation. However, the large-scale application of the computer tomography technique is limited by economic costs and time consumption. Therefore, it is challenging and intractable to indicate the pore structure characteristics of rock. To address this issue, a super-resolution reconstruction algorithm based on convolutional neural networks, residual learning, and attention mechanism was proposed to generate super-resolution images in this study. This algorithm was applied to the reconstruction of carbonate rock and sandstone. The performance of two-dimensional image reconstruction was evaluated by quantitative extraction and qualitative visualization. The results from experiments indicate that the built model performs well on different upscaling factors and is superior to the existing super-resolution approaches based on convolutional neural network. Cited as: Shan, L., Bai, X., Liu, C., Feng, Y., Liu, Y., Qi, Y. Super-resolution reconstruction of digital rock CT images based on residual attention mechanism. Advances in Geo-Energy Research, 2022, 6(2): 157-168. https://doi.org/10.46690/ager.2022.02.07

Performance of low-dose CT image reconstruction for detecting intracerebral hemorrhage: selection of dose, algorithms and their combinations

To investigate the performance of different low-dose CT image reconstruction algorithms for detecting intracerebral hemorrhage. Low-dose CT imaging simulation was performed on CT images of intracerebral hemorrhage at 30%, 25% and 20% of normal dose level (defined as 100% dose). Seven algorithms were tested to reconstruct low-dose CT images for noise suppression, including filtered back projection algorithm (FBP), penalized weighted least squares-total variation (PWLS-TV), non-local mean filter (NLM), block matching 3D (BM3D), residual encoding-decoding convolutional neural network (REDCNN), the FBP convolutional neural network (FBPConvNet) and image restoration iterative residual convolutional network (IRLNet). A deep learning-based model (CNN-LSTM) was used to detect intracerebral hemorrhage on normal dose CT images and low-dose CT images reconstructed using the 7 algorithms. The performance of different reconstruction algorithms for detecting intracerebral hemorrhage was evaluated by comparing the results between normal dose CT images and low-dose CT images. At different dose levels, the low-dose CT images reconstructed by FBP had accuracies of detecting intracerebral hemorrhage of 82.21%, 74.61% and 65.55% at 30%, 25% and 20% dose levels, respectively. At the same dose level (30% dose), the images reconstructed by FBP, PWLS-TV, NLM, BM3D, REDCNN, FBPConvNet and IRLNet algorithms had accuracies for detecting intracerebral hemorrhage of 82.21%, 86.80%, 89.37%, 81.43%, 90.05%, 90.72% and 93.51%, respectively. The images reconstructed by IRLNet at 30%, 25% and 20% dose levels had accuracies for detecting intracerebral hemorrhage of 93.51%, 93.51% and 93.06%, respectively. The performance of reconstructed low-dose CT images for detecting intracerebral hemorrhage is significantly affected by both dose and reconstruction algorithms. In clinical practice, choosing appropriate dose level and reconstruction algorithm can greatly reduce the radiation dose and ensure the detection performance of CT imaging for intracerebral hemorrhage.

Fast Terahertz Imaging Model Based on Group Sparsity and Nonlocal Self-Similarity.

In order to solve the problems of long-term image acquisition time and massive data processing in a terahertz time domain spectroscopy imaging system, a novel fast terahertz imaging model, combined with group sparsity and nonlocal self-similarity (GSNS), is proposed in this paper. In GSNS, the structure similarity and sparsity of image patches in both two-dimensional and three-dimensional space are utilized to obtain high-quality terahertz images. It has the advantages of detail clarity and edge preservation. Furthermore, to overcome the high computational costs of matrix inversion in traditional split Bregman iteration, an acceleration scheme based on conjugate gradient method is proposed to solve the terahertz imaging model more efficiently. Experiments results demonstrate that the proposed approach can lead to better terahertz image reconstruction performance at low sampling rates.

Range-Doppler image reconstruction for collocated MIMO noise radar by sparse recovery

ABSTRACT This letter deals with the range-Doppler image reconstruction problem in co-located multiple input multiple output (MIMO) noise radar, where the waveform diversity is applied not only among different transmit channels but also among different pulse repetition intervals (PRI). The transmit channel diversity and pulse diversity lead to high range-Doppler sidelobes, which would cause the weak target to be overwhelmed by strong target or clutters. In this letter, we propose a sparse recovery-based approach to deal with the range-Doppler image reconstruction problem in collocated MIMO noise radar, by exploiting the 2D-block sparsity of range-Doppler profiles to enhance the range-Doppler image reconstruction performance and its robustness to clutter. Simulation results show that the range-Doppler image reconstruction performance of the proposed method outperforms traditional methods, especially when noise and strong clutter exist.

STLS-LADMM-Net: A Deep Network for SAR Autofocus Imaging

Synthetic aperture radar (SAR) can provide high-resolution electromagnetic backscattering images of the illuminated area, playing a significant role in various applications. However, achieving focused SAR images is challenging under sparse sampling and phase error conditions. By exploiting the sparsity or compressibility priors, the state-of-the-art sparsity-driven SAR imaging methods can reconstruct images under the condition of sparse sampling. However, the handcrafted priors used in these methods limit the imaging performance, and the iterative solution schemes reduce the computational efficiency. Besides, the measurement inaccuracy introduced by the phase error also degrades the reconstruction performance of the sparsity-driven imaging methods. To address these issues, a deep network for SAR autofocus imaging is proposed, which alternately performs image reconstruction and phase error estimation. When performing image reconstruction, the sparsity-cognizant total least-square (S-TLS) model is introduced to handle the problem of measurement inaccuracy, contributing to robust reconstruction performance under the condition of phase error. During the implementation of the deep network, a feature transform operator is used to realize data-driven prior knowledge learning and overcome the limitations of handcrafted priors. Moreover, the deep network approach can significantly improve computational efficiency. Experiments on simulated and real data verify the effectiveness and efficiency of the proposed method.

Image Reconstruction With Deep CNN for Mirrored Aperture Synthesis

In mirrored aperture synthesis (MAS), the existing brightness temperature image reconstruction methods include inverse cosine transform and impulse matrix reconstruction methods. However, the quality of the MAS brightness temperature images reconstructed by the existing methods is still poor and needs to be improved. This article proposes a method of MAS brightness temperature image reconstruction with deep convolutional neural network (CNN). The network includes two fully connected (FC) layers, multiple convolutional layers, and deconvolutional layers, which realize the image reconstruction for MAS. This method uses deep CNN to learn the MAS image reconstruction mapping and system errors, so as to improve the performance of the brightness temperature image reconstruction. Both simulation and experimental results verify that the performance of the proposed MAS-CNN method is better than the existing MAS image reconstruction methods.

Target-Oriented SAR Imaging for SCR Improvement via Deep MF-ADMM-Net

Synthetic aperture radar (SAR) is an important means for target surveillance through reconstructing the microwave image of the observation area. However, under the condition of low signal-to-clutter ratio (SCR), such as a strong sea clutter situation, it is difficult to surveil targets from SAR images acquired by the traditional matched filter-based imaging methods. To improve the target surveillance performance of SAR, this article proposes a target-oriented SAR imaging method, which can enhance the desired target and improve the SCR in the reconstructed SAR images. By separating the target area from the clutter area, we first establish a target-oriented SAR imaging model, where the generalized regularization is used to characterize the features of the target, contributing to the improvement of SCR in the reconstructed image. Then, the imaging model is solved through a deep network, MF-ADMM-Net, which is obtained by unfolding an alternating direction method of multipliers (ADMM)-based iterative solution. In addition, the training strategy is formulated with the consideration of complex values. Experiments are conducted to verify the performance of image reconstruction and SCR improvement of the proposed method, and comparisons show the superiority of MF-ADMM-Net in effect and efficiency.

Single-Pass Object-Adaptive Data Undersampling and Reconstruction for MRI

There is much recent interest in techniques to accelerate the data acquisition process in MRI by acquiring limited measurements. Often sophisticated reconstruction algorithms are deployed to maintain high image quality in such settings. In this work, we propose a data-driven sampler using a convolutional neural network, MNet, to provide object-specific sampling patterns adaptive to each scanned object. The network observes very limited low-frequency k-space data for each object and rapidly predicts the desired undersampling pattern in one go that achieves high image reconstruction quality. We propose an accompanying alternating-type training framework with a mask-backward procedure that efficiently generates training labels for the sampler network and jointly trains an image reconstruction network. Experimental results on the fastMRI knee dataset demonstrate the ability of the proposed learned undersampling network to generate object-specific masks at fourfold and eightfold acceleration that achieve superior image reconstruction performance than several existing schemes. The source code for the proposed joint sampling and reconstruction learning framework is available at https://github.com/zhishenhuang/mri.

Progressive image transmission based on image spatio-temporal decomposition by sigma-delta cellular neural network

In this paper, we propose a novel progressive image transmission framework based on spatio-temporal image decomposition and synthesis by the SD-CNN. In our method, we redesign the baseline SD-CNN and the weighted sum is introduced in the accumulator. This innovation enables lossless or near-lossless progressive image transmission. Experimental results in various test images support that the image reconstruction performance of the SD-CNN has dramatically improved by our method.

DLSLA 3-D SAR Imaging via Sparse Recovery Through Combination of Nuclear Norm and Low-Rank Matrix Factorization

Downward-looking sparse linear array 3-D synthetic aperture radar (DLSLA 3-D SAR) cross-track dimensional imaging always suffers from incomplete observation which does not satisfy the Nyquist sampling theorem and leads to the failure of conventional 3-D frequency-domain methods. Although several sparse reconstruction-based methods have been presented to solve this problem, the basis mismatch issue in sparse reconstruction theory will degrade the image reconstruction performance. To address this issue, this article proposes a novel 3-D imaging method for DLSLA 3-D SAR, which provides another idea for 3-D imaging through sparse recovery. It utilizes recovered full-sampled data to achieve cross-track dimensional imaging instead of using the under-sampled data directly as before. The Along-track-Height plane imaging is first finished by the range-Doppler (RD) algorithm and motion error compensation. Then, an advanced nuclear norm and low-rank matrix factorization (NU-LRMF)-based matrix completion (MC) algorithm and a vector reconstruction framework are built to achieve accurate recovery of full-sampled data. Finally, the cross-track dimensional imaging is completed with recovered full-sampled data by geometric correction and beamforming. Moreover, a fast two-stage iteration strategy for NU-LRMF (TS-NU-LRMF) is also presented to accelerate convergence. The robustness and effectiveness of the proposed 3-D imaging method are verified by several numerical simulations and comparative studies based on both the complex 3-D ship model and the simulated 3-D distributed scenario.

Imaging Cerebral Energy Metabolism in Healthy Infants.

Broadband near-infrared spectroscopy (bNIRS) has the potential to provide non-invasive measures of cerebral haemodynamic changes alongside changes in cellular oxygen utilisation through the measurement of mitochondrial enzyme cytochrome-c-oxidase (oxCCO). It therefore provides the opportunity to explore brain function and specialisation, which remains largely unexplored in infancy. We used bNIRS to measure changes in haemodynamics and changes in oxCCO in 4-to-7-month-old infants over the occipital and right temporal and parietal cortices in response to social and non-social visual and auditory stimuli. Changes in concentration of oxygenated-haemoglobin (Δ[HbO2]), deoxygenated haemoglobin (Δ[HHb]) and change in the oxidation state of oxCCO (Δ[oxCCO]) were calculated using changes in attenuation of light at 120 wavelengths between 780 and900nm, using the UCLn algorithm. For 4 infants, the attenuation changes in a subset of wavelengths were used to perform image reconstruction, in an age-matched infant model, for channels over the right parietal and temporal cortices, using a multispectral approach which allows direct reconstruction of concentration change data. The volumetric reconstructed images were mapped onto the cortical surface to visualise the reconstructed changes in concentration of HbO2 and HHb and changes in metabolism for both social and non-social stimuli. Spatially localised activation was observed for Δ[oxCCO] and Δ[HbO2] over the temporo-parietal region, in response to the social stimulus. This study provides the first reconstructed images of changes in metabolism in healthy, awake infants.