Super-resolution Deep Learning Research Articles

Deep learning super-resolution reconstruction for fast and high-quality cine cardiovascular magnetic resonance.

To compare standard-resolution balanced steady-state free precession (bSSFP) cine images with cine images acquired at low resolution but reconstructed with a deep learning (DL) super-resolution algorithm. Cine cardiovascular magnetic resonance (CMR) datasets (short-axis and 4-chamber views) were prospectively acquired in healthy volunteers and patients at normal (cineNR: 1.89 × 1.96 mm2, reconstructed at 1.04 × 1.04 mm2) and at a low-resolution (2.98 × 3.00 mm2, reconstructed at 1.04 × 1.04 mm2). Low-resolution images were reconstructed using compressed sensing DL denoising and resolution upscaling (cineDL). Left ventricular ejection fraction (LVEF), end-diastolic volume index (LVEDVi), and strain were assessed. Apparent signal-to-noise (aSNR) and contrast-to-noise ratios (aCNR) were calculated. Subjective image quality was assessed on a 5-point Likert scale. Student's paired t-test, Wilcoxon matched-pairs signed-rank-test, and intraclass correlation coefficient (ICC) were used for statistical analysis. Thirty participants were analyzed (37 ± 16 years; 20 healthy volunteers and 10 patients). Short-axis views whole-stack acquisition duration of cineDL was shorter than cineNR (57.5 ± 8.7 vs 98.7 ± 12.4 s; p < 0.0001). No differences were noted for: LVEF (59 ± 7 vs 59 ± 7%; ICC: 0.95 [95% confidence interval: 0.94, 0.99]; p = 0.17), LVEDVi (85.0 ± 13.5 vs 84.4 ± 13.7 mL/m2; ICC: 0.99 [0.98, 0.99]; p = 0.12), longitudinal strain (-19.5 ± 4.3 vs -19.8 ± 3.9%; ICC: 0.94 [0.88, 0.97]; p = 0.52), short-axis aSNR (81 ± 49 vs 69 ± 38; p = 0.32), aCNR (53 ± 31 vs 45 ± 27; p = 0.33), or subjective image quality (5.0 [IQR 4.9, 5.0] vs 5.0 [IQR 4.7, 5.0]; p = 0.99). Deep-learning reconstruction of cine images acquired at a lower spatial resolution led to a decrease in acquisition times of 42% with shorter breath-holds without affecting volumetric results or image quality. Question Cine CMR acquisitions are time-intensive and vulnerable to artifacts. Findings Low-resolution upscaled reconstructions using DL super-resolution decreased acquisition times by 35-42% without a significant difference in volumetric results or subjective image quality. Clinical relevance DL super-resolution reconstructions of bSSFP cine images acquired at a lower spatial resolution reduce acquisition times while preserving diagnostic accuracy, improving the clinical feasibility of cine imaging by decreasing breath hold duration.

Absolute permeability estimation from microtomography rock images through deep learning super-resolution and adversarial fine tuning

The carbon capture and storage (CCS) process has become one of the main technologies used for mitigating greenhouse gas emissions. The success of CCS projects relies on accurate subsurface reservoir petrophysical characterization, enabling efficient storage and captured CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ extrm{CO}_2$$\\end{document} containment. In digital rock physics, X-ray microtomography (μ\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\upmu $$\\end{document}-CT) is applied to characterize reservoir rocks, allowing a more assertive analysis of physical properties such as porosity and permeability, enabling better simulations of porous media flow. Estimating petrophysical properties through numeric simulations usually requires high-resolution images, which are expensive and time-inefficient to obtain with μ\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\upmu $$\\end{document}-CT. To address this, we propose using two deep learning models: a super-resolution model to enhance the quality of low-resolution images and a surrogate model that acts as a substitute for numerical simulations to estimate the petrophysical property of interest. A correction process inspired by generative adversarial network (GAN) adversarial training is applied. In this approach, the super-resolution model acts as a generator, creating high-resolution images, and the surrogate network acts as a discriminator. By adjusting the generator, images that correct the errors in the surrogate’s estimations are produced. The proposed method was applied to the DeePore dataset. The results shows the proposed approach improved permeability estimation overall.

Cross-Modality Reference and Feature Mutual-Projection for 3D Brain MRI Image Super-Resolution.

High-resolution (HR) magnetic resonance imaging (MRI) can reveal rich anatomical structures for clinical diagnoses. However, due to hardware and signal-to-noise ratio limitations, MRI images are often collected with low resolution (LR) which is not conducive to diagnosing and analyzing clinical diseases. Recently, deep learning super-resolution (SR) methods have demonstrated great potential in enhancing the resolution of MRI images; however, most of them did not take the cross-modality and internal priors of MR seriously, which hinders the SR performance. In this paper, we propose a cross-modality reference and feature mutual-projection (CRFM) method to enhance the spatial resolution of brain MRI images. Specifically, we feed the gradients of HR MRI images from referenced imaging modality into the SR network to transform true clear textures to LR feature maps. Meanwhile, we design a plug-in feature mutual-projection (FMP) method to capture the cross-scale dependency and cross-modality similarity details of MRI images. Finally, we fuse all feature maps with parallel attentions to produce and refine the HR features adaptively. Extensive experiments on MRI images in the image domain and k-space show that our CRFM method outperforms existing state-of-the-art MRI SR methods.

2547: Projection-based deep learning super-resolution for CBCT dose reduction

2547: Projection-based deep learning super-resolution for CBCT dose reduction

Deep Learning Super-Resolution Technique Based on Magnetic Resonance Imaging for Application of Image-Guided Diagnosis and Surgery of Trigeminal Neuralgia.

This study aimed to implement a deep learning-based super-resolution (SR) technique that can assist in the diagnosis and surgery of trigeminal neuralgia (TN) using magnetic resonance imaging (MRI). Experimental methods applied SR to MRI data examined using five techniques, including T2-weighted imaging (T2WI), T1-weighted imaging (T1WI), contrast-enhancement T1WI (CE-T1WI), T2WI turbo spin-echo series volume isotropic turbo spin-echo acquisition (VISTA), and proton density (PD), in patients diagnosed with TN. The image quality was evaluated using the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). High-quality reconstructed MRI images were assessed using the Leksell coordinate system in gamma knife radiosurgery (GKRS). The results showed that the PSNR and SSIM values achieved by SR were higher than those obtained by image postprocessing techniques, and the coordinates of the images reconstructed in the gamma plan showed no differences from those of the original images. Consequently, SR demonstrated remarkable effects in improving the image quality without discrepancies in the coordinate system, confirming its potential as a useful tool for the diagnosis and surgery of TN.

NAWQ-SR: A Hybrid-Precision NPU Engine for Efficient On-Device Super-Resolution

In recent years, image and video delivery systems have begun integrating deep learning super-resolution (SR) approaches, leveraging their unprecedented visual enhancement capabilities while reducing reliance on networking conditions. Nevertheless, deploying these solutions on mobile devices still remains an active challenge as SR models are excessively demanding with respect to workload and memory footprint. Despite recent progress on on-device SR frameworks, existing systems either penalize visual quality, lead to excessive energy consumption or make inefficient use of the available resources. This work presents NAWQ-SR, a novel framework for the efficient on-device execution of SR models. Through a novel hybrid-precision quantization technique and a runtime neural image codec, NAWQ-SR exploits the multi-precision capabilities of modern mobile NPUs in order to minimize latency, while meeting user-specified quality constraints. Moreover, NAWQ-SR selectively adapts the arithmetic precision at run time to equip the SR DNN's layers with wider representational power, improving visual quality beyond what was previously possible on NPUs. Altogether, NAWQ-SR achieves an average speedup of 7.9×, 3× and 1.91× over the state-of-the-art on-device SR systems that use heterogeneous processors (MobiSR), CPU (SplitSR) and NPU (XLSR), respectively. Furthermore, NAWQ-SR delivers an average of 3.2× speedup and 0.39 dB higher PSNR over status-quo INT8 NPU designs, but most importantly mitigates the negative effects of quantization on visual quality, setting a new state-of-the-art in the attainable quality of NPU-based SR.

Physics-informed deep learning for T2-deblurred superresolution turbo spin echo MRI.

Deep learning superresolution (SR) is a promising approach to reduce MRI scan time without requiring custom sequences or iterative reconstruction. Previous deep learning SR approaches have generated low-resolution training images by simple k-space truncation, but this does not properly model in-plane turbo spin echo (TSE) MRI resolution degradation, which has variable T2 relaxation effects in different k-space regions. To fill this gap, we developed a T2 -deblurred deep learning SR method for the SR of 3D-TSE images. A SR generative adversarial network was trained using physically realistic resolution degradation (asymmetric T2 weighting of raw high-resolution k-space data). For comparison, we trained the same network structure on previous degradation models without TSE physics modeling. We tested all models for both retrospective and prospective SR with 3 × 3 acceleration factor (in the two phase-encoding directions) of genetically engineered mouse embryo model TSE-MR images. The proposed method can produce high-quality 3 × 3 SR images for a typical 500-slice volume with 6-7 mouse embryos. Because 3 × 3 SR was performed, the image acquisition time can be reduced from 15 h to 1.7 h. Compared to previous SR methods without TSE modeling, the proposed method achieved the best quantitative imaging metrics for both retrospective and prospective evaluations and achieved the best imaging-quality expert scores for prospective evaluation. The proposed T2 -deblurring method improved accuracy and image quality of deep learning-based SR of TSE MRI. This method has the potential to accelerate TSE image acquisition by a factor of up to 9.

Analysis of the Effect of Deep-learning Super-resolution for Fragments Detection Performance Enhancement

The Arena Fragmentation Test(AFT) is designed to analyze warhead performance by measuring fragmentation data. In order to evaluate the results of the AFT, a set of AFT images are captured by high-speed cameras. To detect objects in the AFT image set, ResNet-50 based Faster R-CNN is used as a detection model. However, because of the low resolution of the AFT image set, a detection model has shown low performance. To enhance the performance of the detection model, Super-resolution(SR) methods are used to increase the AFT image set resolution. To this end, The Bicubic method and three SR models: ZSSR, EDSR, and SwinIR are used. The use of SR images results in an increase in the performance of the detection model. While the increase in the number of pixels representing a fragment flame in the AFT images improves the Recall performance of the detection model, the number of pixels representing noise also increases, leading to a slight decreases in Precision performance. Consequently, the F1 score is increased by up to 9 %, demonstrating the effectiveness of SR in enhancing the performance of the detection model.

Improved Double Regression Nonlinear Image Super Resolution Model

Abstract The existing super resolution reconstruction methods are mainly divided into traditional super resolution reconstruction and deep learning super resolution reconstruction. The main problem faced by traditional super resolution reconstruction algorithms, such as image enlargement and space transformation, is how to establish the mapping relationship between the input image and the target image, and express the pixel value of the target image through the mapping relationship. As a prominent problem, the difficulty of super resolution reconstruction lies in the fact that there is no realizable matrix relationship between one - to - many mapping relationships. Based on the U-Net network framework, this paper improves the jump-connected modules. By using the combination of convolutional layer, activation layer and residual channel block, the overall module operation efficiency is increased by 2.4%, the overall PNSR is increased by 0.49db, and the running speed is increased by 0.3ms on average when processing a single image compared with other classical models.

Single-frame deep-learning super-resolution microscopy for intracellular dynamics imaging

Single-molecule localization microscopy (SMLM) can be used to resolve subcellular structures and achieve a tenfold improvement in spatial resolution compared to that obtained by conventional fluorescence microscopy. However, the separation of single-molecule fluorescence events that requires thousands of frames dramatically increases the image acquisition time and phototoxicity, impeding the observation of instantaneous intracellular dynamics. Here we develop a deep-learning based single-frame super-resolution microscopy (SFSRM) method which utilizes a subpixel edge map and a multicomponent optimization strategy to guide the neural network to reconstruct a super-resolution image from a single frame of a diffraction-limited image. Under a tolerable signal density and an affordable signal-to-noise ratio, SFSRM enables high-fidelity live-cell imaging with spatiotemporal resolutions of 30 nm and 10 ms, allowing for prolonged monitoring of subcellular dynamics such as interplays between mitochondria and endoplasmic reticulum, the vesicle transport along microtubules, and the endosome fusion and fission. Moreover, its adaptability to different microscopes and spectra makes it a useful tool for various imaging systems.

Prostatic urinary tract visualization with super-resolution deep learning models.

In urethra-sparing radiation therapy, prostatic urinary tract visualization is important in decreasing the urinary side effect. A methodology has been developed to visualize the prostatic urinary tract using post-urination magnetic resonance imaging (PU-MRI) without a urethral catheter. This study investigated whether the combination of PU-MRI and super-resolution (SR) deep learning models improves the visibility of the prostatic urinary tract. We enrolled 30 patients who had previously undergone real-time-image-gated spot scanning proton therapy by insertion of fiducial markers. PU-MRI was performed using a non-contrast high-resolution two-dimensional T2-weighted turbo spin-echo imaging sequence. Four different SR deep learning models were used: the enhanced deep SR network (EDSR), widely activated SR network (WDSR), SR generative adversarial network (SRGAN), and residual dense network (RDN). The complex wavelet structural similarity index measure (CW-SSIM) was used to quantitatively assess the performance of the proposed SR images compared to PU-MRI. Two radiation oncologists used a 1-to-5 scale to subjectively evaluate the visibility of the prostatic urinary tract. Cohen's weighted kappa (k) was used as a measure of agreement of inter-operator reliability. The mean CW-SSIM in EDSR, WDSR, SRGAN, and RDN was 99.86%, 99.89%, 99.30%, and 99.67%, respectively. The mean prostatic urinary tract visibility scores of the radiation oncologists were 3.70 and 3.53 for PU-MRI (k = 0.93), 3.67 and 2.70 for EDSR (k = 0.89), 3.70 and 2.73 for WDSR (k = 0.88), 3.67 and 2.73 for SRGAN (k = 0.88), and 4.37 and 3.73 for RDN (k = 0.93), respectively. The results suggest that SR images using RDN are similar to the original images, and the SR deep learning models subjectively improve the visibility of the prostatic urinary tract.

Deep learning super-resolution for the reconstruction of full wavefield of Lamb waves

Scanning laser Doppler vibrometer is a popular tool for the acquisition of the full wavefield of propagating guided waves, in particular Lamb waves. Signal processing of such a wavefield enables us to reveal damage in the inspected structure. However, the process of acquiring the full wavefield of guided waves is time-consuming. One possible solution to tackle this problem is to acquire the Lamb waves in a low-resolution form and then apply a compressive sensing (CS) or deep learning-based super-resolution approach to that low-resolution form of full wavefield data. In this work, we applied two deep learning-based super-resolution approaches on a large synthetic low-resolution dataset of full wavefields of propagating Lamb waves in a plate of CFRP. The developed deep learning approaches are elaborated, and the results obtained from both models are compared with the conventional CS approach. Furthermore, the validation of the proposed deep learning models in a real-world scenario is performed by an experiment on a CFRP plate with embedded Teflon inserts simulating delaminations. It is concluded that the performance of both models outperforms the conventional CS approach.

Image super‐resolution based on self‐similarity generative adversarial networks

Self-attention has been successfully leveraged for long-range feature-wise similarities in deep learning super-resolution (SR) methods. However, most of the SR methods only explore the features on the original scale, but do not take full advantage of self-similarities features on different scales especially in generative adversarial networks (GAN). In this paper, self-similarity generative adversarial networks (SSGAN) are proposed as the SR framework. The framework establishes the multi-scale feature correlation by adding two modules to the generative network: downscale attention block (DAB) and upscale attention block (UAB). Specifically, DAB is designed to restore the repetitive details from the corresponding downsampled image, which achieves multi-scale feature restoration through self-similarity. And UAB improves the baseline up-sampling operations and captures low-resolution to high-resolution feature mapping, which enhances the cross-scale repetitive features to reconstruct the high-resolution image. Experimental results demonstrate that the proposed SSGAN achieve better visual performance especially in the similar pattern details.

Paired and Unpaired Deep Learning Methods for Physically Accurate Super-Resolution Carbonate Rock Images

X-ray micro-computed tomography (micro-CT) has been widely leveraged to characterise the pore-scale geometry of subsurface porous rocks. Recent developments in super-resolution (SR) methods using deep learning allow for the digital enhancement of low-resolution (LR) images over large spatial scales, creating SR images comparable to high-resolution (HR) ground truth images. This circumvents the common trade-off between resolution and field-of-view. An outstanding issue is the use of paired LR and HR data, which is often required in the training step of such methods but is difficult to obtain. In this work, we rigorously compare two state-of-the-art SR deep learning techniques, using both paired and unpaired data, with like-for-like ground truth data. The first approach requires paired images to train a convolutional neural network (CNN), while the second approach uses unpaired images to train a generative adversarial network (GAN). The two approaches are compared using a micro-CT carbonate rock sample with complicated micro-porous textures. We implemented various image-based and numerical verifications and experimental validation to quantitatively evaluate the physical accuracy and sensitivities of the two methods. Our quantitative results show that the unpaired GAN approach can reconstruct super-resolution images as precise as the paired CNN method, with comparable training times and dataset requirements. This unlocks new applications for micro-CT image enhancement using unpaired deep learning methods; image registration is no longer needed during the data processing stage. Decoupled images from data storage platforms can be exploited to train networks for SR digital rock applications. This opens up a new pathway for various applications related to multi-scale flow simulations in heterogeneous porous media.

CONTRIBUTION OF SUPER RESOLUTION TO 3D RECONSTRUCTION FROM PAIRS OF SATELLITE IMAGES

Abstract. The photogrammetric 3D stereo reconstruction from pairs of strereo images is rising interest in the past few years in space field downstream. Nowadays, it is conceivable that a large production of DSMs from satellite images can become the primary source of 3D information on a global scale. However, in urban areas, DSMs produced with current technology suffer from poor quality. Indeed, even using very high resolution (VHR) images, there is too little information to generate disparity maps that reproduce very well defined shaped objects such as buildings.To address this issue, one solution may be to artificially increase image resolution beyond the sensor limits. Super resolution (SR) algorithms are designed to recover high frequencies, introducing significant information in a scene characterized by strong and frequent discontinuities such as a city. State-of-the-art methods relying on Deep Learning have shown remarkable results in this sense. The aim of this work is therefore to assess the contribution of single image SR Deep Learning techniques to the stereo matching and DSMs generation in an urban context, highlighting potential advantages and limitations that can show up when introducing such a technology in a multi-view stereo pipeline. The proposed contributions are: a methodology for super resolution of VHR data that takes into account realistic simulation of a satellite product; a testbed for the evaluation of the impact of super resolution on 3D photogrammetric reconstruction; a local analysis of the consequences of deep learning SR of VHR images on stereo matching.

Medical Radiation Exposure Reduction in PET via Super-Resolution Deep Learning Model.

In positron emission tomography (PET) imaging, image quality correlates with the injected [18F]-fluorodeoxyglucose (FDG) dose and acquisition time. If image quality improves from short-acquisition PET images via the super-resolution (SR) deep learning technique, it is possible to reduce the injected FDG dose. Therefore, the aim of this study was to clarify whether the SR deep learning technique could improve the image quality of the 50%-acquisition-time image to the level of that of the 100%-acquisition-time image. One-hundred-and-eight adult patients were enrolled in this retrospective observational study. The supervised data were divided into nine subsets for nested cross-validation. The mean peak signal-to-noise ratio and structural similarity in the SR-PET image were 31.3 dB and 0.931, respectively. The mean opinion scores of the 50% PET image, SR-PET image, and 100% PET image were 3.41, 3.96, and 4.23 for the lung level, 3.31, 3.80, and 4.27 for the liver level, and 3.08, 3.67, and 3.94 for the bowel level, respectively. Thus, the SR-PET image was more similar to the 100% PET image and subjectively improved the image quality, as compared to the 50% PET image. The use of the SR deep-learning technique can reduce the injected FDG dose and thus lower radiation exposure.

Isotropic super-resolution light-sheet microscopy of dynamic intracellular structures at subsecond timescales.

Long-term visualization of the dynamic interactions between intracellular structures throughout the three-dimensional space of whole live cells is essential to better understand their functions, but this task remains challenging due to the limitations of existing three-dimensional fluorescence microscopy techniques, such as an insufficient axial resolution, low volumetric imaging rate and photobleaching. Here, we present the combination of a progressive deep-learning super-resolution strategy with a double-ring-modulated selective plane illumination microscopy design capable of visualizing the dynamics of intracellular structures in live cells for hours at an isotropic spatial resolution of roughly 100 nm in three dimensions at speeds up to roughly 17 Hz. Using this approach, we reveal the complex spatial relationships and interactions between endoplasmic reticulum (ER) and mitochondria throughout live cells, providing new insights into ER-mediated mitochondrial division. We also examined the motion of Drp1 oligomers involved in mitochondrial fission and revealed the dynamic interactions between Drp1 and mitochondria in three dimensions.

Downscaling MODIS spectral bands using deep learning

ABSTRACT MODIS sensors are widely used in a broad range of environmental studies, many of which involve joint analysis of multiple MODIS spectral bands acquired at disparate spatial resolutions. To extract land surface information from multi-resolution MODIS spectral bands, existing studies often downscale lower resolution (LR) bands to match the higher resolution (HR) bands based on simple interpolation or more advanced statistical modeling. Statistical downscaling methods rely on the functional relationship between the LR spectral bands and HR spatial information, which may vary across different land surface types, making statistical downscaling methods less robust. In this paper, we propose an alternative approach based on deep learning to downscale 500 m and 1000 m spectral bands of MODIS to 250 m without additional spatial information. We employ a superresolution architecture based on an encoder decoder network. This deep learning-based method uses a custom loss function and a self-attention layer to preserve local and global spatial relationships of the predictions. We compare our approach with a statistical method specifically developed for downscaling MODIS spectral bands, an interpolation method widely used for downscaling multi-resolution spectral bands, and a deep learning superresolution architecture previously used for downscaling satellite imagery. Results show that our deep learning method outperforms on almost all spectral bands both quantitatively and qualitatively. In particular, our deep learning-based method performs very well on the thermal bands due to the larger scale difference between the input and target resolution. This study demonstrates that our proposed deep learning-based downscaling method can maintain the spatial and spectral fidelity of satellite images and contribute to the integration and enhancement of multi-resolution satellite imagery.

Analysis of a Deep Learning-Based Superresolution Algorithm Tailored to Partial Fourier Gradient Echo Sequences of the Abdomen at 1.5 T: Reduction of Breath-Hold Time and Improvement of Image Quality.

The aim of this study was to investigate the feasibility and impact of a novel deep learning superresolution algorithm tailored to partial Fourier allowing retrospectively theoretical acquisition time reduction in 1.5 T T1-weighted gradient echo imaging of the abdomen. Fifty consecutive patients who underwent a 1.5 T contrast-enhanced magnetic resonance imaging examination of the abdomen between April and May 2021 were included in this retrospective study. After acquisition of a conventional T1-weighted volumetric interpolated breath-hold examination using Dixon for water-fat separation (VIBEStd), the acquired data were reprocessed including a superresolution algorithm that was optimized for partial Fourier acquisitions (VIBESR). To accelerate theoretically the acquisition process, a more aggressive partial Fourier setting was applied in VIBESR reconstructions practically corresponding to a shorter acquisition for the data included in the retrospective reconstruction. Precontrast, dynamic contrast-enhanced, and postcontrast data sets were processed. Image analysis was performed by 2 radiologists independently in a blinded random order without access to clinical data regarding the following criteria using a Likert scale ranging from 1 to 4 with 4 being the best: noise levels, sharpness and contrast of vessels, sharpness and contrast of organs and lymph nodes, overall image quality, diagnostic confidence, and lesion conspicuity.Wilcoxon signed rank test for paired data was applied to test for significance. Mean patient age was 61 ± 14 years. Mean acquisition time for the conventional VIBEStd sequence was 15 ± 1 seconds versus theoretical 13 ± 1 seconds of acquired data used for the VIBESR reconstruction. Noise levels were evaluated to be better in VIBESR with a median of 4 (4-4) versus a median of 3 (3-3) in VIBEStd by both readers (P < 0.001). Sharpness and contrast of vessels as well as organs and lymph nodes were also evaluated to be superior in VIBESR compared with VIBEStd with a median of 4 (4-4) versus a median of 3 (3-3) (P < 0.001). Diagnostic confidence was also rated superior in VIBESR with a median of 4 (4-4) versus a median of 3.5 (3-4) in VIBEStd by reader 1 and with a median of 4 (4-4) for VIBESR and a median of 4 (4-4) for VIBEStd by reader 2 (both P < 0.001). Image enhancement using deep learning-based superresolution tailored to partial Fourier acquisitions of T1-weighted gradient echo imaging of the abdomen provides improved image quality and diagnostic confidence in combination with more aggressive partial Fourier settings leading to shorter scan time.

Can the Structure Similarity of Training Patches Affect the Sea Surface Temperature Deep Learning Super-Resolution?

Meso- and fine-scale sea surface temperature (SST) is an essential parameter in oceanographic research. Remote sensing is an efficient way to acquire global SST. However, single infrared-based and microwave-based satellite-derived SST cannot obtain complete coverage and high-resolution SST simultaneously. Deep learning super-resolution (SR) techniques have exhibited the ability to enhance spatial resolution, offering the potential to reconstruct the details of SST fields. Current SR research focuses mainly on improving the structure of the SR model instead of training dataset selection. Different from generating the low-resolution images by downscaling the corresponding high-resolution images, the high- and low-resolution SST are derived from different sensors. Hence, the structure similarity of training patches may affect the SR model training and, consequently, the SST reconstruction. In this study, we first discuss the influence of training dataset selection on SST SR performance, showing that the training dataset determined by the structure similarity index (SSIM) of 0.6 can result in higher reconstruction accuracy and better image quality. In addition, in the practical stage, the spatial similarity between the low-resolution input and the objective high-resolution output is a key factor for SST SR. Moreover, the training dataset obtained from the actual AMSR2 and MODIS SST images is more suitable for SST SR because of the skin and sub-skin temperature difference. Finally, the SST reconstruction accuracies obtained from different SR models are relatively consistent, yet the differences in reconstructed image quality are rather significant.