Structural Similarity Values Research Articles

Dataset augmentation with multiple contrasts images in super-resolution processing of T1-weighted brain magnetic resonance images.

This study investigated the effectiveness of augmenting datasets for super-resolution processing of brain Magnetic Resonance Images (MRI) T1-weighted images (T1WIs) using deep learning. By incorporating images with different contrasts from the same subject, this study sought to improve network performance and assess its impact on image quality metrics, such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). This retrospective study included 240 patients who underwent brain MRI. Two types of datasets were created: the Pure-Dataset group comprising T1WIs and the Mixed-Dataset group comprising T1WIs, T2-weighted images, and fluid-attenuated inversion recovery images. A U-Net-based network and an Enhanced Deep Super-Resolution network (EDSR) were trained on these datasets. Objective image quality analysis was performed using PSNR and SSIM. Statistical analyses, including paired t test and Pearson's correlation coefficient, were conducted to evaluate the results. Augmenting datasets with images of different contrasts significantly improved training accuracy as the dataset size increased. PSNR values ranged 29.84-30.26dB for U-Net trained on mixed datasets, and SSIM values ranged 0.9858-0.9868. Similarly, PSNR values ranged 32.34-32.64dB for EDSR trained on mixed datasets, and SSIM values ranged 0.9941-0.9945. Significant differences in PSNR and SSIM were observed between models trained on pure and mixed datasets. Pearson's correlation coefficient indicated a strong positive correlation between dataset size and image quality metrics. Using diverse image data obtained from the same subject can improve the performance of deep-learning models in medical image super-resolution tasks.

Eliminating metal artifacts in dental computed tomography using an elaborate sinogram normalization interpolation method with CNR-based metal segmentation

Metal artifact reduction (MAR) in dental cone-beam computed tomography (CBCT) is critical for improving its clinical usefulness and remains a challenging problem. This study presents an elaborate sinogram normalization interpolation (SNI) method with contrast-to-noise ratio (CNR)-based metal segmentation to eliminate metal artifacts in dental CBCT. The proposed MAR method involves three main steps: (1) CNR-based segmentation of a metal trace in the sinogram domain; (2) generation of a residual artifact-reduced prior sinogram; and (3) sinogram completion using the prior sinogram, followed by filtered-backprojection (FBP)-based CBCT reconstruction and metal insertion processes. To verify the efficacy of the proposed method, simulations and experiments were performed using numerical and physical dental phantoms with several metal inserts. The quality of the resulting CBCT images was quantitatively evaluated using the structural similarity (SSIM) metric and compared to those obtained with other interpolation-based MAR methods. A tabletop setup for the micro-CT system was used in the experiment, which comprised an X-ray tube operated under tube conditions of 70 kV p and 5 mA and a flat-panel detector with a pixel size of 198 μm. Our results indicate that the CNR-based metal segmentation method precisely identified traces of metallic objects on the sinogram, and thereby, the proposed SNI-based MAR method considerably reduced metal artifacts in dental CBCT images without introducing any contrast anomalies. The SSIM values of the CBCT images obtained with the proposed MAR method were 0.99 and 0.95 for the simulation and experiment, respectively, which are approximately 11% and 9% greater than those with a simple threshold-based linear interpolation method, respectively, improving the image quality. The proposed MAR method can be applied to reduce metal artifacts in real-world dental CBCT systems.

Image Restoration Technology of Tang Dynasty Tomb Murals Using Adversarial Edge Learning

The city of Xi’an, China houses a vast collection of valuable Tang dynasty (618 AD-907 AD) tomb murals that have experienced various degrees of damage over time due to weathering. This research proposes an adversarial edge learning-based mural restoration technique that provides a fast and accurate automatic repair of the murals. The method uses generative adversarial networks to learn how to repair the edge contour of a damaged mural and restore the missing content. The contour restoration network and mural restoration networks are trained and updated simultaneously to enhance the feature extraction and restoration coordination performance of the network. The experiments were conducted on a self-built Tang dynasty tomb mural painting restoration dataset and compared to several existing algorithms. The proposed algorithm showed a 1.3 increase in peak signal-to-noise ratio (PSNR) and a 1.43% increase in structural similarity (SSIM) value, providing a precise restoration of the structure and contour of the mural content. The algorithm successfully repaired damaged murals with fractures and small missing parts. The suggested method can be useful for both the technical repair and conservation of significant national visual cultural property and the digital restoration of historic murals.

Multiobjective optimization guided by image quality index for limited-angle CT image reconstruction.

Due to the incomplete projection data collected by limited-angle computed tomography (CT), severe artifacts are present in the reconstructed image. Classical regularization methods such as total variation (TV) minimization, ℓ0 minimization, are unable to suppress artifacts at the edges perfectly. Most existing regularization methods are single-objective optimization approaches, stemming from scalarization methods for multiobjective optimization problems (MOP). To further suppress the artifacts and effectively preserve the edge structures of the reconstructed image. This study presents a multiobjective optimization model incorporates both data fidelity term and ℓ0-norm of the image gradient as objective functions. It employs an iterative approach different from traditional scalarization methods, using the maximization of structural similarity (SSIM) values to guide optimization rather than minimizing the objective function.The iterative method involves two steps, firstly, simultaneous algebraic reconstruction technique (SART) optimizes the data fidelity term using SSIM and the Simulated Annealing (SA) algorithm for guidance. The degradation solution is accepted in the form of probability, and guided image filtering (GIF) is introduced to further preserve the image edge when the degradation solution is rejected. Secondly, the result from the first step is integrated into the second objective function as a constraint, we use ℓ0 minimization to optimize ℓ0-norm of the image gradient, and the SSIM, SA algorithm and GIF are introduced to guide optimization process by improving SSIM value like the first step. With visual inspection, the peak signal-to-noise ratio (PSNR), root mean square error (RMSE), and SSIM values indicate that our approach outperforms other traditional methods. The experiments demonstrate the effectiveness of our method and its superiority over other classical methods in artifact suppression and edge detail restoration.

Synthesising two-dimensional mammographic images using compressed sensing-reconstructed digital breast tomosynthesis images

This study presents a method synthesizing a two-dimensional (2D) mammographic image of reasonable quality directly from a 3D digital breast tomosynthesis (DBT) image, reconstructed using an advanced compressed sensing (CS)-based algorithm. This approach aims to reduce the radiation dose required for complementary imaging technologies of digital mammography (DM) and DBT, eliminating the need for additional DM examinations. The method involves three main steps: projection data acquisition from a DBT system, CS-based DBT reconstruction, and synthesis of a 2D mammographic image from the reconstructed DBT image. To verify the efficacy of the proposed method, we conducted both a simulation and an experiment on a numerical breast and commercially available BR3D phantoms, respectively, prior to practical implementation in real-world DBT systems. Our simulation and experimental results indicated that the CS-based algorithm yielded markedly improved DBT reconstruction quality, preserving superior image homogeneity, better edge contour and sharpening, and fewer image artifacts. The measured contrast-to-noise ratio and structural similarity values of the CS-reconstructed DBT images were 10.31 and 0.78, respectively, which were approximately 2.2 and 3.1 times larger, respectively, than those of the filtered-backprojection-reconstructed DBT images in the simulation. The quality of the synthetic mammographic images using the CS-reconstructed DBT images was similar to that of conventional mammographic images obtained using a full dose, indicating the efficacy of the proposed method. Consequently, we successfully reconstructed DBT images of substantially high quality using the CS-based algorithm and synthesised 2D mammographic images of reasonable quality, potentially reducing the radiation dose to patients in the complementary imaging technologies of DM and DBT.

Mask-guided dual-perception generative adversarial network for synthesizing complex maize diseased leaves to augment datasets

In practice, acquiring and annotating data in specialized domains can be costly, thereby constraining the performance and applicability of deep learning. Utilizing generative models to synthesize data proves to be an effective augmentation technique. Therefore, this research proposes a diseased leaf generation pipeline to diversify the maize disease datasets. We introduce the Dual-Perception Cycle-Consistent Generative Adversarial Network (DP-CycleGAN). During training, incorporating our proposed Structure Perception (SP) loss and Texture Perception (TP) loss functions. These losses guide the model's attention areas through activation reconstruction and mask mechanisms, thereby improving the overall perceptual quality of the generated images and the realism of the disease lesions. We constructed a maize leaf mixed disease dataset to simulate the complex conditions of real-world disease occurrence. Experimental results show that the DP-CycleGAN generates higher-quality and more realistic diseased leaf images. Compared to CycleGAN and state-of-the-art method, DP-CycleGAN shows a 29.6% and 15.7% reduction in Fréchet Inception Distance (FID) scores and a 125.3% and 61.5% increase in Structural Similarity (SSIM) values, respectively. Simultaneously, by incorporating synthetic data during training, our approach significantly enhances the performance of the recognition model in scenarios of both data abundance and scarcity, with improvement rates exceeding two times those of existing state-of-the-art methods. This contributes to the application of Artificial Intelligence (AI) in agricultural production practices.

Evaluation of SSIM loss function in RIR generator GANs

This study explores the potential of integrating the structural similarity (SSIM) as a loss function within generative adversarial networks (GANs) to enhance the generation of room impulse responses (RIRs). Neural network-based RIR generators sometimes introduce glitches into the generated RIRs, leading to distortion in the reconstructed signals. In this study, GANs are trained using three different loss functions: Mean Squared Error (MSE), SSIM and a combination of SSIM and MSE. Incorporating SSIM within the loss function improves the quality of generated RIRs and reduces glitches. Empirical findings highlight the effectiveness of GANs trained with the mixed MSE and SSIM loss function, resulting in RIR signals with fewer glitches, lower MSE values, and higher SSIM values. To evaluate the broader applicability of this approach and contribute to available resources, we introduce a novel RIR dataset named GTU-RIR. This dataset is presented alongside existing datasets such as BUT ReverbDB, enabling a comprehensive evaluation of the proposed method.

SR-RDFAN-LOG: Arbitrary-scale logging image super-resolution reconstruction based on residual dense feature aggregation

In the process of oil and gas exploration and development, ultrasonic logging imaging technology visually displays the lithology, structure, fractures, pores, and other characteristics of the wellbore. This technology has been widely applied in the exploration and development of oil and gas resources. However, the low circumferential sampling rate negatively affects the imaging quality and resolution during drilling logging. To address the issues of low resolution and blurriness in ultrasonic logging images caused by these factors, as well as the limitation of existing super-resolution reconstruction algorithms that can only reconstruct fixed scales, this paper presents an Image Super-Resolution Reconstruction Network called Residual Dense Feature Aggregation (SR-RDFAN-LOG), which is specifically designed to improve ultrasonic logging image resolution. The proposed network consists of two stages, including encoding and decoding. In the encoding stage, a local feature extraction module is initially designed to integrate contextual features. Subsequently, a residual dense module is introduced for extracting deep features. Finally, a spatial attention module is constructed to focus on high-frequency details. In the decoding stage, a multilayer perceptron is used to decode the feature maps and achieve super-resolution reconstruction at arbitrary scales. The method is tested on an ultrasonic logging images test set with in-distribution scale factors of × 2 and × 4, and out-of-distribution scale factors of × 8 and × 16. When the scaling factors are × 2 and × 4, compared to the best-performing method, the peak signal-to-noise ratio (PSNR) increased by 0.4892 dB and 0.3849 dB, respectively, the structural similarity (SSIM) values increased by 0.0002 and 0.0012, respectively, and the learned perceptual image patch similarity (LPIPS) decreased by 0.0001 and 0.0008, respectively. Compared to the bicubic interpolation method, the LPIPS decreased by 0.0001 and 0.0008 at scale factors of × 8 and × 16, PSNR increased by 1.9614 dB and 2.1002 dB, respectively, and SSIM values increased by 0.0476 and 0.033, respectively. The LPIPS values decreased by 0.2233 and 0.1070, respectively. The experimental results demonstrate that our method outperforms other mainstream methods, achieves super-resolution reconstruction at arbitrary scale, and provides support for subsequent fracture-cave evaluation and oil and gas exploration.

Ultrasound image denoising autoencoder model based on lightweight attention mechanism.

The presence of noise in medical ultrasound images significantly degrades image quality and affects the accuracy of disease diagnosis. The convolutional neural network-denoising autoencoder (CNN-DAE) model extracts feature information by stacking regularly sized kernels. This results in the loss of texture detail, the over-smoothing of the image, and a lack of generalizability for speckle noise. A lightweight attention denoise-convolutional neural network (LAD-CNN) is proposed in the present study. Two different lightweight attention blocks (i.e., the lightweight channel attention (LCA) block and the lightweight large-kernel attention (LLA) block are concatenated into the downsampling stage and the upsampling stage, respectively. A skip connection is included before the upsampling layer to alleviate the problem of gradient vanishing during backpropagation. The effectiveness of our model was evaluated using both subjective visual effects and objective evaluation metrics. With the highest peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) values at all noise levels, the proposed model outperformed the other models. In the test of brachial plexus ultrasound images, the average PSNR of our model was 0.15 higher at low noise levels and 0.33 higher at high noise levels than the suboptimal model. In the test of fetal ultrasound images, the average PSNR of our model was 0.23 higher at low noise levels and 0.20 higher at high noise levels than the suboptimal model. The statistical analysis showed that the p values were less than 0.05, which indicated a statistically significant difference between our model and the other models. The results of this study suggest that the proposed LAD-CNN model is more efficient in denoising and preserving image details than both conventional denoising algorithms and existing deep-learning algorithms.

Low-dose CT reconstruction based on high-dimensional partial differential equation projection recovery

We propose a low-dose CT reconstruction method using partial differential equation (PDE) denoising under high-dimensional constraints. The projection data were mapped into a high-dimensional space to construct a high-dimensional representation of the data, which were updated by moving the points in the high-dimensional space. The data were denoised using partial differential equations and the CT image was reconstructed using the FBP algorithm. Compared with those by FBP, PWLS-QM and TGV-WLS methods, the relative root mean square error of the Shepp-Logan image reconstructed by the proposed method were reduced by 68.87%, 50.15% and 27.36%, the structural similarity values were increased by 23.50%, 8.83% and 1.62%, and the feature similarity values were increased by 17.30%, 2.71% and 2.82%, respectively. For clinical image reconstruction, the proposed method, as compared with FBP, PWLS-QM and TGV-WLS methods, resulted in reduction of the relative root mean square error by 42.09%, 31.04% and 21.93%, increased the structural similarity values by 18.33%, 13.45% and 4.63%, and increased the feature similarity values by 3.13%, 1.46% and 1.10%, respectively. The new method can effectively reduce the streak artifacts and noises while maintaining the spatial resolution in reconstructed low-dose CT images.

Generative adversarial network-based synthesis of contrast-enhanced MR images from precontrast images for predicting histological characteristics in breast cancer

Objective. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is a sensitive tool for assessing breast cancer by analyzing tumor blood flow, but it requires gadolinium-based contrast agents, which carry risks such as brain retention and astrocyte migration. Contrast-free MRI is thus preferable for patients with renal impairment or who are pregnant. This study aimed to investigate the feasibility of generating contrast-enhanced MR images from precontrast images and to evaluate the potential use of synthetic images in diagnosing breast cancer. Approach. This retrospective study included 322 women with invasive breast cancer who underwent preoperative DCE-MRI. A generative adversarial network (GAN) based postcontrast image synthesis (GANPIS) model with perceptual loss was proposed to generate contrast-enhanced MR images from precontrast images. The quality of the synthesized images was evaluated using the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). The diagnostic performance of the generated images was assessed using a convolutional neural network to predict Ki-67, luminal A and histological grade with the area under the receiver operating characteristic curve (AUC). The patients were divided into training (n = 200), validation (n = 60), and testing sets (n = 62). Main results. Quantitative analysis revealed strong agreement between the generated and real postcontrast images in the test set, with PSNR and SSIM values of 36.210 ± 2.670 and 0.988 ± 0.006, respectively. The generated postcontrast images achieved AUCs of 0.918 ± 0.018, 0.842 ± 0.028 and 0.815 ± 0.019 for predicting the Ki-67 expression level, histological grade, and luminal A subtype, respectively. These results showed a significant improvement compared to the use of precontrast images alone, which achieved AUCs of 0.764 ± 0.031, 0.741 ± 0.035, and 0.797 ± 0.021, respectively. Significance. This study proposed a GAN-based MR image synthesis method for breast cancer that aims to generate postcontrast images from precontrast images, allowing the use of contrast-free images to simulate kinetic features for improved diagnosis.

Target Image Processing based on Super-resolution Reconstruction and Machine Learning Algorithm

This article proposes a target image processing method based on super-resolution reconstruction and machine learning algorithms, which solves the low-resolution problem in medical images during imaging. This method uses nonlocal autoregressive learning based on a medical image super-resolution reconstruction method. The autoregressive model is introduced into the sparse representation-based medical image super-resolution reconstruction model by utilizing medical image data inherent nonlocal similarity characteristics. At the same time, a clustering algorithm is used to obtain a classification dictionary, improving experimental efficiency. The experimental results show that ten randomly selected CT/MR images are used as test images, and each image's peak signal-to-noise ratio and structural similarity values are calculated separately. Compared with other methods, the method proposed in this paper is significantly better and can achieve ideal results, with the highest value being 31.49. This method demonstrates the feasibility of using super-resolution reconstruction and machine learning algorithms in medical image resolution.

Application of Generative Adversarial Network in Image Color Correction

The popularity of electronic products has increased with the development of technology. Electronic devices allow people to obtain information through the transmission of images. However, color distortion can occur during the transmission process, which may hinder the usefulness of the images. To this end, a deep residual network and a deep convolutional network were used to define the generator and discriminator. Then, self-attention-enhanced convolution was applied to the generator network to construct an image resolution correction model based on coupled generative adversarial networks. On this basis, a generative network model integrating multi-scale features and contextual attention mechanism was constructed to achieve image restoration. Finally, performance and image restoration application tests were conducted on the constructed model. The test showed that when the coupled generative adversarial network was tested on the Set5 dataset, the image peak signal-to-noise ratio and image structure similarity values were 31.2575 and 0.8173. On the Set14 dataset, they were 30.8521 and 0.8079, respectively. The multi-scale feature-fusion algorithm was tested on the BSDS100 dataset with an image peak signal-to-noise ratio of 30.2541 and an image structure similarity value of 0.8352. Based on the data presented, it can be concluded that the image correction model constructed in this study has a strong image restoration ability. The reconstructed image has the highest similarity with the real high-resolution image and a low distortion rate. It can achieve the task of repairing problems such as color distortion during image transmission. In addition, this study can provide technical support for similar information correction and restoration work.

Feasibility of improving image performance in photon-counting computed tomography using X-ray spectrum filtration

Photon-counting computed tomography (PCCT) is an emerging technology based on new energy-resolving X-ray detectors (photon-counting detectors (PCDs)) that provide promising image performance compared to the conventional energy-integrating CT (EICT). It has the potential to provide higher resolution and contrast, lower radiation dose, and fewer artifacts, which has led to significant research interest. This study proposes a method to further improve the image performance of the PCCT using an X-ray spectrum filtration technique. We conducted a feasibility study via simulations using aluminum (Al), beryllium (Be), sodium (Na), nickel (Ni), tin (Sn), neodymium (Nd), and tantalum (Ta) incorporating varying thicknesses using a PCD simulation toolkit (PcTK). The PCCT system used in this simulation was modeled to have a cadmium telluride-based PCD with four multi-energy thresholds of E = 40, 60, 80, 100 keV, assuming that the object received the same number of photons. Using the PCCT images obtained with the highest threshold (i.e., E4 = 100 keV), the image quality was evaluated quantitatively in terms of the signal difference-to-noise ratio (SDNR) and structural similarity (SSIM). Among the filtrations selected in this simulation, a filtration of 2.0 mmAl (inherent) and 1.5 mmSn (added) showed the best image quality. The SDNR and SSIM values measured in the PCCT image obtained with an added filtration of 1.5 mmSn were 3.73 and 0.81, approximately 1.9 and 1.8 times higher than those with no added filtration, respectively. Consequently, the X-ray spectrum filtration technique in PCCT is useful to further improve image performance.

Improving noise characteristics using a modified image pyramid with guided filtering and Bayesian shrinkage threshold in low-dose animal radiography

Radiography is a commonly used diagnostic tool in animal practice to provide clinical information. The recent market for animal radiography is rapidly growing owing to the increasing awareness of the welfare of companion animals and the expansion of animal clinics. As the clinical use of radiography in animal hospitals has grown rapidly, radiologists continuously seek ways to reduce the radiation dose to animals. This study proposes an effective noise removal method using a modified image pyramid with guided filtering and Bayesian shrinkage threshold for low-dose animal radiography. This study aims to model a modified image pyramid-based denoising algorithm and to confirm its applicability in an animal radiographic system. We conducted an experiment on a companion dog and cat using a commercially-available veterinary radiographic system and quantitatively evaluated the image quality using the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). The PSNR and SSIM values measured in the denoised image of the dog using the proposed denoising algorithm were 39.54 and 0.96, respectively, which are improvements of approximately 11.3% and 5.5% of the values obtained in the original noisy image. Consequently, the proposed method is highly efficient in reducing image noise in animal radiography, thereby improving the image quality.

Fabric defect image generation method based on the dual-stage W-net generative adversarial network

Due to the intricate and diverse nature of textile defects, detecting them poses an exceptionally challenging task. In comparison with conventional defect detection methods, deep learning-based defect detection methods generally exhibit superior precision. However, utilizing deep learning for defect detection requires a substantial volume of training data, which can be particularly challenging to accumulate for textile flaws. To augment the fabric defect dataset and enhance fabric defect detection accuracy, we propose a fabric defect image generation method based on Pix2Pix generative adversarial network. This approach devises a novel dual-stage W-net generative adversarial network. By increasing the network depth, this model can effectively extract intricate textile image features, thereby enhancing its ability to expand information sharing capacity. The dual-stage W-net generative adversarial network allows generating desired defects on defect-free textile images. We conduct quality assessment of the generated fabric defect images resulting in peak signal-to-noise ratio and structural similarity values exceeding 30 and 0.930, respectively, and a learned perceptual image patch similarity value no greater than 0.085, demonstrating the effectiveness of fabric defect data augmentation. The effectiveness of dual-stage W-net generative adversarial network is established through multiple comparative experiments evaluating the generated images. By examining the detection performance before and after data augmentation, the results demonstrate that mean average precision improves by 6.13% and 14.57% on YOLO V5 and faster recurrent convolutional neural networks detection models, respectively.

Deep learning-based harmonization of trabecular bone microstructures between high- and low-resolution CT imaging.

Osteoporosis is a bone disease related to increased bone loss and fracture-risk. The variability in bone strength is partially explained by bone mineral density (BMD), and the remainder is contributed by bone microstructure. Recently, clinical CT has emerged as a viable option for in vivo bone microstructural imaging. Wide variations in spatial-resolution and other imaging features among different CT scanners add inconsistency to derived bone microstructural metrics, urging the need for harmonization of image data from different scanners. This paper presents a new deep learning (DL) method for the harmonization of bone microstructural images derived from low- and high-resolution CT scanners and evaluates the method's performance at the levels of image data as well as derived microstructural metrics. We generalized a three-dimensional (3D) version of GAN-CIRCLE that applies two generative adversarial networks (GANs) constrained by the identical, residual, and cycle learning ensemble (CIRCLE). Two GAN modules simultaneously learn to map low-resolution CT (LRCT) to high-resolution CT (HRCT) and vice versa. Twenty volunteers were recruited. LRCT and HRCT scans of the distal tibia of their left legs were acquired. Five-hundred pairs of LRCT and HRCT image blocks of voxels were sampled for each of the twelve volunteers and used for training in supervised as well as unsupervised setups. LRCT and HRCT images of the remaining eight volunteers were used for evaluation. LRCT blocks were sampled at 32 voxel intervals in each coordinate direction and predicted HRCT blocks were stitched to generate a predicted HRCT image. Mean±standard deviation of structural similarity (SSIM) values between predicted and true HRCT using both 3DGAN-CIRCLE-based supervised (0.84±0.03) and unsupervised (0.83±0.04) methods were significantly (p<0.001) higher than the mean SSIM value between LRCT and true HRCT (0.75±0.03). All Tb measures derived from predicted HRCT by the supervised 3DGAN-CIRCLE showed higher agreement (CCC [0.956 0.991]) with the reference values from true HRCT as compared to LRCT-derived values (CCC [0.732 0.989]). For all Tb measures, except Tb plate-width (CCC=0.866), the unsupervised 3DGAN-CIRCLE showed high agreement (CCC [0.920 0.964]) with the true HRCT-derived reference measures. Moreover, Bland-Altman plots showed that supervised 3DGAN-CIRCLE predicted HRCT reduces bias and variability in residual values of different Tb measures as compared to LRCT and unsupervised 3DGAN-CIRCLE predicted HRCT. The supervised 3DGAN-CIRCLE method produced significantly improved performance (p<0.001) for all Tb measures as compared to the two DL-based supervised methods available in the literature. 3DGAN-CIRCLE, trained in either unsupervised or supervised fashion, generates HRCT images with high structural similarity to the reference true HRCT images. The supervised 3DGAN-CIRCLE improves agreements of computed Tb microstructural measures with their reference values and outperforms the unsupervised 3DGAN-CIRCLE. 3DGAN-CIRCLE offers a viable DL solution to retrospectively improve image resolution, which may aid in data harmonization in multi-site longitudinal studies where scanner mismatch is unavoidable.

DOVE: Doodled vessel enhancement for photoacoustic angiography super resolution

Deep-learning-based super-resolution photoacoustic angiography (PAA) has emerged as a valuable tool for enhancing the resolution of blood vessel images and aiding in disease diagnosis. However, due to the scarcity of training samples, PAA super-resolution models do not generalize well, especially in the challenging in-vivo imaging of organs with deep tissue penetration. Furthermore, prolonged exposure to high laser intensity during the image acquisition process can lead to tissue damage and secondary infections. To address these challenges, we propose an approach doodled vessel enhancement (DOVE) that utilizes hand-drawn doodles to train a PAA super-resolution model. With a training dataset consisting of only 32 real PAA images, we construct a diffusion model that interprets hand-drawn doodles as low-resolution images. DOVE enables us to generate a large number of realistic PAA images, achieving a 49.375% fool rate, even among experts in photoacoustic imaging. Subsequently, we employ these generated images to train a self-similarity-based model for super-resolution. During cross-domain tests, our method, trained solely on generated images, achieves a structural similarity value of 0.8591, surpassing the scores of all other models trained with real high-resolution images. DOVE successfully overcomes the limitation of insufficient training samples and unlocks the clinic application potential of super-resolution-based biomedical imaging.

A New Two-Stream Structure Preserving Network for X-Ray Enhancement

Generative adversarial network (GAN) is applied in X-ray enhancement, but the tiny structure of X-ray is hard to be maintained by the generator of the GAN. To solve this problem, a new two-stream GAN (named T-S GAN) is proposed to enhance X-rays and guarantee better structure preserving at the same time. In the proposed generator of T-S GAN, the two-stream feature maps are extracted from original and detail X-rays, and then the two-stream features are fused by a coarse-to-fine strategy. Meanwhile, by introducing detail and structure constraints, a mixed loss function is designed for optimizing the T-S GAN. Lastly, to prove the correctness and effectiveness of T-S GAN, a novel dataset of X-ray with finger and wrist fracture is originally created. There are plenty of hairline fractures in the dataset, which can be used to solve the problem of verifying that the tiny structure is preserved by T-S GAN. Experimental results show that the proposed method, compared with the state-of-the-art methods, has lower mean absolute error (MAE), higer peak signal to noise ratio (PSNR) and structure similarity (SSIM) values. A combination of T-S GAN and the state-of-the-art object detection method can be potentially applied for modern medical detection.

KernelFlexSR: a self-adaptive super-resolution algorithm with multi-path convolution and residual network for dynamic kernel enhancement

Machine learning-based image super-resolution (SR) has garnered increasing research interest in recent years. However, there are two issues that have not been adequately addressed. The first issue is that existing SR methods often overlook the importance of improving the quality of the training dataset, which is a crucial factor in determining SR performance, regardless of the training method employed. The second issue is that while some studies report high numerical metrics, the visual results remain unsatisfactory. To address the first problem, we propose a new image down-sampling method to obtain higher-quality training datasets. To tackle the second problem, we present a new image super-resolution model based on a large-size convolution kernel and a multi-path algorithm. Specifically, we use an adaptive large-size convolutional kernel to extract features from the image based on the size of the input image, and a residual network to generate a deeper model to retain more details of the original input image. Experimental results demonstrate that the proposed multilayer downsampling method (MDM) can significantly improve the visual quality compared to traditional downsampling methods. Moreover, our proposed method achieves the best peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) values compared to several typical SR algorithms. Furthermore, subjective evaluation by human observers reveals that our method retains more details of the original image and produces smoother high-resolution images. Our proposed method effectively addresses the two aforementioned issues, which leads to improved SR performance in terms of both quantitative and qualitative measures.