Search

A major concern in data-driven deep learning (DL) is how to maximize the capability of a model for limited datasets. The lack of high-performance datasets limits intelligent agriculture development. Recent studies have shown that image enhancement techniques can alleviate the limitations of datasets on model performance. Existing image enhancement algorithms mainly perform in the same category and generate highly correlated samples. Directly using authentic images to expand the dataset, the environmental noise in the image will seriously affect the model's accuracy. Hence, this paper designs an automatic leaf segmentation algorithm (AISG) based on the EISeg segmentation method, separating the leaf information with disease spot characteristics from the background noise in the picture. This algorithm enhances the network model's ability to extract disease features. In addition, the Cycle-GAN network is used for minor sample data enhancement to realize cross-category image transformation. Then, MobileNet was trained by transfer learning on an enhanced dataset. The experimental results reveal that the proposed method achieves a classification accuracy of 98.61% for the ten types of tomato diseases, surpassing the performance of other existing methods. Our method is beneficial in solving the problems of low accuracy and insufficient training data in tomato disease detection. This method can also provide a reference for the detection of other types of plant diseases. [ABSTRACT FROM AUTHOR]

Computed tomography (CT) imaging is vital for diagnosing and monitoring diseases in both humans and animals, yet radiation exposure remains a significant concern, especially in animal imaging. Low-dose CT (LDCT) minimizes radiation exposure but often compromises image quality due to a reduced signal-to-noise ratio (SNR). Recent advancements in deep learning, particularly with CycleGAN, offer promising solutions for denoising LDCT images, though challenges in preserving anatomical detail and image sharpness persist. This study introduces a novel framework tailored for animal LDCT imaging, integrating deep learning techniques within the CycleGAN architecture. Key components include BlurPool for mitigating high-resolution image distortion, PixelShuffle for enhancing expressiveness, hierarchical feature synthesis (HFS) networks for feature retention, and spatial channel squeeze excitation (scSE) blocks for contrast reproduction. Additionally, a multi-scale discriminator enhances detail assessment, supporting effective adversarial learning. Rigorous experimentation on veterinary CT images demonstrates our framework's superiority over traditional denoising methods, achieving significant improvements in noise reduction, contrast enhancement, and anatomical structure preservation. Extensive evaluations show that our method achieves a precision of 0.93 and a recall of 0.94. This validates our approach's efficacy, highlighting its potential to enhance diagnostic accuracy in veterinary imaging. We confirm the scSE method's critical role in optimizing performance, and robustness to input variations underscores its practical utility. [ABSTRACT FROM AUTHOR]

To propose a style transfer model for multi-contrast magnetic resonance imaging (MRI) images with a cycle-consistent generative adversarial network (CycleGAN) and evaluate the image quality and prognosis prediction performance for glioblastoma (GBM) patients from the extracted radiomics features. Style transfer models of T1 weighted MRI image (T1w) to T2 weighted MRI image (T2w) and T2w to T1w with CycleGAN were constructed using the BraTS dataset. The style transfer model was validated with the Cancer Genome Atlas Glioblastoma Multiforme (TCGA-GBM) dataset. Moreover, imaging features were extracted from real and synthesized images. These features were transformed to rad-scores by the least absolute shrinkage and selection operator (LASSO)-Cox regression. The prognosis performance was estimated by the Kaplan–Meier method. For the accuracy of the image quality of the real and synthesized MRI images, the MI, RMSE, PSNR, and SSIM were 0.991 ± 2.10 × 10 - 4 , 2.79 ± 0.16, 40.16 ± 0.38, and 0.995 ± 2.11 × 10 - 4 , for T2w, and.992 ± 2.63 × 10 - 4 , 2.49 ± 6.89 × 10 - 2 , 40.51 ± 0.22, and 0.993 ± 3.40 × 10 - 4 for T1w, respectively. The survival time had a significant difference between good and poor prognosis groups for both real and synthesized T2w (p < 0.05). However, the survival time had no significant difference between good and poor prognosis groups for both real and synthesized T1w. On the other hand, there was no significant difference between the real and synthesized T2w in both good and poor prognoses. The results of T1w were similar in the point that there was no significant difference between the real and synthesized T1w. It was found that the synthesized image could be used for prognosis prediction. The proposed prognostic model using CycleGAN could reduce the cost and time of image scanning, leading to a promotion to build the patient's outcome prediction with multi-contrast images. [ABSTRACT FROM AUTHOR]

In medical diagnostic systems, the most challenging task is to segment and classify the varieties of skin disorders from dermoscopic images. For this purpose, Bootstrapping of Fine-tuned Segmentation and Classification Network (BF-SegClassNet) model was designed, which uses (i) cycle-Generative Adversarial Network (GAN) as domain adaptation, (ii) modified SegNet as segmentation and (iii) fine-tuned ResNet18 with Bootstrapping as classification. But, the efficiency of cycle-GAN was degraded if the source domain differs largely from the target domain. Hence, in this article, a Fuzzy Transfer Learning (FTL) model is developed based on fuzzy logic as domain adaptation. In this model, 2 different stages are performed such as training and adaptation. During the training stage, the source labeled data is used to build the Fuzzy Inference System (FIS), which extracts information from the source and transfers it to the target domain. The fuzzy sets and fuzzy rules created by an Adhoc Data-Driven Learning (ADDL) activity are included in the FIS. The created source FIS and the target data are used in the adaptation stage to adapt the fuzzy rule and the fuzzy rule base from the FIS to extract dissimilarities in the data and help bridge the contextual gap between the source and target. Thus, this FTL model is applied instead of cycleGAN to create more samples, which are further partitioned and classified by the BF-SegClassNet model efficiently. Finally, the testing outcomes exhibit that the FTL model attains a mean accuracy of 98.08% for the HAM dataset compared to the other GAN models. [ABSTRACT FROM AUTHOR]

A major concern in data-driven deep learning (DL) is how to maximize the capability of a model for limited datasets. The lack of high-performance datasets limits intelligent agriculture development. Recent studies have shown that image enhancement techniques can alleviate the limitations of datasets on model performance. Existing image enhancement algorithms mainly perform in the same category and generate highly correlated samples. Directly using authentic images to expand the dataset, the environmental noise in the image will seriously affect the model’s accuracy. Hence, this paper designs an automatic leaf segmentation algorithm (AISG) based on the EISeg segmentation method, separating the leaf information with disease spot characteristics from the background noise in the picture. This algorithm enhances the network model’s ability to extract disease features. In addition, the Cycle-GAN network is used for minor sample data enhancement to realize cross-category image transformation. Then, MobileNet was trained by transfer learning on an enhanced dataset. The experimental results reveal that the proposed method achieves a classification accuracy of 98.61% for the ten types of tomato diseases, surpassing the performance of other existing methods. Our method is beneficial in solving the problems of low accuracy and insufficient training data in tomato disease detection. This method can also provide a reference for the detection of other types of plant diseases.

Computed tomography (CT) imaging is vital for diagnosing and monitoring diseases in both humans and animals, yet radiation exposure remains a significant concern, especially in animal imaging. Low-dose CT (LDCT) minimizes radiation exposure but often compromises image quality due to a reduced signal-to-noise ratio (SNR). Recent advancements in deep learning, particularly with CycleGAN, offer promising solutions for denoising LDCT images, though challenges in preserving anatomical detail and image sharpness persist. This study introduces a novel framework tailored for animal LDCT imaging, integrating deep learning techniques within the CycleGAN architecture. Key components include BlurPool for mitigating high-resolution image distortion, PixelShuffle for enhancing expressiveness, hierarchical feature synthesis (HFS) networks for feature retention, and spatial channel squeeze excitation (scSE) blocks for contrast reproduction. Additionally, a multi-scale discriminator enhances detail assessment, supporting effective adversarial learning. Rigorous experimentation on veterinary CT images demonstrates our framework’s superiority over traditional denoising methods, achieving significant improvements in noise reduction, contrast enhancement, and anatomical structure preservation. Extensive evaluations show that our method achieves a precision of 0.93 and a recall of 0.94. This validates our approach’s efficacy, highlighting its potential to enhance diagnostic accuracy in veterinary imaging. We confirm the scSE method’s critical role in optimizing performance, and robustness to input variations underscores its practical utility.

The fault diagnosis method based on domain adaptation is a hot topic in recent years. It is difficult to collect a complete data set containing all fault categories in practice under the same working condition, leading to fault categories knowledge loss in the single source domain. To resolve the problem, a domain adaptation method for bearing fault diagnosis using multiple incomplete source data is proposed in this study. First, the cycle generative adversarial network is used to learn the mapping between multi-source domains to complement the missing category data. Then, considering the domain mismatch problem, a multi-source domain adaption model based on anchor adapters is developed to obtain general domain invariant diagnosis knowledge. Finally, the fault diagnosis model is established by an ensemble of multi-classifier results. Extensive experiments on bearing data sets demonstrate that the proposed method in fault diagnosis with multiple incomplete source data is effective and has a good diagnosis performance. [ABSTRACT FROM AUTHOR]

Machine learning algorithm has been applied to shear wave velocity (Vs) inversion in surface wave tomography, where a set of starting 1-D Vs profiles and their corresponding synthetic dispersion curves are used in network training. Previous studies showed that the performance of such trained network is dependent on the diversity of the training data set, which limits its application to previously poorly understood regions. Here, we present an improved semi-supervised algorithm-based network that takes both model-generated and observed surface wave dispersion data in the training process. The algorithm is termed Wasserstein cycle-consistent generative adversarial networks (Wasserstein Cycle-GAN [Wcycle-GAN]). Different from conventional supervised approaches, the GAN architecture enables the inclusion of unlabeled data (the observed surface wave dispersion) in the training process that can complement the model-generated data set. The cycle-consistency and Wasserstein metric significantly improve the training stability of the proposed algorithm. We benchmark the Wcycle-GAN method using 4,076 pairs of fundamental mode Rayleigh wave phase and group velocity dispersion curves derived in periods from 3 to 16 s in Southern California. The final 3-D Vs model given by the best trained network shows large-scale features consistent with the surface geology. The resulting Vs model has reasonable data misfits and provides sharper images of structures near faults in the top 15 km compared with those from conventional machine learning methods.

Vegetable diseases and pests continue to be a major threat to food security. Manual identification is time-consuming and labor-intensive due to the wide variety of diseases and pests, and it is also prone to errors. In this paper, a novel and integrated deep learning framework based on convolutional neural networks was proposed for leek disease identification using an improved Cycle-GAN network, gated recurrent units, and recurrent convolutional auto-encoders. We collected healthy and diseased leek leaves under realistic conditions for the experimental implementation. We concentrated on the identification of five leek diseases: virus, rust, blight, sclerotium rolfsii, and botrytis cinerea. It is discovered that the proposed method's evaluation metrics produce an accuracy of 99.58%, a specificity of 99.62%, a sensitivity of 97.96%, a precision of 99.77%, and an f1-score of 98.19%, which is significantly higher than other cutting-edge deep learning models such as ShuffleNet-V3, Inception-V4, and Efficient-B7. Many aspects of the proposed model implementation provide competitive advantages, such as high recognition precision, strong adaptive ability, and good generalization. In conclusion, the proposed system can accurately classify leek diseases, which is very beneficial in increasing the efficiency of plant pathology image analysis. [ABSTRACT FROM AUTHOR]

Background and Purpose: Addressing the need for accurate dose calculation in MRI-only radiotherapy, the generation of synthetic Computed Tomography (sCT) from MRI has emerged. Deep learning (DL) techniques, have shown promising results in achieving high sCT accuracies. However, existing sCT synthesis methods are often center-specific, posing a challenge to their generalizability. To overcome this limitation, recent studies have proposed approaches, such as multicenter training . Material and methods: The purpose of this work was to propose a multicenter sCT synthesis by DL, using a 2D cycle-GAN on 128 prostate cancer patients, from four different centers. Four cases were compared: monocenter cases, monocenter training and test on another center, multicenter trainings and a test on a center not included in the training and multicenter trainings with an included center in the test. Trainings were performed using 20 patients. sCT accuracy evaluation was performed using Mean Absolute Error, Mean Error and Peak-Signal-to-Noise-Ratio. Dose accuracy was assessed with gamma index and Dose Volume Histogram comparison. Results: Qualitative, quantitative and dose results show that the accuracy of sCTs for monocenter trainings and multicenter trainings using a seen center in the test did not differ significantly. However, when the test involved an unseen center, the sCT quality was inferior. Conclusions: The aim of this work was to propose generalizable multicenter training for MR-to-CT synthesis. It was shown that only a few data from one center included in the training cohort allows sCT accuracy equivalent to a monocenter study.

In the realm of computer vision, image transformations play a pivotal role across various domains such as healthcare, image enhancement, artist painting identification, genome sequencing, and more. While supervised learning demands a substantial volume of annotated images for training models, Cycle-GAN emerges as a potent solution for training models with fewer paired sources and target images in an unsupervised manner. This study introduces a novel system aimed at generating Monet-style paintings from realistic images, leveraging the Cycle-GAN methodology. Given the scarcity of Monet paintings, our system employs a combination of generator and discriminator neural networks to produce new Monet-style artworks. The model is trained using Cycle-GAN in conjunction with deep learning optimizers like RMSprop, ADAM, and SGD. The training dataset, Monet2Photo, comprises two distinct image categories: Monet paintings (300 samples) and natural photographs (7028 samples). The Monet-style images are utilized for training the model, while the raw photo images serve as the test set. Notably, the proposed model exhibits commendable performance, particularly when utilizing the SGD optimizer, as evidenced by favorable outcomes in terms of generator and discriminator losses.

Abstract The two main research threads in computer-based music generation are the construction of autonomous music-making systems and the design of computer-based environments to assist musicians. In the symbolic domain, the key problem of automatically arranging a piece of music was extensively studied, while relatively fewer systems tackled this challenge in the audio domain. In this contribution, we propose CycleDRUMS, a novel method for generating drums given a bass line. After converting the waveform of the bass into a mel-spectrogram, we can automatically generate original drums that follow the beat, sound credible, and be directly mixed with the input bass. We formulated this task as an unpaired image-to-image translation problem, and we addressed it with CycleGAN, a well-established unsupervised style transfer framework designed initially for treating images. The choice to deploy raw audio and mel-spectrograms enabled us to represent better how humans perceive music and to draw sounds for new arrangements from the vast collection of music recordings accumulated in the last century. In the absence of an objective way of evaluating the output of both generative adversarial networks and generative music systems, we further defined a possible metric for the proposed task, partially based on human (and expert) judgment. Finally, as a comparison, we replicated our results with Pix2Pix, a paired image-to-image translation network, and we showed that our approach outperforms it.

Image dehazing is a fundamental problem in computer vision. However, GT images for supervised dehazing network training are virtually impossible to obtain in the real world. Therefore, unsupervised image dehazing is of great significance. In this paper, a Cycle Generative Adversarial Network (Cycle-GAN) based on the Homology Isomerism Discriminator (HID)-assisted Detail Generation Module (DGM) is proposed to achieve image dehazing under unsupervised training. In order to enable the generator to recover the details of the entire image, especially in images with complex structures, DGM is developed to boost the details performance of the output dehazing results. Then, HID is proposed to boost dehazing performance based on heterogeneous features and Channel-Spatial Attention (CSA) and complement DGM by varied guidance to the generator. Next, the loss functions are updated according to the image dehazing task based on Cycle-GAN under unsupervised learning. Finally, the experimental results under ablation study and comparison to state-of-the-art demonstrate that the proposed method has attractive results in both visual experience and quantitative metrics under various datasets. [ABSTRACT FROM AUTHOR]

Facial recognition technology and related research have matured over time, but research in the field of individual animal recognition is still very limited. Therefore, this article focuses on the identification of cashmere goats with similar characteristics. First, the single shot multibox detector network was used to process the dataset. Next, transfer learning was applied to learn the characteristics of the goats, as well as the loss function is composed of Triplet Loss and Label Smoothing CrossEntropy Loss function. The result of Label Smoothing CrossEntropy Loss function is fused by multiple different branches, which is convenient for classification. We added a small number of images of 24 different breeds of sheep to each cashmere goat dataset with different ID to promote the distance between training individuals, and then used the trained model to find the number of goats with the lowest recognition accuracy. The Cycle‐Consistent Adversarial Network (Cycle‐GAN) learned the goat dataset with a high error rate in individual identification. Unlike previous studies using the Cycle‐GAN, we took the novel approach of using this network to learn and combine the features seen in photos of cashmere goats. Since the learned features were all observed in the same goats, this method achieved better results in learning the features of the goats. Finally, we found that recognition can be performed on this data with an accuracy of 93.75%. These results suggest that identification based on deep learning has a high accuracy rate, as well as great value in identifying individual cashmere goats. [ABSTRACT FROM AUTHOR]

The two main research threads in computer-based music generation are the construction of autonomous music-making systems and the design of computer-based environments to assist musicians. In the symbolic domain, the key problem of automatically arranging a piece of music was extensively studied, while relatively fewer systems tackled this challenge in the audio domain. In this contribution, we propose CycleDRUMS, a novel method for generating drums given a bass line. After converting the waveform of the bass into a mel-spectrogram, we can automatically generate original drums that follow the beat, sound credible, and be directly mixed with the input bass. We formulated this task as an unpaired image-to-image translation problem, and we addressed it with CycleGAN, a well-established unsupervised style transfer framework designed initially for treating images. The choice to deploy raw audio and mel-spectrograms enabled us to represent better how humans perceive music and to draw sounds for new arrangements from the vast collection of music recordings accumulated in the last century. In the absence of an objective way of evaluating the output of both generative adversarial networks and generative music systems, we further defined a possible metric for the proposed task, partially based on human (and expert) judgment. Finally, as a comparison, we replicated our results with Pix2Pix, a paired image-to-image translation network, and we showed that our approach outperforms it. [ABSTRACT FROM AUTHOR]

As an active microwave coherent imaging technology, synthetic aperture radar (SAR) images suffer from severe speckle noise and low-resolution problems due to the limitations of the imaging system, which cause difficulties in image interpretation and target detection. However, the existing SAR super-resolution (SR) methods usually reconstruct the images by a determined degradation model and hardly consider multiplicative speckle noise, meanwhile, most SR models are trained with synthetic datasets in which the low-resolution (LR) images are down-sampled from their high-resolution (HR) counterparts. These constraints cause a serious domain gap between the synthetic and real SAR images. To solve the above problems, this paper proposes an unsupervised blind SR method for SAR images by introducing SAR priors in a cycle-GAN framework. First, a learnable probabilistic degradation model combined with SAR noise priors was presented to satisfy various SAR images produced from different platforms. Then, a degradation model and a SR model in a unified cycle-GAN framework were trained simultaneously to learn the intrinsic relationship between HR–LR domains. The model was trained with real LR and HR SAR images instead of synthetic paired images to conquer the domain gap. Finally, experimental results on both synthetic and real SAR images demonstrated the high performance of the proposed method in terms of image quality and visual perception. Additionally, we found the proposed SR method demonstrates the tremendous potential for target detection tasks by reducing missed detection and false alarms significantly. [ABSTRACT FROM AUTHOR]

Abstract Background Coronary computed tomography angiography examinations are increasingly becoming established as a minimally invasive method for diagnosing coronary diseases. However, although various imaging and processing methods have been developed, coronary artery calcification remains a major limitation in the evaluation of the vascular lumen. Subtraction coronary computed tomography angiography (Sub-CCTA) is a method known to be able to reduce the influence of coronary artery calcification and is therefore feasible for improving the diagnosis of significant stenosis in patients with severe calcification. However, Sub-CCTA still involves some problems, such as the increased radiation dose due to plain (mask) imaging, extended breath-holding time, and misregistration due to differences in the imaging phase. Therefore, we considered using artificial intelligence instead of Sub-CCTA to visualize the coronary lumen with high calcification. Given this background, the present study aimed to evaluate the diagnostic performance of a deep learning-based lumen extraction method (DL-LEM) to detect significant stenosis on CCTA in 99 consecutive patients (891 segments) with severe coronary calcification from November 2015 to March 2018. We also estimated the impact of DL-LEM on the medical economics in Japan. Results The DL-LEM slightly improved the per-segment diagnostic accuracy from 74.5 to 76.4%, and the area under the curve (AUC) slightly improved from 0.752 to 0.767 (p = 0.030). When analyzing the 228 segments that could not be evaluated because of severe calcification on the original CCTA images, the DL-LEM improved the accuracy from 35.5 to 42.5%, and the AUC improved from 0.500 to 0.587 (p = 0.00018). As a result, DL-LEM analysis could have avoided invasive coronary angiography in 4/99 cases (per patient). From the calculated results, it was estimated that the number of exams that can be avoided in Japan in one year is approximately 747 for invasive coronary angiography, 219 for fractional flow reserve, and 248 for nuclear exam. The total amount of medical fee that could be reduced was 225,629,368 JPY. Conclusions These findings suggest that the DL-LEM may improve the diagnostic performance in detecting significant stenosis in patients with severe coronary calcification. In addition, the results suggest that not a small medical economic effect can be expected.

Defect detection of petrochemical pipelines is an important task for industrial production safety. At present, pipeline defect detection mainly relies on closed circuit television method (CCTV) to take video of the pipeline inner wall and then detect the defective area manually, so the detection is very time-consuming and has a high rate of false and missed detections. To solve the above issues, we proposed an automatic defect detection system for petrochemical pipeline based on Cycle-GAN and improved YOLO v5. Firstly, in order to create the pipeline defect dataset, the original pipeline videos need pre-processing, which includes frame extraction, unfolding, illumination balancing, and image stitching to create coherent and tiled pipeline inner wall images. Secondly, aiming at the problems of small amount of samples and the imbalance of defect and non-defect classes, a sample enhancement strategy based on Cycle-GAN is proposed to generate defect images and expand the data set. Finally, in order to detect defective areas on the pipeline and improve the detection accuracy, a robust defect detection model based on improved YOLO v5 and Transformer attention mechanism is proposed, with the average precision and recall as 93.10% and 90.96%, and the F1-score as 0.920 on the test set. The proposed system can provide reference for operators in pipeline health inspection, improving the efficiency and accuracy of detection. [ABSTRACT FROM AUTHOR]

• A novel Cycle-GAN for specular highlight removal based on unpaired data. • The network is guided by a highlight mask to focus on highlight regions. • The training is divided into two steps through a luminance guidance structure. • The proposed method obtained high-quality highlight-free images even on unpaired natural images Specular highlight removal is an important yet challenging problem in image enhancement. Recent methods based on deep learning and trained by massive paired or unpaired data have demonstrated promising performance for this task. Methods based on unpaired data have recently gained more attention for easier training data collection. In this paper, we present a Mask-Guided Cycle-GAN for specular highlight removal on unpaired data. Incorporating the idea that specular highlight mainly has characteristics in lightness, we attempt to train a module only on luminance channel before considering all channels, and then adopt the training results to guide the subsequent highlight removal module. We further convert the highlight removal problem to image-to-image translation by using cycle-consistent adversarial network (Cycle-GAN). In the proposed network, a non-negative matrix factorization (NMF) based method is incorporated to obtain accurate highlight masks. The proposed method is evaluated using the specular highlight image quadruples (SHIQ) and the LIME datasets, and the advantages are demonstrated via comparative experimental results. [ABSTRACT FROM AUTHOR]

• A method for image generation was proposed for expanding medical image datasets of liver CT and MRI. • A correction module has been introduced, comprising an encoder and a decoder. The encoder extracts potential features, and the decoder eliminates feature blur, thereby significantly enhancing the quality of liver medical image generation. • We have introduced an additional loss function to transform the unsupervised process of image generation into a semi-supervised learning process, effectively resolving the feature perturbation problem caused by cycle consistency loss. • This method can effectively expand existing liver medical image datasets and shows significant advantages in addressing the scarcity of liver medical image sets. A high-performance computer-aided diagnosis (CAD) system can enhance the accuracy of liver cancer diagnosis, enabling early detection, diagnosis, and treatment. However, the availability of liver medical image datasets for training CAD systems is limited, and the existing augmentation methods generate images of lower quality, thereby limiting the performance of CAD systems. Therefore, this paper proposes a high-quality liver medical image generation algorithm based on an improved Cycle generative adversarial network (ICycle-GAN). Firstly, a correction network module based on an encoder-decoder structure is introduced into a Cycle generative adversarial network (Cycle-GAN). This module incorporates residual connections to efficiently extract latent feature representations from medical images and optimize them to generate higher-quality images. Secondly, a new loss function is embedded in the network based on the principle of loss correction. This loss function treats blurry images as noisy labels, transforming the unsupervised learning process of medical image transformation into a semi-supervised learning process. Finally, in the comparative experiments, the objective evaluation metrics including structural similarity index measure (SSIM), peak signal-to-noise ratio (PSNR), normalized mean absolute error (NMAE) and Fréchet Inception Distance (FID) for the generated liver computer tomography (CT) and magnetic resonance imaging (MRI) images by our proposed algorithm outperform the four mainstream medical image generation algorithms currently available. Moreover, the subjective visual quality of the generated images is also superior to that of the compared methods. [ABSTRACT FROM AUTHOR]

Background: Coronary computed tomography angiography examinations are increasingly becoming established as a minimally invasive method for diagnosing coronary diseases. However, although various imaging and processing methods have been developed, coronary artery calcification remains a major limitation in the evaluation of the vascular lumen. Subtraction coronary computed tomography angiography (Sub-CCTA) is a method known to be able to reduce the influence of coronary artery calcification and is therefore feasible for improving the diagnosis of significant stenosis in patients with severe calcification. However, Sub-CCTA still involves some problems, such as the increased radiation dose due to plain (mask) imaging, extended breath-holding time, and misregistration due to differences in the imaging phase. Therefore, we considered using artificial intelligence instead of Sub-CCTA to visualize the coronary lumen with high calcification. Given this background, the present study aimed to evaluate the diagnostic performance of a deep learning-based lumen extraction method (DL-LEM) to detect significant stenosis on CCTA in 99 consecutive patients (891 segments) with severe coronary calcification from November 2015 to March 2018. We also estimated the impact of DL-LEM on the medical economics in Japan. Results: The DL-LEM slightly improved the per-segment diagnostic accuracy from 74.5 to 76.4%, and the area under the curve (AUC) slightly improved from 0.752 to 0.767 (p = 0.030). When analyzing the 228 segments that could not be evaluated because of severe calcification on the original CCTA images, the DL-LEM improved the accuracy from 35.5 to 42.5%, and the AUC improved from 0.500 to 0.587 (p = 0.00018). As a result, DL-LEM analysis could have avoided invasive coronary angiography in 4/99 cases (per patient). From the calculated results, it was estimated that the number of exams that can be avoided in Japan in one year is approximately 747 for invasive coronary angiography, 219 for fractional flow reserve, and 248 for nuclear exam. The total amount of medical fee that could be reduced was 225,629,368 JPY. Conclusions: These findings suggest that the DL-LEM may improve the diagnostic performance in detecting significant stenosis in patients with severe coronary calcification. In addition, the results suggest that not a small medical economic effect can be expected. [ABSTRACT FROM AUTHOR]

As an active microwave coherent imaging technology, synthetic aperture radar (SAR) images suffer from severe speckle noise and low-resolution problems due to the limitations of the imaging system, which cause difficulties in image interpretation and target detection. However, the existing SAR super-resolution (SR) methods usually reconstruct the images by a determined degradation model and hardly consider multiplicative speckle noise, meanwhile, most SR models are trained with synthetic datasets in which the low-resolution (LR) images are down-sampled from their high-resolution (HR) counterparts. These constraints cause a serious domain gap between the synthetic and real SAR images. To solve the above problems, this paper proposes an unsupervised blind SR method for SAR images by introducing SAR priors in a cycle-GAN framework. First, a learnable probabilistic degradation model combined with SAR noise priors was presented to satisfy various SAR images produced from different platforms. Then, a degradation model and a SR model in a unified cycle-GAN framework were trained simultaneously to learn the intrinsic relationship between HR–LR domains. The model was trained with real LR and HR SAR images instead of synthetic paired images to conquer the domain gap. Finally, experimental results on both synthetic and real SAR images demonstrated the high performance of the proposed method in terms of image quality and visual perception. Additionally, we found the proposed SR method demonstrates the tremendous potential for target detection tasks by reducing missed detection and false alarms significantly.

Objective . Auto-segmentation in mouse micro-CT enhances the efficiency and consistency of preclinical experiments but often struggles with low-native-contrast and morphologically complex organs, such as the spleen, resulting in poor segmentation performance. While CT contrast agents can improve organ conspicuity, their use complicates experimental protocols and reduces feasibility. We developed a 3D Cycle Generative Adversarial Network (CycleGAN) incorporating anatomy-constrained U-Net models to leverage contrast-enhanced CT (CECT) insights to improve unenhanced native CT (NACT) segmentation. Approach. We employed a standard CycleGAN with an anatomical loss function to synthesize virtual CECT images from unpaired NACT scans at two different resolutions. Prior to training, two U-Nets were trained to automatically segment six major organs in NACT and CECT datasets, respectively. These pretrained 3D U-Nets were integrated during the CycleGAN training, segmenting synthetic images, and comparing them against ground truth annotations. The compound loss within the CycleGAN maintained anatomical fidelity. Full image processing was achieved for low-resolution datasets, while high-resolution datasets employed a patch-based method due to GPU memory constraints. Automated segmentation was applied to original NACT and synthetic CECT scans to evaluate CycleGAN performance using the Dice Similarity Coefficient (DSC) and the 95th percentile Hausdorff Distance (HD 95p ). Main results. High-resolution scans showed improved auto-segmentation, with an average DSC increase from 0.728 to 0.773 and a reduced HD95p from 1.19 mm to 0.94 mm. Low-resolution scans benefited more from synthetic contrast, showing a DSC increase from 0.586 to 0.682 and an HD 95p reduction from 3.46 mm to 1.24 mm. Significance. Implementing CycleGAN to synthesize CECT scans substantially improved the visibility of the mouse spleen, leading to more precise auto-segmentation. This approach shows the potential in preclinical imaging studies where contrast agent use is impractical., (© 2024 IOP Publishing Ltd.)

Disease has always been one of the main reasons for the decline of apple quality and yield, which directly harms the development of agricultural economy. Therefore, precise diagnosis of apple diseases and correct decision making are important measures to reduce agricultural losses and promote economic growth. In this paper, a novel Multi-scale Dense classification network is adopted to realize the diagnosis of 11 types of images, including healthy and diseased apple fruits and leaves. The diagnosis of different kinds of diseases and the same disease with different grades was accomplished. First of all, to solve the problem of insufficient images of anthracnose and ring rot, Cycle-GAN algorithm was applied to achieve dataset expansion on the basis of traditional image augmentation methods. Cycle-GAN learned the image characteristics of healthy apples and diseased apples to generate anthracnose and ring rot lesions on the surface of healthy apple fruits. The diseased apple images generated by Cycle-GAN were added to the training set, which improved the diagnosis performance compared with other traditional image augmentation methods. Subsequently, DenseNet and Multi-scale connection were adopted to establish two kinds of models, Multi-scale Dense Inception-V4 and Multi-scale Dense Inception-Resnet-V2, which facilitated the reuse of image features of the bottom layers in the classification neural networks. Both models accomplished the diagnosis of 11 different types of images. The classification accuracy was 94.31 and 94.74%, respectively, which exceeded DenseNet-121 network and reached the state-of-the-art level. [ABSTRACT FROM AUTHOR]

Disease has always been one of the main reasons for the decline of apple quality and yield, which directly harms the development of agricultural economy. Therefore, precise diagnosis of apple diseases and correct decision making are important measures to reduce agricultural losses and promote economic growth. In this paper, a novel Multi-scale Dense classification network is adopted to realize the diagnosis of 11 types of images, including healthy and diseased apple fruits and leaves. The diagnosis of different kinds of diseases and the same disease with different grades was accomplished. First of all, to solve the problem of insufficient images of anthracnose and ring rot, Cycle-GAN algorithm was applied to achieve dataset expansion on the basis of traditional image augmentation methods. Cycle-GAN learned the image characteristics of healthy apples and diseased apples to generate anthracnose and ring rot lesions on the surface of healthy apple fruits. The diseased apple images generated by Cycle-GAN were added to the training set, which improved the diagnosis performance compared with other traditional image augmentation methods. Subsequently, DenseNet and Multi-scale connection were adopted to establish two kinds of models, Multi-scale Dense Inception-V4 and Multi-scale Dense Inception-Resnet-V2, which facilitated the reuse of image features of the bottom layers in the classification neural networks. Both models accomplished the diagnosis of 11 different types of images. The classification accuracy was 94.31 and 94.74%, respectively, which exceeded DenseNet-121 network and reached the state-of-the-art level.

Defect detection of petrochemical pipelines is an important task for industrial production safety. At present, pipeline defect detection mainly relies on closed circuit television method (CCTV) to take video of the pipeline inner wall and then detect the defective area manually, so the detection is very time-consuming and has a high rate of false and missed detections. To solve the above issues, we proposed an automatic defect detection system for petrochemical pipeline based on Cycle-GAN and improved YOLO v5. Firstly, in order to create the pipeline defect dataset, the original pipeline videos need pre-processing, which includes frame extraction, unfolding, illumination balancing, and image stitching to create coherent and tiled pipeline inner wall images. Secondly, aiming at the problems of small amount of samples and the imbalance of defect and non-defect classes, a sample enhancement strategy based on Cycle-GAN is proposed to generate defect images and expand the data set. Finally, in order to detect defective areas on the pipeline and improve the detection accuracy, a robust defect detection model based on improved YOLO v5 and Transformer attention mechanism is proposed, with the average precision and recall as 93.10% and 90.96%, and the F1-score as 0.920 on the test set. The proposed system can provide reference for operators in pipeline health inspection, improving the efficiency and accuracy of detection.

• Body-size normalization of pediatric images helps training of a neural network. • Increasing training data size with adjacent anatomic images can be beneficial. • Preserved air shapes in sCT showed a potential for verifying anterior proton beam. To train a deep neural network for correcting abdominal and pelvic cone-beam computed tomography (CBCT) of children and young adults in the presence of diverse patient size, anatomic extent, and scan parameters. Pretreatment CBCT and planning/repeat CT image pairs from 64 children and young adults treated with proton therapy (aged 1–23 years) were analyzed. To evaluate the impact of anatomic extent in CBCT and data size in the training data, we compared the performance of three cycle-consistent generative adversarial network models that were separately trained by three datasets comprising abdominal (n = 21), pelvic (n = 29), and combined abdominal-pelvic image pairs (n = 50), respectively. The maximum body width of each patient was normalized to a fixed width before training and model application to reduce the impact of variations in body size. The corrected CBCT images by the three models were comparatively evaluated against the repeat CT closest in time to the CBCT (median gap, 0 days; range, 0–6 days) in HU accuracy, estimated dose distribution, and proton range. The network model trained by the combined dataset significantly outperformed the abdomen and pelvis models in mean absolute HU error of the corrected CBCT from 14 testing patients (47 ± 7 HU versus 51 ± 8 HU; paired Wilcoxon signed-rank test, P < 0.01). The larger error (60 ± 7 HU) without the body-size normalization confirmed the efficacy of the preprocessing. The model trained with the combined dataset resulted in gamma passing rates of 98.5 ± 1.9% (2%/2 mm criterion) and the range (80% distal fall-off) differences from the reference within ±3 mm for 91.2 ± 11.5% beamlets. Combining data from adjacent anatomic sites and normalizing age-dependent body sizes in children and young adults were beneficial in training a neural network to accurately estimate proton dose from CBCT despite limited training data size and anatomic diversities. [ABSTRACT FROM AUTHOR]

As a widely deployed security scheme, text-based completely automated public Turing tests to tell computers and humans apart (CAPTCHAs) have become increasingly unable to resist machine learning-based attacks. So far, many researchers have conducted studies on approaches for attacking text-based CAPTCHAs deployed by different companies, such as Microsoft, Amazon, and Apple, and achieved specific results. However, most of these attacks have shortcomings, such as the poor portability of attack methods, which require a series of data preprocessing steps and rely on large amounts of labeled CAPTCHAs. In this study, we propose an efficient and simple end-to-end attack method based on cycle-consistent generative adversarial networks (Cycle-GANs). Compared to previous studies, our approach significantly reduces the cost of data labeling. Additionally, this method has high portability. It can attack ordinary text-based CAPTCHA schemes only by modifying a few configuration parameters, which makes the attack easier to execute. First, we train CAPTCHA synthesizers based on the Cycle-GAN to generate some fake samples. Basic recognizers based on a convolutional recurrent neural network are trained using the fake data. Subsequently, an active transfer learning method is employed to optimize the basic recognizer utilizing tiny amounts of labeled real-world CAPTCHA samples. Our approach efficiently cracked the CAPTCHA schemes deployed by 10 popular websites, indicating that our attack method may be universal. Additionally, we analyzed the current most popular anti-recognition mechanisms. The results show that the combination of more anti-recognition mechanisms can improve the security of CAPTCHAs. However, the improvement is limited. Conversely, generating more complex CAPTCHAs may cost more resources and reduce the usability of CAPTCHAs. [ABSTRACT FROM AUTHOR]

Magnetic resonance imaging (MRI) presents a detailed image of the internal organs via a magnetic field. Given MRI's non-invasive advantage in repeated imaging, the low-contrast MR images in the target area make segmentation of tissue a challenging problem. This study shows the potential advantages of synthetic high tissue contrast (HTC) images through image-to-image translation techniques. Mainly, we use a novel cycle generative adversarial network (Cycle-GAN), which provides a mechanism of attention to increase the contrast within the tissue. The attention block and training on HTC images are beneficial to our model to enhance tissue visibility. We use a multistage architecture to concentrate on a single tissue as a preliminary and filter out the irrelevant context in every stage in order to increase the resolution of HTC images. The multistage architecture reduces the gap between source and target domains and alleviates synthetic images' artefacts. We apply our HTC image synthesising method to two public datasets. In order to validate the effectiveness of these images we use HTC MR images in both end-to-end and two-stage segmentation structures. The experiments on three segmentation baselines on BraTS'18 demonstrate that joining the synthetic HTC images in the multimodal segmentation framework develops the average Dice similarity scores (DSCs) of 0.8%, 0.6%, and 0.5% respectively on the whole tumour (WT), tumour core (TC), and enhancing tumour (ET) while removing one real MRI channels from the segmentation pipeline. Moreover, segmentation of infant brain tissue in T1w MR slices through our framework improves DSCs approximately 1% in cerebrospinal fluid (CSF), grey matter (GM), and white matter (WM) compared to state-of-the-art segmentation techniques. The source code of synthesising HTC images is publicly available. • Our framework increases the contrast among tissues in MR images (MRI-to-HTC). • Training on HTC images and an attention block guide model to attend certain tissues. • Multi-stage model decreases the gap between the source and target domain. • Synthetic HTC MR images improve Dice in the segmentation of tumour and infant dataset. [ABSTRACT FROM AUTHOR]

Deep learning has been used to analyze and diagnose various skin diseases through medical imaging. However, recent researches show that a well-trained deep learning model may not generalize well to data from different cohorts due to domain shift. Simple data fusion techniques such as combining disease samples from different data sources are not effective to solve this problem. In this paper, we present two methods for a novel task of cross-domain skin disease recognition. Starting from a fully supervised deep convolutional neural network classifier pre-trained on ImageNet, we explore a two-step progressive transfer learning technique by fine-tuning the network on two skin disease datasets. We then propose to adopt adversarial learning as a domain adaptation technique to perform invariant attribute translation from source to target domain in order to improve the recognition performance. In order to evaluate these two methods, we analyze generalization capability of the trained model on melanoma detection, cancer detection, and cross-modality learning tasks on two skin image datasets collected from different clinical settings and cohorts with different disease distributions. The experiments prove the effectiveness of our method in solving the domain shift problem. [ABSTRACT FROM AUTHOR]

Purpose: The incorporation of cone‐beam computed tomography (CBCT) has allowed for enhanced image‐guided radiation therapy. While CBCT allows for daily 3D imaging, images suffer from severe artifacts, limiting the clinical potential of CBCT. In this work, a deep learning‐based method for generating high quality corrected CBCT (CCBCT) images is proposed. Methods: The proposed method integrates a residual block concept into a cycle‐consistent adversarial network (cycle‐GAN) framework, called res‐cycle GAN, to learn a mapping between CBCT images and paired planning CT images. Compared with a GAN, a cycle‐GAN includes an inverse transformation from CBCT to CT images, which constrains the model by forcing calculation of both a CCBCT and a synthetic CBCT. A fully convolution neural network with residual blocks is used in the generator to enable end‐to‐end CBCT‐to‐CT transformations. The proposed algorithm was evaluated using 24 sets of patient data in the brain and 20 sets of patient data in the pelvis. The mean absolute error (MAE), peak signal‐to‐noise ratio (PSNR), normalized cross‐correlation (NCC) indices, and spatial non‐uniformity (SNU) were used to quantify the correction accuracy of the proposed algorithm. The proposed method is compared to both a conventional scatter correction and another machine learning‐based CBCT correction method. Results: Overall, the MAE, PSNR, NCC, and SNU were 13.0 HU, 37.5 dB, 0.99, and 0.05 in the brain, 16.1 HU, 30.7 dB, 0.98, and 0.09 in the pelvis for the proposed method, improvements of 45%, 16%, 1%, and 93% in the brain, and 71%, 38%, 2%, and 65% in the pelvis, over the CBCT image. The proposed method showed superior image quality as compared to the scatter correction method, reducing noise and artifact severity. The proposed method produced images with less noise and artifacts than the comparison machine learning‐based method. Conclusions: The authors have developed a novel deep learning‐based method to generate high‐quality corrected CBCT images. The proposed method increases onboard CBCT image quality, making it comparable to that of the planning CT. With further evaluation and clinical implementation, this method could lead to quantitative adaptive radiation therapy. [ABSTRACT FROM AUTHOR]

The performance of face photo-sketch translation has improved a lot thanks to deep neural networks. GAN based methods trained on paired images can produce high-quality results under laboratory settings. Such paired datasets are, however, often very small and lack diversity. Meanwhile, Cycle-GANs trained with unpaired photo-sketch datasets suffer from the steganography phenomenon, which makes them not effective to face photos in the wild. In this paper, we introduce a semi-supervised approach with a noise-injection strategy, named Semi-Cycle-GAN (SCG), to tackle these problems. For the first problem, we propose a pseudo sketch feature representation for each input photo composed from a small reference set of photo-sketch pairs, and use the resulting pseudo pairs to supervise a photo-to-sketch generator G p 2 s. The outputs of G p 2 s can in turn help to train a sketch-to-photo generator G s 2 p in a self-supervised manner. This allows us to train G p 2 s and G s 2 p using a small reference set of photo-sketch pairs together with a large face photo dataset (without ground-truth sketches). For the second problem, we show that the simple noise-injection strategy works well to alleviate the steganography effect in SCG and helps to produce more reasonable sketch-to-photo results with less overfitting than fully supervised approaches. Experiments show that SCG achieves competitive performance on public benchmarks and superior results on photos in the wild. • A semi-supervised learning framework based on Cycle-GAN for face sketch translation. • The network can be trained with lots of photos without GT, improving generalization. • A simple noise-injection strategy to solve invisible steganography in Cycle-GAN. [ABSTRACT FROM AUTHOR]

Background and Purpose: Addressing the need for accurate dose calculation in MRI-only radiotherapy, the generation of synthetic Computed Tomography (sCT) from MRI has emerged. Deep learning (DL) techniques, have shown promising results in achieving high sCT accuracies. However, existing sCT synthesis methods are often center-specific, posing a challenge to their generalizability. To overcome this limitation, recent studies have proposed approaches, such as multicenter training . Material and methods: The purpose of this work was to propose a multicenter sCT synthesis by DL, using a 2D cycle-GAN on 128 prostate cancer patients, from four different centers. Four cases were compared: monocenter cases, monocenter training and test on another center, multicenter trainings and a test on a center not included in the training and multicenter trainings with an included center in the test. Trainings were performed using 20 patients. sCT accuracy evaluation was performed using Mean Absolute Error, Mean Error and Peak-Signal-to-Noise-Ratio. Dose accuracy was assessed with gamma index and Dose Volume Histogram comparison. Results: Qualitative, quantitative and dose results show that the accuracy of sCTs for monocenter trainings and multicenter trainings using a seen center in the test did not differ significantly. However, when the test involved an unseen center, the sCT quality was inferior. Conclusions: The aim of this work was to propose generalizable multicenter training for MR-to-CT synthesis. It was shown that only a few data from one center included in the training cohort allows sCT accuracy equivalent to a monocenter study., Competing Interests: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (© 2023 The Author(s).)

In an automated box de-palletisation system that utilises robots, vision-based box recognition on the pallet plays the main role in providing picking guidelines. The complexity of the working condition and the target object, particularly the cluttered arrangement and various outer surfaces of the boxes, significantly affect the quality of the outcome. Typically, a large-scale vision dataset is required to train a deep learning object-detection model. However, considerable effort and time is required to achieve this. Therefore, this study proposes a Mask R-CNN-based detection approach for box objects, which is supported by a cycle generative adversarial network (Cycle GAN). The purpose of the Cycle-GAN is to optimise the outer surfaces of boxes by automatically erasing tags, stickers, labels, and symbols that exist on the boxes before loading them to the Mask R-CNN for detection. Subsequently, the obtained result was combined with the output from the developed boundary-enhancing technique that was applied to a depth map. Consequently, the box detection performance was significantly improved, and it was confirmed through experiments with a practical robot system in picking tasks. In the experiments, the success rate of the proposed method was validated using 200 cases of orderly and disorderly arrangements of boxes, respectively. Furthermore, the metric of the mean absolute error between the predicted picking point and the ground truth values for the test cases in the implementation process for the robot operation was also researched. [ABSTRACT FROM AUTHOR]

In estimating optical flow using convolutional neural networks, a gap in accuracy exists between supervised and unsupervised training because supervised methods can obtain more explicit clues from GT data. In this study, we bridge this disparity by introducing a cycle adversarial mechanism to provide additional flow priors. Specifically, this mechanism eliminates the requirement for paired data and enables unsupervised training. Furthermore, it provides more explicit clues for the network and optimizes the flow performance by adversarial training between the generator and the discriminator. Additionally, we propose a redesigned generator named the Recursion-Attention Flow Generator (RAFG) and Homology Isomerism Discriminator (HID) to adapt the UDF-GAN to flow estimation tasks. Finally, we use SpyNet as a teacher network to direct the UDF-GAN to improve the robustness to mode collapse in the early training phase. The experiments were conducted on various public flow datasets, in contrast to several state-of-the-art flow models for flow estimation, that involve CNN- and GAN-based methods trained in unsupervised and supervised manners, respectively. • We introduced the cycle adversarial mechanism into the flow task to provide additional clues for estimation. • We redesigned the generator and discriminator to adapt the network for flow estimation. • We introduced SpyNet as a teacher network to guide the generator directly in the early training phase. [ABSTRACT FROM AUTHOR]