Cycle-consistent Generative Adversarial Network Research Articles

Vortex Beam Transmission Compensation in Atmospheric Turbulence Using CycleGAN

To improve the robustness of vortex beam transmission and detection in the face of atmospheric turbulence and to guarantee accurate recognition of orbital angular momentum (OAM), we present an end-to-end dynamic compensation technique for vortex beams using an improved cycle-consistent generative adversarial network (CycleGAN). This approach transforms the problem of vortex beam distortion compensation into one of image translation. The Pix2pix and CycleGAN models were extended with a structural similarity loss function to constrain turbulence distortion compensation in luminance, contrast and structure. Experiments were designed to evaluate the compensation performance from subjective and objective indicators. The simulation results demonstrate that the optical OAM intensity map is very similar to that of the target OAM light after compensation. The mean value of structural similarity is close to 1. The recognition accuracy of the OAM is improved by 4.4% compared to no distortion compensation, demonstrating that the improved CycleGAN-based compensation scheme can guarantee excellent detection accuracy without reconstructing the wavefront and saving optical hardware. The method can be implemented in real-time optical communications in atmospheric turbulence environments.

Comparison and evaluation of different deep learning models of synthetic CT generation from CBCT for nasopharynx cancer adaptive proton therapy.

Cone-beam computed tomography (CBCT) scanning is used for patient setup in image-guided radiotherapy. However, its inaccurate CT numbers limit its applicability in dose calculation and treatment planning. This study compares four deep learning methods for generating synthetic CT (sCT) to determine which method is more appropriate and offers potential for further clinical exploration in adaptive proton therapy for nasopharynx cancer. CBCTs and deformed planning CT (dCT) from 75 patients (60/5/10 for training, validation and testing) were used to compare cycle-consistent Generative Adversarial Network(cycleGAN), Unet, Unet+cycleGAN and conditionalGenerative Adversarial Network (cGAN) for sCT generation. The sCT images generated by each method were evaluated against dCT images using mean absolute error (MAE), structural similarity (SSIM), peak signal-to-noise ratio (PSNR), spatial non-uniformity (SNU) and radial averaging in the frequency domain. In addition, dosimetric accuracy was assessed through gamma analysis, differences in water equivalent thickness (WET), and dose-volume histogram metrics. The cGAN model has demonstrated optimal performance in the four models across various indicators. In terms of image quality under global condition, the average MAE has been reduced to 16.39HU, SSIM has increased to 95.24%, and PSNR has increased to 28.98. Regarding dosimetric accuracy, the gamma passing rate (2%/2mm) has reached 99.02%, and the WET difference is only 1.28mm. The D95 value of CTVs coverage and Dmax value of spinal cord, brainstem show no significant differences between dCT and sCT generated by cGAN model. The cGAN model has been shown to be a more suitable approach for generating sCT using CBCT, considering its characteristics and concepts. The resulting sCT has the potential for application in adaptive proton therapy.

Semi-Supervised MIMO Detection Using Cycle-Consistent Generative Adversarial Network

In this paper, a new semi-supervised deep multiple-input multiple-output (MIMO) detection approach using a cycle-consistent generative adversarial network (CycleGAN) is proposed for communication systems without any prior knowledge of underlying channel distributions. Specifically, we propose the CycleGAN detector by constructing a bidirectional loop of two modified least squares generative adversarial networks (LS-GAN). The forward LS-GAN learns to model the transmission process, while the backward LS-GAN learns to detect the received signals. By optimizing the cycle-consistency of the transmitted and received signals through this loop, the proposed method is trained online and semi-supervisedly using both the pilots and the received payload data. As such, the demand on labelled training dataset is considerably controlled, and thus the overhead is effectively reduced. Numerical results show that the proposed CycleGAN detector achieves better performance in terms of both bit error-rate (BER) and achievable rate than existing semi-blind deep learning (DL) detection methods as well as conventional linear detectors, especially when considering signal distortion due to the nonlinearity of power amplifiers (PA) at the transmitter.

Synthetic MRI-Assisted and Self-Supervised Adaptive Segmentation of Organs-at-Risk (OARs) in MRI-Based Radiation Therapy

This study proposes a self-supervised solution for OAR segmentation, combining patch-based adaptation and unsupervised synthesis of T2-weighted MRI data to finetune the segmentation model. The aim is to improve adaptation to patient anatomy, overcome limited annotated MRI data, and enhance the generalizability of automatic segmentation models for gynecological cancers. The study used a patch-based cycle consistent generative adversarial network (cycle-GAN) for unsupervised MRI synthesis from CT scans of 20 patients, and a residual U-Net model for OARs segmentation. The segmentation model was trained and validated on synthetic MRI (sMRI) of 103 and 25 patient scans respectively, then finetuned on 78 MRI scans from radiation therapy fractions of 15 additional patients through three-fold cross validation. Self-supervised adaptation was applied, incorporating affine and elastic deformations, intensity shifting, and scaling. The model was trained on 96 × 96 × 96 sub-volumes and validated on entire pelvic sections of the same images. A combination of Dice and weighted cross entropy (CE) losses, with weights assigned for bladder (1), small bowel (1), rectum (2), sigmoid (2), left femoral head (0), and right femoral head (0), was used for OAR segmentation. The performance was evaluated against the model trained only on a limited number of acquired MRI data, as well as sMRI pretrained models with encoder weight freezing and either equal weighting or soft-tissue adjusted weighting. Our sMRI-assisted approach showed improved performance for challenging pelvic OARs compared to the method using only the acquired MRI data. The self-supervised fraction-adaptive segmentation results indicated better performance in target soft-tissues when using at least one treatment fraction for organ-specific adaptation. Our framework leverages pre-existing CT planning data for gynecological cancers to enhance the segmentation performance of OARs during MR-guided adaptive treatments. This approach offers substantial benefits for the radiation therapy workflow, including reduced variability in per-fraction segmentation and clinical burden. Further studies that involve human expert evaluations will be conducted to assess the practicality of this approach in radiation therapy.

Deep learning-enabled realistic virtual histology with ultraviolet photoacoustic remote sensing microscopy

The goal of oncologic surgeries is complete tumor resection, yet positive margins are frequently found postoperatively using gold standard H&E-stained histology methods. Frozen section analysis is sometimes performed for rapid intraoperative margin evaluation, albeit with known inaccuracies. Here, we introduce a label-free histological imaging method based on an ultraviolet photoacoustic remote sensing and scattering microscope, combined with unsupervised deep learning using a cycle-consistent generative adversarial network for realistic virtual staining. Unstained tissues are scanned at rates of up to 7 mins/cm2, at resolution equivalent to 400x digital histopathology. Quantitative validation suggests strong concordance with conventional histology in benign and malignant prostate and breast tissues. In diagnostic utility studies we demonstrate a mean sensitivity and specificity of 0.96 and 0.91 in breast specimens, and respectively 0.87 and 0.94 in prostate specimens. We also find virtual stain quality is preferred (P = 0.03) compared to frozen section analysis in a blinded survey of pathologists.

An Improved CycleGAN for Data Augmentation in Person Re-Identification

Person re-identification (ReID) has attracted more and more attention, which is to retrieve interested persons across multiple non-overlapping cameras. Matching the same person between different camera styles has always been an enormous challenge. In the existing work, cross-camera styles images generated by the cycle-consistent generative adversarial network (CycleGAN) only transfer the camera resolution and ambient lighting. The generated images produce considerable redundancy and inappropriate pictures at the same time. Although the data is added to prevent over-fitting, it also makes significant noise, so the accuracy is not significantly improved. In this paper, an improved CycleGAN is proposed to generate images for achieving improved data augmentation. The transfer of pedestrian posture is added at the same time as transferring the image style. It not only increases the diversity of pedestrian posture but also reduces the domain gap caused by the style change between cameras. Besides, through the multi-pseudo regularized label (MpRL), the generated images are assigned virtual tags dynamically in training. Through many experimental evaluations, we have achieved a very high identification accuracy on Market-1501, DukeMTMC-reID, and CUHK03-NP datasets. On the three datasets, the quantitative results of mAP are 96.20%, 93.72%, and 86.65%, and the quantitative results of rank-1 are 98.27%, 95.37%, and 90.71%, respectively. The experimental results fully show the superiority of our proposed method.

Deep learning enhanced fluorescence emission difference microscopy

Fluorescence emission difference microscopy (FED) obtains super-resolution microscopic images by extracting the information of the intensity differences between the solid and doughnut confocal images. Due to the mismatch of the outer contour of the solid and doughnut point spread function, FED suffers the problem of information loss. Here, we present a framework for deep-learning-enhanced FED (DL-FED) microscopy through cycle-consistent generative adversarial network (CycleGAN) based image reconstruction. Using this framework, we effectively avoid the information loss and enhance the spatial resolution of the fluorescence images acquired by the standard FED. We also demonstrate that the standard FED images can be transformed to match the results of Airyscan-based FED and saturated-FED, which can effectively enhance the signal-to-noise ratio and avoid the photo-bleaching effect. The validity of DL-FED is demonstrated by simulations and experiments based on fluorescent nanoparticles and biological cells. Featuring the potential to realize a high imaging speed, this approach may be widely applied in live cells investigations.

Identifying underlying individuality across running, walking, and handwriting patterns with conditional cycle-consistent generative adversarial networks.

In recent years, the analysis of movement patterns has increasingly focused on the individuality of movements. After long speculations about weak individuality, strong individuality is now accepted, and the first situation-dependent fine structures within it are already identified. Methodologically, however, only signals of the same movements have been compared so far. The goal of this work is to detect cross-movement commonalities of individual walking, running, and handwriting patterns using data augmentation. A total of 17 healthy adults (35.8 ± 11.1years, eight women and nine men) each performed 627.9 ± 129.0 walking strides, 962.9 ± 182.0 running strides, and 59.25 ± 1.8 handwritings. Using the conditional cycle-consistent generative adversarial network (CycleGAN), conditioned on the participant's class, a pairwise transformation between the vertical ground reaction force during walking and running and the vertical pen pressure during handwriting was learned in the first step. In the second step, the original data of the respective movements were used to artificially generate the other movement data. In the third step, whether the artificially generated data could be correctly assigned to a person via classification using a support vector machine trained with original data of the movement was tested. The classification F1-score ranged from 46.8% for handwriting data generated from walking data to 98.9% for walking data generated from running data. Thus, cross-movement individual patterns could be identified. Therefore, the methodology presented in this study may help to enable cross-movement analysis and the artificial generation of larger amounts of data.

Bidirectional Mapping Perception-enhanced Cycle-consistent Generative Adversarial Network for Super-resolution of Brain MRI images.

As an effective tool for visualizing neurodegeneration, high-resolution structural magnetism facilitates quantitative image analysis and clinical applications. Super-resolution reconstruction technology allows to improve the resolution of images without upgrading the scanning hardware. However, existing super-resolution techniques relied on paired image data sets and lacked further quantitative analysis of the generated images. In this study, we proposed a semi-supervised generative adversarial network (GAN) model for super-resolution of brain MRI, and the synthetic images were evaluated using various quantitative measures. This model adopted the cycle-consistency structure to allow for a mixture of unpaired data for training. Perceptual loss was further introduced into the model to preserve detailed texture features at high frequencies. 363 subjects with both high-resolution (HR) and low-resolution (LR) scans and 217 subjects with HR scans only were used for model derivation, training, and validation. We extracted multiple voxel-based and surface-based morphological features of the synthetic and real 3D HR images for comparison. We further evaluated the synthetic images in the differential diagnosis of diseases. Our model achieved superior mean absolute error (0.049±0.021), mean squared error (0.0059±0.0043), peak signal-to-noise ratio (29.41±3.71), structural similarity index measure (0.914±0.048). Eight morphological metrics, both voxel-based and surface-based, showed significant agreement (P<0.0001). The gap of accuracy in disease diagnosis between synthetic and real HR images was within 5% and significantly outperformed the LR images. Our proposed model enables the reconstruction of HR MRI and could be used accurately for image quantification.Clinical relevance- Quantitative evaluation of the synthetic high-resolution images was used to determine whether the synthetic images have sufficient realism and diversity.

DFC-dehaze: an improved cycle-consistent generative adversarial network for unpaired image dehazing

DFC-dehaze: an improved cycle-consistent generative adversarial network for unpaired image dehazing

PSSGAN: Towards spectrum shift based perceptual quality enhancement for fluorescence imaging.

Fluorescence imaging has demonstrated great potential for malignant tissue inspection. However, poor imaging quality of medical fluorescent images inevitably brings challenges to disease diagnosis. Though improvement of image quality can be achieved by translating the images from low-quality domain to high-quality domain, fewer scholars have studied the spectrum translation and the prevalent cycle-consistent generative adversarial network (CycleGAN) is powerless to grasp local and semantic details, leading to produce unsatisfactory translated images. To enhance the visual quality by shifting spectrum and alleviate the under-constraint problem of CycleGAN, this study presents the design and construction of the perception-enhanced spectrum shift GAN (PSSGAN). Besides, by introducing the constraint of perceptual module and relativistic patch, the model learns effective biological structure details of image translation. Moreover, the interpolation technique is innovatively employed to validate that PSSGAN can vividly show the enhancement process and handle the perception-fidelity trade-off dilemma of fluorescent images. A novel no reference quantitative analysis strategy is presented for medical images. On the open data and collected sets, PSSGAN provided 15.32% ∼ 35.19% improvement in structural similarity and 21.55% ∼ 27.29% improvement in perceptual quality over the leading method CycleGAN. Extensive experimental results indicated that our PSSGAN achieved superior performance and exhibited vital clinical significance.

Non-contrast CT synthesis using patch-based cycle-consistent generative adversarial network (Cycle-GAN) for radiomics and deep learning in the era of COVID-19

Handcrafted and deep learning (DL) radiomics are popular techniques used to develop computed tomography (CT) imaging-based artificial intelligence models for COVID-19 research. However, contrast heterogeneity from real-world datasets may impair model performance. Contrast-homogenous datasets present a potential solution. We developed a 3D patch-based cycle-consistent generative adversarial network (cycle-GAN) to synthesize non-contrast images from contrast CTs, as a data homogenization tool. We used a multi-centre dataset of 2078 scans from 1,650 patients with COVID-19. Few studies have previously evaluated GAN-generated images with handcrafted radiomics, DL and human assessment tasks. We evaluated the performance of our cycle-GAN with these three approaches. In a modified Turing-test, human experts identified synthetic vs acquired images, with a false positive rate of 67% and Fleiss’ Kappa 0.06, attesting to the photorealism of the synthetic images. However, on testing performance of machine learning classifiers with radiomic features, performance decreased with use of synthetic images. Marked percentage difference was noted in feature values between pre- and post-GAN non-contrast images. With DL classification, deterioration in performance was observed with synthetic images. Our results show that whilst GANs can produce images sufficient to pass human assessment, caution is advised before GAN-synthesized images are used in medical imaging applications.

Speckle noise reduction for medical ultrasound images based on cycle-consistent generative adversarial network

Medical ultrasound (US) images are corrupted by speckle noise, which can adversely affect the disease diagnosis and treatment. Recently, the cycle-consistent adversarial network (CycleGAN) has been used in the style transfer for both natural and medical images. In this study, we aim to develop a US despeckling method based on the CycleGAN by the style transfer between the speckled noisy data domain and noise-free data domain by forming a bi-directional universal mapping. The inputs of noisy and noise-free images are designed in the CycleGAN model. For simulation work, we use both the multiplicative model and the spatial impulse response model to obtain noisy images from noise-free images. However, noise-free US images are not clinically available. Hence, for the real US despeckling scenario, the clinical images of hearts, lymph nodes, and breast tumors are used as noisy images; and the high-quality images that are derived from the clinical images by despeckling with the Gabor-based anisotropic diffusion (GAD) and selected with a new metric named the edge-to-noise ratio, are used as the noise-free images. We compare our CycleGAN based denoising method with nine existing denoising methods, i.e., the speckle reduction anisotropic diffusion, GAD, non-local means, wavelet transform, unbiased non-local means, statistical nearest-neighbors, TVHTVM, improved non-local self-similarity measures, and generative adversarial network. Our method outperforms other methods by visual assessment and quantitative comparison, which demonstrates its superiority for noise reduction and detail preservation.

Analysis of reconstructed multisource and multiscale 3-D digital rocks based on the cycle-consistent generative adversarial network method

SUMMARY Digital rock physics (DRP) is important for characterizing the pore characteristics and petrophysical parameters of rocks from a microscopic perspective. Among the digital rock reconstruction methods, the most widely used is the computer tomography (CT) scan method. However, the pore structure of rocks has multiscale features, and CT scan images with a single resolution cannot completely describe the pore structure characteristics of rocks. In this paper, we analysed reconstructed multisource and multiscale 3-D digital rocks based on the cycle-consistent generative adversarial network (CycleGAN) method. This method fully integrates the advantages of the large field of view of low-resolution images and the high-precision features of high-resolution images. To investigate the performance of the method, two sequences of CT scan images of a sandstone (sample A) and a carbonate rock (sample B) collected from oilfields were studied. Moreover, due to the high resolution of scanning electron microscope (SEM) images, we also constructed 3-D digital rocks with different resolutions of the Berea sandstone based on the combination of CT scan images and SEM images. Finally, the statistical properties and absolute permeabilities were calculated to evaluate the accuracies of reconstructed multisource and multiscale 3-D digital rocks. The results show that the reconstructed multiscale digital rocks based on the CycleGAN method have good accuracy in terms of statistical properties and petrophysical properties. Considering the computational cost and computational accuracy, for high- and low-resolution CT scan images and CT scan images with SEM images as training image data sources, we suggest that the resolution of the reconstructed multiscale digital cores is 4–8 times and 4–16 times higher than that of the low-resolution CT scan images, respectively. The findings of our research will be helpful in gaining insight into the petrophysical properties of heterogeneous rocks.

Smoke Image Segmentation Algorithm Suitable for Low-Light Scenes

The real-time monitoring and analysis system based on video images has been implemented to detect fire accidents on site. While most segmentation methods can accurately segment smoke areas in bright and clear images, it becomes challenging to obtain high performance due to the low brightness and contrast of low-light smoke images. An image enhancement model cascaded with a semantic segmentation model was proposed to enhance the segmentation effect of low-light smoke images. The modified Cycle-Consistent Generative Adversarial Network (CycleGAN) was used to enhance the low-light images, making smoke features apparent and improving the detection ability of the subsequent segmentation model. The smoke segmentation model was based on Transformers and HRNet, where semantic features at different scales were fused in a dense form. The addition of attention modules of spatial dimension and channel dimension to the feature extraction units established the relationship mappings between pixels and features in the two-dimensional spatial directions, which improved the segmentation ability. Through the Foreground Feature Localization Module (FFLM), the discrimination between foreground and background features was increased, and the ability of the model to distinguish the thinner positions of smoke edges was improved. The enhanced segmentation method achieved a segmentation accuracy of 91.68% on the self-built dataset with synthetic low-light images and an overall detection time of 120.1 ms. This method can successfully meet the fire detection demands in low-light environments at night and lay a foundation for expanding the all-weather application of initial fire detection technology based on image analysis.

A transfer learning strategy based on numerical simulation driving 1D Cycle-GAN for bearing fault diagnosis

Most transfer learning (TL) models generally need the fault data from similar scenarios to achieve cross-domain bearing fault diagnosis. However, due to the bearings are mostly in normal state and the fault experiment cost is high, it is hard to collect the complete fault data in any real scenario. Thus, the problem of no fault data still limits the application of various intelligent diagnosis methods, including TL methods. To solve this problem, this paper proposed a novel transfer learning (TL) strategy by driving the one-dimensional (1D) cycle-consistent generative adversarial network (Cycle-GAN) with numerical simulation. In this strategy, the bearing dynamic model is constructed to generate the simulated vibration signals under normal and fault states. Furthermore, a 1D Cycle-GAN is first devised as the TL model for signal-to-signal translation. By using all simulated signals (normal and fault) and real normal signals, this 1D Cycle-GAN is trained to translate each simulated signal into a nearly real signal. Two experiments indicate that this strategy can generate satisfactory fault signals to solve the problem of no fault data in bearing fault diagnosis.

A deep learning solution for particle size analysis in low resolution inline microscopy images based on generative adversarial network

We introduce a new strategy for image analysis of inline microscopy monitoring estimate particle size distribution using deep learning. The proposed method consists of two major components: First, a novel way to generate training image-label pairs with a high-level of credibility via a Cycle-consistent Generative Adversarial Network (CycleGAN), and second, a Mask-RCNN model trained with the generated data for the particle detection task. The proposed methodology eliminates the need for manual labeling in the training phase which is a labor-intensive step and can result in labeling errors given the fuzziness of these images. We studied the application of this strategy to images acquired with a particle vision and measurement (PVM) probe.The proposed methodology was applied to images of two particle morphologies with different sizes and concentrations. Our results showed that the proposed methodology could be inexpensively used to determine qualitative trends between crystal size distributions. This trend information is a very important aspect of crystallization process monitoring and is often enough to determine what is controlling the crystallization. Therefore, we see our approach as a step in the right direction to provide insights into the particularly challenging PVM inline microscopy monitoring process without the need for offline sampling.

A projection-domain correction method in CBCT reconstruction for head and neck radiotherapy using cycle-GAN and nonlocal means filter.

In radiation treatments for head and neck tumors, cone-beam computed tomography (CBCT) is employed for patient positioning and dose calculation of adaptive radiotherapy. However, the quality of CBCT is degraded by the scatter and noise, majorly impacting the accuracy of patient positioning and dose calculation. To improve the quality of CBCT for patients with head and neck cancer, a projection-domain CBCT correction method was proposed using a cycle-consistent generative adversarial network (cycle-GAN) and a nonlocal means filter based on a reference digitally reconstructed radiograph (DRR). A cycle-GAN was initially trained to learn mapping from CBCT projections to a DRR using the data obtained from 30 patients. For each patient, 671 CBCT projections were measured for CBCT reconstruction. Moreover, 360 Digital Reconstructed Radiographs (DRR) were computed from each patient's planning computed tomography (CT), whose projection angles ranged from 0° to 359° with an interval of 1°. By applying the trained generator of the cycle-GAN to the unseen CBCT projection, a synthetic DRR with considerably less scatter was obtained. However, annular artifacts were observed in the CBCT reconstructed with synthetic DRR. To address this issue, a nonlocal means filter based on reference DRR was used to further correct the synthetic DRR, which corrected the synthetic DRR using the calculated DRR as a reference image. Finally, the CBCT with no annular artifact and little noise was reconstructed with the corrected synthetic DRR. The proposed method was tested using the data of six patients. The corrected synthetic DRR and CBCT were compared with the corresponding real DRR and CT images. The structural preservation ability of the proposed method was evaluated using the Dice coefficients of the automatically extracted nasal cavity. Moreover, The image quality of CBCT corrected with the proposed method was objectively assessed with an five-point human scoring system and compared with CT, original CBCT and CBCT corrected with other strategies. The mean absolute value (MAE) of the relative error between the corrected synthetic and real DRR was <8%. The MAE between the corrected CBCT and corresponding CT was <30 HU. Moreover, the Dice coefficient of nasal cavity between the corrected CBCT image and the original image exceeded 98.8 for all the patients. Last but not least, the objective assessment of image qualityshowed the proposed method had an average score of 4.2 in overall image quality, which was higher than that of the original CBCT, CBCT reconstructed with synthetic DRR, and CBCT reconstructed with projections filtered with NLMF only. The proposed method can considerably improve the CBCT image quality with little anatomical distortion, improving the accuracy of radiotherapy for head and neck patients. This article is protected by copyright. All rights reserved.

A semi-supervised material quantitative intelligent imaging algorithm for spectral CT based on prior information perception learning

To propose a semi-supervised material quantitative intelligent imaging algorithm based on prior information perception learning (SLMD-Net) to improve the quality and precision of spectral CT imaging. The algorithm includes a supervised and a self- supervised submodule. In the supervised submodule, the mapping relationship between low and high signal-to-noise ratio (SNR) data was constructed through mean square error loss function learning based on a small labeled dataset. In the self- supervised sub-module, an image recovery model was utilized to construct the loss function incorporating the prior information from a large unlabeled low SNR basic material image dataset, and the total variation (TV) model was used to to characterize the prior information of the images. The two submodules were combined to form the SLMD-Net method, and pre-clinical simulation data were used to validate the feasibility and effectiveness of the algorithm. Compared with the traditional model-driven quantitative imaging methods (FBP-DI, PWLS-PCG, and E3DTV), data-driven supervised-learning-based quantitative imaging methods (SUMD-Net and BFCNN), a material quantitative imaging method based on unsupervised learning (UNTV-Net) and semi-supervised learning-based cycle consistent generative adversarial network (Semi-CycleGAN), the proposed SLMD-Net method had better performance in both visual and quantitative assessments. For quantitative imaging of water and bone materials, the SLMD-Net method had the highest PSNR index (31.82 and 29.06), the highest FSIM index (0.95 and 0.90), and the lowest RMSE index (0.03 and 0.02), respectively) and achieved significantly higher image quality scores than the other 7 material decomposition methods (P < 0.05). The material quantitative imaging performance of SLMD-Net was close to that of the supervised network SUMD-Net trained with labeled data with a doubled size. A small labeled dataset and a large unlabeled low SNR material image dataset can be fully used to suppress noise amplification and artifacts in basic material decomposition in spectral CT and reduce the dependence on labeled data-driven network, which considers more realistic scenario in clinics.

Data-Driven Parameterized Corner Synthesis for Efficient Validation of Perception Systems for Autonomous Driving

Today's automotive cyber-physical systems for autonomous driving aim to enhance driving safety by replacing the uncertainties posed by human drivers with standard procedures of automated systems. However, the accuracy of in-vehicle perception systems may significantly vary under different operational conditions (e.g., fog density, light condition, etc.) and consequently degrade the reliability of autonomous driving. A perception system for autonomous driving must be carefully validated with an extremely large dataset collected under all possible operational conditions in order to ensure its robustness. The aforementioned dataset required for validation, however, is expensive or even impossible to acquire in practice, since most operational corners rarely occur in a real-world environment. In this paper, we propose to generate synthetic datasets at a variety of operational corners by using a parameterized cycle-consistent generative adversarial network (PCGAN) . The proposed PCGAN is able to learn from an image dataset recorded at real-world operational conditions with only a few samples at corners and synthesize a large dataset at a given operational corner. By taking STOP sign detection as an example, our numerical experiments demonstrate that the proposed approach is able to generate high-quality synthetic datasets to facilitate accurate validation.