Generative Network Architecture Research Articles

CLGAN: a generative adversarial network (GAN)-based video prediction model for precipitation nowcasting

Abstract. The prediction of precipitation patterns up to 2 h ahead, also known as precipitation nowcasting, at high spatiotemporal resolutions is of great relevance in weather-dependent decision-making and early warning systems. In this study, we are aiming to provide an efficient and easy-to-understand deep neural network – CLGAN (convolutional long short-term memory generative adversarial network) – to improve the nowcasting skills of heavy precipitation events. The model constitutes a generative adversarial network (GAN) architecture, whose generator is built upon a u-shaped encoder–decoder network (U-Net) and is equipped with recurrent long short-term memory (LSTM) cells to capture spatiotemporal features. The optical flow model DenseRotation and the competitive video prediction models ConvLSTM (convolutional LSTM) and PredRNN-v2 (predictive recurrent neural network version 2) are used as the competitors. A series of evaluation metrics, including the root mean square error, the critical success index, the fractions skill score, and object-based diagnostic evaluation, are utilized for a comprehensive comparison against competing baseline models. We show that CLGAN outperforms the competitors in terms of scores for dichotomous events and object-based diagnostics. A sensitivity analysis on the weight of the GAN component indicates that the GAN-based architecture helps to capture heavy precipitation events. The results encourage future work based on the proposed CLGAN architecture to improve the precipitation nowcasting and early warning systems.

Multi-contrast digital histopathology of mouse organs using quantitative phase imaging and virtual staining.

Quantitative phase imaging (QPI) has emerged as a new digital histopathologic tool as it provides structural information of conventional slide without staining process. It is also capable of imaging biological tissue sections with sub-nanometer sensitivity and classifying them using light scattering properties. Here we extend its capability further by using optical scattering properties as imaging contrast in a wide-field QPI. In our first step towards validation, QPI images of 10 major organs of a wild-type mouse have been obtained followed by H&E-stained images of the corresponding tissue sections. Furthermore, we utilized deep learning model based on generative adversarial network (GAN) architecture for virtual staining of phase delay images to a H&E-equivalent brightfield (BF) image analogues. Using the structural similarity index, we demonstrate similarities between virtually stained and H&E histology images. Whereas the scattering-based maps look rather similar to QPI phase maps in the kidney, the brain images show significant improvement over QPI with clear demarcation of features across all regions. Since our technology provides not only structural information but also unique optical property maps, it could potentially become a fast and contrast-enriched histopathology technique.

Application of generative adversarial neural networks for the formation of databases in scanning tunneling microscopy

We discuss the development of a technique for automatic generation of databases of images obtained with a scanning tunneling microscope. An analysis of state-of-the-art methods and means of automatic processing of images obtained from probe and electron microscopes is carried out. We proposed using generative-adversarial networks for generating images taken with a scanning tunneling microscope to form training databases of images. A process of training and comparison of deep convolutional generative adversarial network (DCGAN) architectures using the OpenCV and Keras libraries together with TensorFlow is described, with the best of them identified by computing the metrics IS, FID, and KID. The scaling of images obtained from DCGAN is performed using a method of fine tuning of a super-resolution generative adversarial neural network (SRGAN) and bilinear interpolation based on the Python programming language. An analysis of calculated quantitative metrics values shows that the best results of image generation are obtained using DCGAN96 and SRGAN. It is found that FID and KID metric values for SRGAN method are better than values for bilinear interpolation in all cases except for DCGAN32. All calculations are performed on a GTX GeForce 1070 video card. A method for automatic generation of a scanning tunneling microscope image database based on the stepwise application of DCGAN and SRGAN is developed. Results of generation and comparison of the original image, the one obtained with DCGAN96 and the enlarged image with SRGAN are presented.

Medigan: a Python library of pretrained generative models for medical image synthesis.

Deep learning has shown great promise as the backbone of clinical decision support systems. Synthetic data generated by generative models can enhance the performance and capabilities of data-hungry deep learning models. However, there is (1)limited availability of (synthetic) datasets and (2)generative models are complex to train, which hinders their adoption in research and clinical applications. To reduce this entry barrier, we explore generative model sharing to allow more researchers to access, generate, and benefit from synthetic data. We propose medigan, a one-stop shop for pretrained generative models implemented as an open-source framework-agnostic Python library. After gathering end-user requirements, design decisions based on usability, technical feasibility, and scalability are formulated. Subsequently, we implement medigan based on modular components for generative model (i)execution, (ii)visualization, (iii)search & ranking, and (iv)contribution. We integrate pretrained models with applications across modalities such as mammography, endoscopy, x-ray, and MRI. The scalability and design of the library are demonstrated by its growing number of integrated and readily-usable pretrained generative models, which include 21 models utilizing nine different generative adversarial network architectures trained on 11 different datasets. We further analyze three medigan applications, which include (a)enabling community-wide sharing of restricted data, (b)investigating generative model evaluation metrics, and (c)improving clinical downstream tasks. In (b), we extract Fréchet inception distances (FID) demonstrating FID variability based on image normalization and radiology-specific feature extractors. medigan allows researchers and developers to create, increase, and domain-adapt their training data in just a few lines of code. Capable of enriching and accelerating the development of clinical machine learning models, we show medigan's viability as platform for generative model sharing. Our multimodel synthetic data experiments uncover standards for assessing and reporting metrics, such as FID, in image synthesis studies.

Data Augmentation and Classification of Sea–Land Clutter for Over-the-Horizon Radar Using AC-VAEGAN

In the sea-land clutter classification of sky-wave over-the-horizon-radar (OTHR), the imbalanced and scarce data leads to a poor performance of the deep learning-based classification model. To solve this problem, this paper proposes an improved auxiliary classifier generative adversarial network (AC-GAN) architecture, namely auxiliary classifier variational autoencoder generative adversarial network (AC-VAEGAN). AC-VAEGAN can synthesize higher quality sea-land clutter samples than AC-GAN and serve as an effective tool for data augmentation. Specifically, a one-dimensional convolutional AC-VAEGAN architecture is designed to synthesize sea-land clutter samples. Additionally, an evaluation method combining both traditional evaluation of GAN domain and statistical evaluation of signal domain is proposed to evaluate the quality of synthetic samples. Using a dataset of OTHR sea-land clutter, both the quality of the synthetic samples and the performance of data augmentation of AC-VAEGAN are verified. Further, the effect of AC-VAEGAN as a data augmentation method on the classification performance of imbalanced and scarce sea-land clutter samples is validated. The experiment results show that the quality of samples synthesized by AC-VAEGAN is better than those synthesized by the state-of-the-art GAN-based methods, and the data augmentation method with AC-VAEGAN is able to improve the classification performance in the case of imbalanced and scarce sea-land clutter samples.

A hierarchical GAN method with ensemble CNN for accurate nodule detection.

Lung cancer can evolve into one of the deadliest diseases whose early detection is one of the major survival factors. However, early detection is a challenging task due to the unclear structure, shape, and the size of the nodule. Hence, radiologists need automated tools to make accurate decisions. This paper develops a new approach based on generative adversarial network (GAN) architecture for nodule detection to propose a two-step GAN model containing lung segmentation and nodule localization. The first generator comprises a U-net network, while the second utilizes a mask R-CNN. The task of lung segmentation involves a two-class classification of the pixels in each image, categorizing lung pixels in one class and the rest in the other. The classifier becomes imbalanced due to numerous non-lung pixels, decreasing the model performance. This problem is resolved by using the focal loss function for training the generator. Moreover, a new loss function is developed as the nodule localization generator to enhance the diagnosis quality. Discriminator nets are implemented in GANs as an ensemble of convolutional neural networks (ECNNs), using multiple CNNs and connecting their outputs to make a final decision. Several experiments are designed to assess the model on the well-known LUNA dataset. The experiments indicate that the proposed model can reduce the error of the state-of-the-art models on the IoU criterion by about 35 and 16% for lung segmentation and nodule localization, respectively. Unlike recent studies, the proposed method considers two loss functions for generators, further promoting the goal achievements. Moreover, the network of discriminators is regarded as ECNNs, generating rich features for decisions.

A tomographic spherical mass map emulator of the KiDS-1000 survey using conditional generative adversarial networks

Large sets of matter density simulations are becoming increasingly important in large-scale structure cosmology. Matter power spectra emulators, such as the Euclid Emulator and CosmicEmu, are trained on simulations to correct the non-linear part of the power spectrum. Map-based analyses retrieve additional non-Gaussian information from the density field, whether through human-designed statistics such as peak counts, or via machine learning methods such as convolutional neural networks. The simulations required for these methods are very resource-intensive, both in terms of computing time and storage. This creates a computational bottleneck for future cosmological analyses, as well as an entry barrier for testing new, innovative ideas in the area of cosmological information retrieval. Map-level density field emulators, based on deep generative models, have recently been proposed to address these challenges. In this work, we present a novel mass map emulator of the KiDS-1000 survey footprint, which generates noise-free spherical maps in a fraction of a second. It takes a set of cosmological parameters (Ω M , σ 8) as input and produces a consistent set of 5 maps, corresponding to the KiDS-1000 tomographic redshift bins. To construct the emulator, we use a conditional generative adversarial network architecture and the spherical convolutional neural network DeepSphere, and train it on N-body-simulated mass maps. We compare its performance using an array of quantitative comparison metrics: angular power spectra Cℓ , pixel/peaks distributions, Cℓ correlation matrices, and Structural Similarity Index. Overall, the average agreement on these summary statistics is <10% for the cosmologies at the centre of the simulation grid, and degrades slightly on grid edges. However, the quality of the generated maps is worse at high negative κ values or large scale, which can significantly affect summaries sensitive to such observables. Finally, we perform a mock cosmological parameter estimation using the emulator and the original simulation set. We find good agreement in these constraints, for both likelihood and likelihood-free approaches. The emulator is available at tfhub.dev/cosmo-group-ethz/models/kids-cgan.

Uncertainty-aware physics-driven deep learning network for free-breathing liver fat and R2 * quantification using self-gated stack-of-radial MRI.

To develop a deep learning-based method for rapid liver proton-density fat fraction (PDFF) and R2 * quantification with built-in uncertainty estimation using self-gated free-breathing stack-of-radial MRI. This work developed an uncertainty-aware physics-driven deep learning network (UP-Net) to (1) suppress radial streaking artifacts because of undersampling after self-gating, (2) calculate accurate quantitative maps, and (3) provide pixel-wise uncertainty maps. UP-Net incorporated a phase augmentation strategy, generative adversarial network architecture, and an MRI physics loss term based on a fat-water and R2 * signal model. UP-Net was trained and tested using free-breathing multi-echo stack-of-radial MRI data from 105 subjects. UP-Net uncertainty scores were calibrated in a validation dataset and used to predict quantification errors for liver PDFF and R2 * in a testing dataset. Compared with images reconstructed using compressed sensing (CS), UP-Net achieved structural similarity index >0.87 and normalized root mean squared error <0.18. Compared with reference quantitative maps generated using CS and graph-cut (GC) algorithms, UP-Net achieved low mean differences (MD) for liver PDFF (-0.36%) and R2 * (-0.37 s-1 ). Compared with breath-holding Cartesian MRI results, UP-Net achieved low MD for liver PDFF (0.53%) and R2 * (6.75 s-1 ). UP-Net uncertainty scores predicted absolute liver PDFF and R2 * errors with low MD of 0.27% and 0.12 s-1 compared to CS + GC results. The computational time for UP-Net was 79 ms/slice, whereas CS + GC required 3.2 min/slice. UP-Net rapidly calculates accurate liver PDFF and R2 * maps from self-gated free-breathing stack-of-radial MRI. The pixel-wise uncertainty maps from UP-Net predict quantification errors in the liver.

Analysis of convolutional neural network image classifiers in a hierarchical max-pooling model with additional local pooling

In practical applications of image classification, methods based on convolutional neural networks are used as standard nowadays. However, the success of these methods is not explained sufficiently from a theoretical point of view. For this, a statistical model for image classification, namely a hierarchical max-pooling model with additional local pooling, is introduced. Here the additional local pooling enables the hierarchical model to combine parts of the image which have a variable relative distance towards each other. Various convolutional neural network image classifiers are introduced and analyzed in view of their rate of convergence of the misclassification risk of the estimates towards the optimal misclassification risk. This analysis provides a theoretical explanation why general convolutional neural network architectures that include some kind of local pooling layers are useful in some image classification situations and gives theoretical hints for choosing the right network architecture. Furthermore, the finite sample size performance of these network architectures is illustrated by applying them to simulated and real data.

Series-Parallel Generative Adversarial Network Architecture for Translating from Fundus Structure Image to Fluorescence Angiography

Although fundus fluorescein angiography (FFA) is a very effective retinal imaging tool for ophthalmic diagnosis, the requirement of intravenous injection of harmful fluorescein dye limits its application. As a screening diagnostic method that reduces the frequency of intravenous injection, a series-parallel generative adversarial network (GAN) architecture for translating fundus structure image to FFA images is proposed herein, using deep learning-based software that only needs an intravenous injection for the training process. Firstly, the fundus structure image and the corresponding FFA images of three phases are collected. Secondly, our series-parallel GAN is trained to translate FFA images from fundus structure image with the supervision of FFA images. Thirdly, the trained series-parallel GAN model is used to translate FFA images by only using fundus structure image. By comparing the FFA images translated by our algorithm, Sequence GAN, pix2pix, and cycleGAN, we show the advancement of our algorithm. To further confirm the advancements of our algorithm, we evaluate the peak signal-to-noise ratio (PSNR), structural similarity (SSIM) index, and mean-squared error (MSE) of our algorithm, Sequence GAN, pix2pix, and cycleGAN. To demonstrate the performance of our method, we show some typical FFA images translated by our algorithm.

Generative and reinforcement learning approaches for the automated de novo design of bioactive compounds

Deep generative neural networks have been used increasingly in computational chemistry for de novo design of molecules with desired properties. Many deep learning approaches employ reinforcement learning for optimizing the target properties of the generated molecules. However, the success of this approach is often hampered by the problem of sparse rewards as the majority of the generated molecules are expectedly predicted as inactives. We propose several technical innovations to address this problem and improve the balance between exploration and exploitation modes in reinforcement learning. In a proof-of-concept study, we demonstrate the application of the deep generative recurrent neural network architecture enhanced by several proposed technical tricks to design inhibitors of the epidermal growth factor (EGFR) and further experimentally validate their potency. The proposed technical solutions are expected to substantially improve the success rate of finding novel bioactive compounds for specific biological targets using generative and reinforcement learning approaches.

A Multi-Purpose Shallow Convolutional Neural Network for Chart Images.

Charts are often used for the graphical representation of tabular data. Due to their vast expansion in various fields, it is necessary to develop computer algorithms that can easily retrieve and process information from chart images in a helpful way. Convolutional neural networks (CNNs) have succeeded in various image processing and classification tasks. Nevertheless, the success of training neural networks in terms of result accuracy and computational requirements requires careful construction of the network layers’ and networks’ parameters. We propose a novel Shallow Convolutional Neural Network (SCNN) architecture for chart-type classification and image generation. We validate the proposed novel network by using it in three different models. The first use case is a traditional SCNN classifier where the model achieves average classification accuracy of 97.14%. The second use case consists of two previously introduced SCNN-based models in parallel, with the same configuration, shared weights, and parameters mirrored and updated in both models. The model achieves average classification accuracy of 100%. The third proposed use case consists of two distinct models, a generator and a discriminator, which are both trained simultaneously using an adversarial process. The generated chart images are plausible to the originals. Extensive experimental analysis end evaluation is provided for the classification task of seven chart classes. The results show that the proposed SCNN is a powerful tool for chart image classification and generation, comparable with Deep Convolutional Neural Networks (DCNNs) but with higher efficiency, reduced computational time, and space complexity.

Manifold-aware synthesis of high-resolution diffusion from structural imaging.

The physical and clinical constraints surrounding diffusion-weighted imaging (DWI) often limit the spatial resolution of the produced images to voxels up to eight times larger than those of T1w images. The detailed information contained in accessible high-resolution T1w images could help in the synthesis of diffusion images with a greater level of detail. However, the non-Euclidean nature of diffusion imaging hinders current deep generative models from synthesizing physically plausible images. In this work, we propose the first Riemannian network architecture for the direct generation of diffusion tensors (DT) and diffusion orientation distribution functions (dODFs) from high-resolution T1w images. Our integration of the log-Euclidean Metric into a learning objective guarantees, unlike standard Euclidean networks, the mathematically-valid synthesis of diffusion. Furthermore, our approach improves the fractional anisotropy mean squared error (FA MSE) between the synthesized diffusion and the ground-truth by more than 23% and the cosine similarity between principal directions by almost 5% when compared to our baselines. We validate our generated diffusion by comparing the resulting tractograms to our expected real data. We observe similar fiber bundles with streamlines having <3% difference in length, <1% difference in volume, and a visually close shape. While our method is able to generate diffusion images from structural inputs in a high-resolution space within 15 s, we acknowledge and discuss the limits of diffusion inference solely relying on T1w images. Our results nonetheless suggest a relationship between the high-level geometry of the brain and its overall white matter architecture that remains to be explored.

Constrained and unconstrained deep image prior optimization models with automatic regularization.

Deep Image Prior (DIP) is currently among the most efficient unsupervised deep learning based methods for ill-posed inverse problems in imaging. This novel framework relies on the implicit regularization provided by representing images as the output of generative Convolutional Neural Network (CNN) architectures. So far, DIP has been shown to be an effective approach when combined with classical and novel regularizers. Unfortunately, to obtain appropriate solutions, all the models proposed up to now require an accurate estimate of the regularization parameter. To overcome this difficulty, we consider a locally adapted regularized unconstrained model whose local regularization parameters are automatically estimated for additively separable regularizers. Moreover, we propose a novel constrained formulation in analogy to Morozov’s discrepancy principle which enables the application of a broader range of regularizers. Both the unconstrained and the constrained models are solved via the proximal gradient descent-ascent method. Numerical results demonstrate the robustness with respect to image content, noise levels and hyperparameters of the proposed models on both denoising and deblurring of simulated as well as real natural and medical images.

Deep convolutional generative adversarial networks for modeling complex hydrological structures in Monte-Carlo simulation

• A hybrid method MC-GAN is proposed for hydrological modeling. • An improved DCGAN model is embedded into the Monte-Carlo simulation process. • Joint loss functions are used to improve the performance of MC-GAN. • Various hydrological patterns can be reproduced efficiently. • Multiple-scale realizations can be obtained by using the same trained model. Characterization of complex subsurface structures is challenging due to the demand to preserve geological realism of the training images in earth and environmental sciences. In this work, we propose a novel method to reconstruct complex hydrological structures by using deep convolutional generative adversarial networks (DCGAN) in the Monte-Carlo simulation process, named MC-GAN. Network architectures for reconstructing both two-dimensional (2D) and three-dimensional (3D) complex spatial structures are provided in this method. We first exploit the robust DCGAN to reproduce abundant and various spatial pattern blocks. Then, we combine the various heterogeneous patterns to reconstruct a complex hydrological structure by using the Monte-Carlo stochastic simulation process. The method is able to represent multiple-scale spatial structures under the premise of using the same generative adversarial network architecture. It not only ensures the simulation efficiency, but also makes the heterogeneous patterns in the realizations more diverse. Three sets of training images were used to test the capability of the proposed method. The experiment results demonstrate that our method can accurately characterize complex heterogeneous spatial structures. At the same time, the trained deep learning model can be reused effectively to generate multiple-scale spatial structures.

Generation of Individualized Synthetic Data for Augmentation of the Type 1 Diabetes Data Sets Using Deep Learning Models.

In this paper, we present a methodology based on generative adversarial network architecture to generate synthetic data sets with the intention of augmenting continuous glucose monitor data from individual patients. We use these synthetic data with the aim of improving the overall performance of prediction models based on machine learning techniques. Experiments were performed on two cohorts of patients suffering from type 1 diabetes mellitus with significant differences in their clinical outcomes. In the first contribution, we have demonstrated that the chosen methodology is able to replicate the intrinsic characteristics of individual patients following the statistical distributions of the original data. Next, a second contribution demonstrates the potential of synthetic data to improve the performance of machine learning approaches by testing and comparing different prediction models for the problem of predicting nocturnal hypoglycemic events in type 1 diabetic patients. The results obtained for both generative and predictive models are quite encouraging and set a precedent in the use of generative techniques to train new machine learning models.

Detection of Adversarial DDoS Attacks Using Symmetric Defense Generative Adversarial Networks

DDoS (distributed denial of service) attacks consist of a large number of compromised computer systems that launch joint attacks at a targeted victim, such as a server, website, or other network equipment, simultaneously. DDoS has become a widespread and severe threat to the integrity of computer networks. DDoS can lead to system paralysis, making it difficult to troubleshoot. As a critical component of the creation of an integrated defensive system, it is essential to detect DDoS attacks as early as possible. With the popularization of artificial intelligence, more and more researchers have applied machine learning (ML) and deep learning (DL) to the detection of DDoS attacks and have achieved satisfactory accomplishments. The complexity and sophistication of DDoS attacks have continuously increased and evolved since the first DDoS attack was reported in 1996. Regarding the headways in this problem, a new type of DDoS attack, named adversarial DDoS attack, is investigated in this study. The generating adversarial DDoS traffic is carried out using a symmetric generative adversarial network (GAN) architecture called CycleGAN to demonstrate the severe impact of adversarial DDoS attacks. Experiment results reveal that the synthesized attack can easily penetrate ML-based detection systems, including RF (random forest), KNN (k-nearest neighbor), SVM (support vector machine), and naïve Bayes. These alarming results intimate the urgent need for countermeasures against adversarial DDoS attacks. We present a novel DDoS detection framework that incorporates GAN with a symmetrically built generator and discriminator defense system (SDGAN) to deal with these problems. Both symmetric discriminators are intended to simultaneously identify adversarial DDoS traffic. As demonstrated by the experimental results, the suggested SDGAN can be an effective solution against adversarial DDoS attacks. We train SDGAN on adversarial DDoS data generated by CycleGAN and compare it to four previous machine learning-based detection systems. SDGAN outperformed the other machine learning models, with a TPR (true positive rate) of 87.2%, proving its protection ability. Additionally, a more comprehensive test was undertaken to evaluate SDGAN’s capacity to defend against unseen adversarial threats. SDGAN was evaluated using non-training data-generated adversarial traffic. SDGAN remained effective, with a TPR of around 70.9%, compared to RF’s 9.4%.

Computed tomography image generation from magnetic resonance imaging using Wasserstein metric for MR‐only radiation therapy

AbstractMagnetic resonance imaging (MRI) and computed tomography (CT) are the prevalent imaging techniques used in treatment planning in radiation therapy. Since MR‐only radiation therapy planning (RTP) is needed in the future for new technologies like MR‐LINAC (medical linear accelerator), MR to CT synthesis model benefits in CT synthesis from MR images generated via MR‐LINAC. A Wasserstein generative adversarial network (WGAN) architecture with a residual UNet based generator and a patch‐based discriminator is proposed in this paper. The WGAN uses the Wasserstein metric with gradient penalty along with mean absolute error (MAE) as the loss function. The WGAN is trained and tested on an NVIDIA Tesla V100 GPU server using mutually aligned MR‐CT brain images of 26 patients. The proposed model generates synthetic CTs with an average PSNR of 31.09 dB, SSIM of 0.9265, MAE of 48.39 HU, and RMSE of 172.75 HU. The treatment planning was performed on real and synthetic CTs. The dose distributions of the synthetic CTs obtained were similar to that of the corresponding real CTs. The synthetic CTs exhibit good visual quality as well. The experimental outcomes and image quality of the model surpass many of the state‐of‐the‐art methods. The proposed CT synthesizer supports MR‐only radiation therapy while minimizing treatment costs and radiation exposure. A significant increase in the training set can further improve the CT prediction accuracy of the proposed model.

Synthetic 18F-FDG PET Image Generation Using a Combination of Biomathematical Modeling and Machine Learning.

Simple SummaryTraining computer-assisted algorithms on medical images, particularly 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) due to its excellent diagnostic accuracy, is difficult, considering small/fragmented samples or privacy concerns. In computer-vision, deep learning-based models, unlike the conventional data augmentation methods, are highly sought after for creating massive medical samples. For this reason, we developed a multi-scale computational framework to generate synthetic 18F-FDG PET images similar to the real ones in different stages of solid tumor growth and angiogenesis. The framework is developed based on the bio-physiological phenomena of FDG radiotracer uptake in solid tumors using a biomathematical model and a generative adversarial network (GAN)-based architecture. The non-invasive augmented 18F-FDG PET images can be used in clinical practice without the need to manage the patient data. In addition, our spatiotemporal mathematical model can calculate the distribution of various radiopharmaceuticals in different tumor-associated vasculatures.No previous works have attempted to combine generative adversarial network (GAN) architectures and the biomathematical modeling of positron emission tomography (PET) radiotracer uptake in tumors to generate extra training samples. Here, we developed a novel computational model to produce synthetic 18F-fluorodeoxyglucose (18F-FDG) PET images of solid tumors in different stages of progression and angiogenesis. First, a comprehensive biomathematical model is employed for creating tumor-induced angiogenesis, intravascular and extravascular fluid flow, as well as modeling of the transport phenomena and reaction processes of 18F-FDG in a tumor microenvironment. Then, a deep convolutional GAN (DCGAN) model is employed for producing synthetic PET images using 170 input images of 18F-FDG uptake in each of 10 different tumor microvascular networks. The interstitial fluid parameters and spatiotemporal distribution of 18F-FDG uptake in tumor and healthy tissues have been compared against previously published numerical and experimental studies, indicating the accuracy of the model. The structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) of the generated PET sample and the experimental one are 0.72 and 28.53, respectively. Our results demonstrate that a combination of biomathematical modeling and GAN-based augmentation models provides a robust framework for the non-invasive and accurate generation of synthetic PET images of solid tumors in different stages.

Diagnostic-Quality Guided Wave Signals Synthesized Using Generative Adversarial Neural Networks

Guided waves are a potent tool in structural health monitoring, with promising machine learning algorithm applications due to the complexity of their signals. However, these algorithms usually require copious amounts of data to be trained. Collecting the correct amount and distribution of data is costly and time-consuming, and sometimes even borderline impossible due to the necessity of introducing damage to vital machinery to collect signals for various damaged scenarios. This data scarcity problem is not unique to guided waves or structural health monitoring, and has been partly addressed in the field of computer vision using generative adversarial neural networks. These networks generate synthetic data samples based on the distribution of the data they were trained on. Though there are multiple researched methods for simulating guided wave signals, the problem is not yet solved. This work presents a generative adversarial network architecture for guided waves generation and showcases its capabilities when working with a series of pitch-catch experiments from the OpenGuidedWaves database. The network correctly generates random signals and can accurately reconstruct signals it has not seen during training. The potential of synthetic data to be used for training other algorithms was confirmed in a simple damage detection scenario, with the classifiers trained exclusively on synthetic data and evaluated on real signals. As a side effect of the signal reconstruction process, the network can also compress the signals by 98.44% while retaining the damage index information they carry.