Cycle Generative Adversarial Network Research Articles

Integrating Contrastive Learning and Cycle Generative Adversarial Networks for Non-invasive Fetal ECG Extraction.

Fetal electrocardiogram (FECG) contains crucial information about the fetus during pregnancy, making the extraction of FECG signal essential for monitoring fetal health. However, extracting FECG signal from abdominal electrocardiogram (AECG) poses several challenges: (1) FECG signal is often contaminated by noise, and (2) FECG signal is frequently overshadowed by high-amplitude maternal electrocardiogram (MECG). To address these issues and enhance the accuracy of signal extraction, this paper proposes an improved Cycle Generative Adversarial Networks (CycleGAN) with integrated contrastive learning for FECG signal extraction. The model introduces a dual-attention mechanism in the generator of the generative adversarial network, incorporating amulti-head self-attention(MSA) module and achannel-wise self-attention (CSA) module to enhance the quality of generated signals. Additionally, a contrastive triplet loss is integrated into the CycleGAN loss function, optimizing training to increase the similarity between the extracted FECG signal and the scalp fetal electrocardiogram. The proposed method is evaluated using the ADFECG dataset and the PCDB dataset both from the Physionet. In terms of signal extraction quality, Mean Squared Error is reduced to 0.036, Mean Absolute Error (MAE) to 0.009, and Pearson Correlation Coefficient reaches 0.924. When validating the model performance, Structural Similarity Index achieves 95.54%, Peak Signal-to-Noise Ratio (PSNR) reaches 38.87dB, and R-squared (R2) attains 95.12%. Furthermore, the positive predictive value (PPV), sensitivity (SEN) and F1-score for QRS wave cluster detection on the ADFECG dataset also reached 99.56%, 99.43% and 99.50%, respectively. On the PCDB dataset, the positive predictive value (PPV), sensitivity (SEN) and F1-score for QRS wave cluster detection also reached 98.24%, 98.60% and 98.42%, respectively. All of them are higher than other methods. Therefore, the proposed model has important applications in effective monitoring of fetal health during pregnancy.

Depth estimation from monocular endoscopy using simulation and image transfer approach

Obtaining accurate distance or depth information in endoscopy is crucial for the effective utilization of navigation systems. However, due to space constraints, incorporating depth cameras into endoscopic systems is often impractical. Our goal is to estimate depth images directly from endoscopic images using deep learning. This study presents a three-step methodology for training a depth-estimation network model. Initially, simulated endoscopy images and corresponding depth maps are generated using Unity based on a colon surface model obtained from segmented computed tomography colonography data. Subsequently, a cycle generative adversarial network model is employed to enhance the realism of the simulated endoscopy images. Finally, a deep learning model is trained using the synthesized endoscopy images and depth maps to estimate depths accurately. The performance of the proposed approach is evaluated and compared against prior studies utilizing unsupervised training methods. The results demonstrate the superior precision of the proposed technique in estimating depth images within endoscopy. The proposed depth estimation method holds promise for advancing the field by enabling enhanced navigation, improved lesion marking capabilities, and ultimately leading to better clinical outcomes.

Single-Image-Based Deep Learning for Precise Atomic Defect Identification.

Defect engineering is widely used to impart the desired functionalities on materials. Despite the widespread application of atomic-resolution scanning transmission electron microscopy (STEM), traditional methods for defect analysis are highly sensitive to random noise and human bias. While deep learning (DL) presents a viable alternative, it requires extensive amounts of training data with labeled ground truth. Herein, employing cycle generative adversarial networks (CycleGAN) and U-Nets, we propose a method based on a single experimental STEM image to tackle high annotation costs and image noise for defect detection. Not only atomic defects but also oxygen dopants in monolayer MoS2 are visualized. The method can be readily extended to other two-dimensional systems, as the training is based on unit-cell-level images. Therefore, our results outline novel ways to train the model with minimal data sets, offering great opportunities to fully exploit the power of DL in the materials science community.

Unsupervised deep learning bias correction of CMIP6 global ensemble precipitation predictions with cycle generative adversarial network

Abstract Climate change significantly impacts agricultural production, ecosystem stability, and socioeconomic development. Global climate models (GCMs) serve as the primary tool for simulating historical and future precipitation patterns. However, due to issues such as coarse resolution, boundary condition, and parameterization, model outputs require bias correction (BC). With the evolution of deep learning techniques, supervised convolutional neural network (CNN) frameworks have gained popularity in the area of climate model BC but face limitations in spatial correlation assumptions and data sparsity, particularly for extreme precipitation This study proposed an unsupervised learning approach using cycle generative adversarial network (CycleGAN) to correct the ensemble mean bias of models and compare its performance with CNN and Quantile Mapping methods. The results demonstrate that the proposed CycleGAN approach outperforms both CNN and Quantile Mapping in ensemble mean BC. It effectively learns the overall distribution of precipitation through an adversarial process and yields better extreme precipitation predictions.

DermoGAN: multi-task cycle generative adversarial networks for unsupervised automatic cell identification on in-vivo reflectance confocal microscopy images of the human epidermis.

Accurate identification of epidermal cells on reflectance confocal microscopy (RCM) images is important in the study of epidermal architecture and topology of both healthy and diseased skin. However, analysis of these images is currently done manually and therefore time-consuming and subject to human error and inter-expert interpretation. It is also hindered by low image quality due to noise and heterogeneity. We aimed to design an automated pipeline for the analysis of the epidermal structure from RCM images. Two attempts have been made at automatically localizing epidermal cells, called keratinocytes, on RCM images: the first is based on a rotationally symmetric error function mask, and the second on cell morphological features. Here, we propose a dual-task network to automatically identify keratinocytes on RCM images. Each task consists of a cycle generative adversarial network. The first task aims to translate real RCM images into binary images, thus learning the noise and texture model of RCM images, whereas the second task maps Gabor-filtered RCM images into binary images, learning the epidermal structure visible on RCM images. The combination of the two tasks allows one task to constrict the solution space of the other, thus improving overall results. We refine our cell identification by applying the pre-trained StarDist algorithm to detect star-convex shapes, thus closing any incomplete membranes and separating neighboring cells. The results are evaluated both on simulated data and manually annotated real RCM data. Accuracy is measured using recall and precision metrics, which is summarized as the -score. We demonstrate that the proposed fully unsupervised method successfully identifies keratinocytes on RCM images of the epidermis, with an accuracy on par with experts' cell identification, is not constrained by limited available annotated data, and can be extended to images acquired using various imaging techniques without retraining.

P-049 Identification of human spermatozoa with zona pellucida-binding ability using deep learning

Abstract Study question Can a deep learning model be used for identification of fertilization-competent human spermatozoa based on their distinctive morphological features? Summary answer Our results validated the clinical significance of image analysis of sperm morphology using deep learning to identify high-quality spermatozoa with zona pellucida (ZP)-binding ability. What is known already Sperm morphology is manually evaluated as part of the conventional semen analysis in clinical settings. However, the clinical predictive power of sperm morphology remains debatable due to its lack of association with the fertilization potential of spermatozoa. Human zona pellucida (ZP)-binding ability is a measurable, independent sperm parameter associated with in vitro fertilization (IVF) and pregnancy outcome. Supervised deep learning is an advanced branch to systematically identify the underlying relationship between the input images and designated output. A well-trained model with high generalization ability can generate plausible predictions for unseen images collected under the same condition as the training datasets. Study design, size, duration Human spermatozoa and oocytes were obtained with written consents from the donors attending our assisted reproduction program. Acrosome intact, ZP-bound spermatozoa were recovered after 30 min of gamete coincubation. Unbound spermatozoa were collected from clinical samples of spermatozoa with defective ZP-binding ability leading to fertilization failure following conventional insemination. Sperm images of infertile men (n = 165) with fertilization rates including 0-40% (low), 41-70% (intermediate) and 71-100% (high) following conventional insemination were collected for validation. Participants/materials, setting, methods Diff-Quik stained images of spermatozoa were processed by the K-means clustering algorithm to generate grey-scaled images of individual sperm heads. All sperm images of clinical samples were subjected an additional image conversion step using Cycle Generative Adversarial Network (CycleGAN) to ensure that their microenvironments are similar to those of the laboratory samples. A VGG 13 model was fine-tuned to classify a total of 1,083 images of ZP-bound and unbound spermatozoa over 50 epochs. Main results and the role of chance The newly fine-tuned VGG-13 model demonstrated excellent performance in classifying the ZP-bound and unbound spermatozoa, as reflected by the high sensitivity of 97.6%, specificity of 96.0%, and accuracy of 96.4%. It had a high discriminative power to distinguish between ZP-bound and unbound spermatozoa with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.992. Saliency maps indicated that the model focused on the sperm head and mid-pieces in all images, with a primary emphasis on the anterior regions (43.0%±5.0%). The model was further validated on a vast collection of over 40,000 sperm images collected from the aforementioned fertilization groups. Our results revealed that the high fertilization rate group had a significantly higher percentage of predicted ZP-binding ability than the low fertilization rate group, as determined by the model (16.8%±7.2 vs. 3.2%±1.4; p < 0.05). By analyzing sperm images from infertile men (n = 113) in the high and low fertilization rates groups, a clinical threshold of 4.8% was identified using ROS curve (AUC:0.970) and Youden’s index, which can be used to distinguish between ZP-bound and unbound spermatozoa of clinical samples with a specificity of 93.8% and a sensitivity of 90.9%. Limitations, reasons for caution Our deep learning model is only applicable to diff-quik stained images of spermatozoa captured at high magnification. Further investigation is needed to examine its classification performance on images captured by different platforms and settings, ensuring consistent and reliable prediction accuracy regardless of varying degrees of image quality. Wider implications of the findings Our newly fine-tuned model can be used to evaluate sperm fertilizing potential, irrespective of the parameters evaluated in conventional semen analysis and to identify normospermic patients with defective ZP-binding ability who may benefit from intracytoplasmic sperm injection (ICSI) to improve their fertilization outcome. Trial registration number not applicable

HDMA-CGAN: Advancing Image Style Transfer with Deep Learning

The integration of artificial intelligence (AI) and deep learning heralds a transformative era in pattern recognition and computer vision, notably in image style transfer. We introduce the hierarchical dynamic multi-attention cycle generative adversarial network (HDMA-CGAN), an innovative deep learning architecture poised to redefine image style transfer capabilities. HDMA-CGAN employs a novel multi-attention mechanism and color optimization strategies, enabling precise style replication with improved fidelity and vibrancy. Our model surpasses existing benchmarks in image quality, validated by leading metrics such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and Fréchet inception distance (FID). Although HDMA-CGAN advances the state of the art, it necessitates high computational resources and faces challenges with very high-resolution images. Future work could explore optimizing the model’s efficiency for real-time applications and extending its application to video content. This work enhances the tools available for visual content creation and digital media enhancement, leveraging advanced pattern recognition and AI techniques to significantly impact computer vision and image processing.

Data Augmentation Based on Image Translation for Bayesian Inference-Based Damage Diagnostics of Miter Gates

Abstract Structural health monitoring (SHM) data is the essential foundation for any SHM structural integrity assessment, including large civil infrastructure such as the miter gate application in this work. For some applications, the amount of monitoring data is limited due to various reasons such as a lack of sensor deployment investment, sensor reliability, inaccessibility of measurement locations, expensive duty cycles, etc. This limited data could result in uncertainty in structural health assessment. This paper addresses this challenging issue by proposing a data augmentation method based on image translation for Bayesian inference-based damage diagnostics. In particular, we translate the monitoring data of one miter gate to that of another, thereby increasing the volume of monitoring data available for assessing the structural health of a target miter gate. This translation starts with converting the monitoring data of different miter gates into images. After that, Cycle Generative Adversarial Networks (CycleGAN) are employed to accomplish the task of image translation among different miter gates. The translated images (i.e., synthetic monitoring data) are then used together with the true monitoring data for damage diagnostics. Two types of CycleGAN architectures are investigated and compared using a case study. Results of the case study show that the proposed data augmentation method can effectively improve the accuracy and confidence of damage diagnostics of miter gates. It demonstrates the potential of integrating synthetic data generation with probabilistic model updating in structural health monitoring.

IFSrNet: Multi-Scale IFS Feature-Guided Registration Network Using Multispectral Image-to-Image Translation

Multispectral image registration is the process of aligning the spatial regions of two images with different distributions. One of the main challenges it faces is to resolve the severe inconsistencies between the reference and target images. This paper presents a novel multispectral image registration network, Multi-scale Intuitionistic Fuzzy Set Feature-guided Registration Network (IFSrNet), to address multispectral image registration. IFSrNet generates pseudo-infrared images from visible images using Cycle Generative Adversarial Network (CycleGAN), which is equipped with a multi-head attention module. An end-to-end registration network encodes the input multispectral images with intuitionistic fuzzification, which employs an improved feature descriptor—Intuitionistic Fuzzy Set–Scale-Invariant Feature Transform (IFS-SIFT)—to guide its operation. The results of the image registration will be presented in a direct output. For this task we have also designed specialised loss functions. The results of the experiment demonstrate that IFSrNet outperforms existing registration methods in the Visible–IR dataset. IFSrNet has the potential to be employed as a novel image-to-image translation paradigm.

A Two-Stage Generative Model with CycleGAN and Joint Diffusion for MRI-based Brain Tumor Detection.

Accuratedetection and segmentation of brain tumors is critical for medical diagnosis. However, current supervised learning methods require extensively annotated images and the state-of-the-art generative models used in unsupervised methods often have limitations in covering the whole data distribution. In this paper, we propose a novel framework Two-Stage Generative Model (TSGM) that combines Cycle Generative Adversarial Network (CycleGAN) and Variance Exploding stochastic differential equation using joint probability (VE-JP) to improve brain tumor detection and segmentation. The CycleGAN is trained on unpaired data to generate abnormal images from healthy images as data prior. Then VE-JP is implemented to reconstruct healthy images using synthetic paired abnormal images as a guide, which alters only pathological regions but not regions of healthy. Notably, our method directly learned the joint probability distribution for conditional generation. The residual between input and reconstructed images suggests the abnormalities and a thresholding method is subsequently applied to obtain segmentation results. Furthermore, the multimodal results are weighted with different weights to improve the segmentation accuracy further. We validated our method on three datasets, and compared with other unsupervised methods for anomaly detection and segmentation. The DSC score of 0.8590 in BraTs2020 dataset, 0.6226 in ITCS dataset and 0.7403 in In-house dataset show that our method achieves better segmentation performance and has better generalization.

Restoration of magnetohydrodynamic-corrupted 12-lead electrocardiogram to enhance cardiac monitoring during magnetic resonance imaging

The Magnetohydrodynamic (MHD) effect on the bloodstream, induced by the static magnetic field of Magnetic Resonance Imaging (MRI) devices, distorts Electrocardiogram (ECG) components, and poses challenges to cardiac gating and monitoring during MRI examinations. Restoring ECG components from MHD-induced artifacts has always been a challenge, primarily due to the influence of numerous device-induced and physiological variables and the absence of paired ground truth clean ECG for validation. In this research, at first, we employ unpaired restoration of MHD-corrupted 12-lead ECG signals using one dimensional (1D) Cycle Generative Adversarial Networks (CycleGANs) for channel and patient orientation-specific restoration. Subsequently, a 1D-segmentation-based comprehensive restoration scheme is proposed based on the phase 1 outputs to simultaneously restore 12-channel ECG signals, irrespective of internal and external factors affecting ECG. By employing the proposed Magnetohydrodynamic Generative Adversarial Network (MHDGAN) and MHD-to-ECG Network (MHD2ECGNet) for unpaired and comprehensive restoration, we achieved an overall spectral correlation improvement of 92.56% and 77.64%, respectively. In R-peak detection, MHDGAN demonstrated an overall precision, recall, and F1-score of 98.14%, 94.12%, and 96.01%, respectively. The extracted heart rate (HR) and heart rate variability (HRV) from the restored waveforms closely match the ground truth, as confirmed through quantitative and qualitative assessments. The proposed framework has proven effective in restoring MHD-corrupted ECG waveforms, even in the absence of paired ground truth labels and amidst compounding distortions from multiple sources.

Data augmentation method for insulators based on Cycle GAN

Data augmentation is an important task of using existing data to expand data sets. Using generative countermeasure network technology to realize data augmentation has the advantages of high-quality generated samples, simple training, and fewer restrictions on the number of generated samples. However, in the field of transmission line insulator images, the freely synthesized samples are prone to produce fuzzy backgrounds and disordered samples of the main insulator features. To solve the above problems, this paper uses cycle generative adversarial network (Cycle-GAN) used for domain conversion in the generation countermeasure network as the initial framework and uses the self-attention mechanism and the channel attention mechanism to assist the conversion to realize the mutual conversion of different insulator samples. The attention module with prior knowledge is used to build the generation countermeasure network, and the GAN model with local controllable generation is built to realize the directional generation of insulator belt defect samples. The experimental results show that the samples obtained by this method are improved in a number of quality indicators, and the quality effect of the samples obtained is excellent, which has a reference value for the data expansion of insulator images.

A cycle generative adversarial network for generating synthetic contrast-enhanced computed tomographic images from non-contrast images in the internal jugular lymph node-bearing area.

The objectives of this study were to create a mutual conversion system between contrast-enhanced computed tomography (CECT) and non-CECT images using a cycle generative adversarial network (cycleGAN) for the internal jugular region. Image patches were cropped from CT images in 25 patients who underwent both CECT and non-CECT imaging. Using a cycleGAN, synthetic CECT and non-CECT images were generated from original non-CECT and CECT images, respectively. The peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) were calculated. Visual Turing tests were used to determine whether oral and maxillofacial radiologists could tell the difference between synthetic versus original images, and receiver operating characteristic (ROC) analyses were used to assess the radiologists' performances in discriminating lymph nodes from blood vessels. The PSNR of non-CECT images was higher than that of CECT images, while the SSIM was higher in CECT images. The Visual Turing test showed a higher perceptual quality in CECT images. The area under the ROC curve showed almost perfect performances in synthetic as well as original CECT images. In conclusion, synthetic CECT images created by cycleGAN appeared to have the potential to provide effective information in patients who could not receive contrast enhancement.

Enhanced cell segmentation with limited training datasets using cycle generative adversarial networks

Deep learning is transforming bioimage analysis, but its application in single-cell segmentation is limited by the lack of large, diverse annotated datasets. We addressed this by introducing a CycleGAN-based architecture, cGAN-Seg, that enhances the training of cell segmentation models with limited annotated datasets. During training, cGAN-Seg generates annotated synthetic phase-contrast or fluorescent images with morphological details and nuances closely mimicking real images. This increases the variability seen by the segmentation model, enhancing the authenticity of synthetic samples and thereby improving predictive accuracy and generalization. Experimental results show that cGAN-Seg significantly improves the performance of widely used segmentation models over conventional training techniques. Our approach has the potential to accelerate the development of foundation models for microscopy image analysis, indicating its significance in advancing bioimage analysis with efficient training methodologies.

Deep learning empowers photothermal microscopy with super-resolution capabilities.

In the past two decades, photothermal microscopy (PTM) has achieved sensitivity at the level of a single particle or molecule and has found applications in the fields of material science and biology. PTM is a far-field imaging method; its resolution is restricted by the diffraction limits. In our previous work, the modulated difference PTM (MDPTM) was proposed to improve the lateral resolution, but its resolution improvement was seriously constrained by information loss and artifacts. In this Letter, a deep learning approach of the cycle generative adversarial network (Cycle GAN) is employed for further improving the resolution of PTM, called DMDPTM. The point spread functions (PSFs) of both PTM and MDPTM are optimized and act as the second generator of Cycle GAN. Besides, the relationship between the sample's volume and the photothermal signal is utilized during dataset construction. The images of both PTM and MDPTM are utilized as the inputs of the Cycle GAN to incorporate more information. In the simulation, DMDPTM quantitatively distinguishes a distance of 60 nm between two nanoparticles (each with a diameter of 60 nm), demonstrating a 4.4-fold resolution enhancement over the conventional PTM. Experimentally, the super-resolution capability of DMDPTM is verified by restored images of Au nanoparticles, achieving the resolution of 114 nm. Finally, the DMDPTM is successfully employed for the imaging of carbon nanotubes. Therefore, the DMDPTM will serve as a powerful tool to improve the lateral resolution of PTM.

Faulty rolling bearing digital twin model and its application in fault diagnosis with imbalanced samples

The simulation signals generated by the bearing dynamics model have a big gap with the actual signals, which limits their efficacy in bearing fault diagnosis. Therefore, it is valuable to build an accurate digital twin model of faulty rolling bearing. Firstly, a multi-degree-of-freedom bearing fault dynamics model is constructed in the virtual space for generating the vibration responses of bearing parts. Then considering that the frequency spectrum contains more characteristic information than the time-domain signal, a frequency-domain bi-directional long short-term memory (Bi-LSTM) cycle generative adversarial network (CycleGAN) named FBC-GAN is proposed to construct the frequency-domain coupling mapping relationship between the multipart vibration responses and the measured signals. In the proposed network, Bi-LSTM is used for enhancing the feature extraction ability. Meantime, a new spectrum-constraint loss is proposed to ensure the frequency-domain mapping. Next, the simulated fault bearing signals close to the actual signals are generated by FBC-GAN and Fourier transform. Finally, the results of two experiments show the superiority of the proposed method over other advanced data augmentation methods in bearing fault diagnosis with the imbalanced samples.

Gan-based data augmentation to improve breast ultrasound and mammography mass classification

Data imbalance is a common problem in breast cancer diagnosis, to address this challenge, the research explores the use of Generative Adversarial Networks (GANs) to generate synthetic medical data. Various GAN methods, including Wasserstein GAN with Gradient Penalty (WGAN-GP), Cycle GAN, Conditional GAN, and Spectral Normalization GAN (SNGAN), were tested for data augmentation in breast regions of interest (ROIs) using mammography and ultrasound databases. The study employed real, synthetic, and hybrid ROIs (128x128 pixels) to train a Resnet network for classifying as benign (B) or malignant (M) classes. The quality and diversity of the synthetic data were assessed using several metrics: Fréchet Inception Distance (FID), Kernel Inception Distance (KID), Structural Similarity Index (SSIM), Multi-Scale SSIM (MS-SSIM), Blind Reference Image Spatial Quality Evaluator (BRISQUE), Naturalness Image Quality Evaluator (NIQE), and Perception-based Image Quality Evaluator (PIQE).Results revealed that the SNGAN model (FID = 52.89) was most effective for augmenting mammography data, while CGAN (FID = 116.03) excelled with ultrasound data. Cycle GAN and WGAN-GP, though demonstrating lower KID values, did not perform better than SNGAN and CGAN. The lower average MS-SSIM values suggested that SNGAN and CGAN produced a high diversity of synthetic images. However, lower SSIM, BRISQUE, NIQE, and PIQE values indicated poor quality in both real and synthetic images. Classification results showed high accuracy without data augmentation in both US (93.1 %B/94.9 %M) and mammography (80.9 %B/76.9 %M). The research concludes that preprocessing and characterizing ROIs by abnormality type is crucial to generate diverse synthetic data and improve accuracy in the classification process using combined GANs and CNN models.

Data enhancement method based on cyclic generation adversarial network

This study introduces a novel data augmentation technique employing Cycle Generative Adversarial Networks (CycleGAN) to mitigate the challenges posed by the paucity of image datasets in deep learning domains. Through the adept training of a CycleGAN model, this method substantially enriches image datasets, thereby enhancing the efficiency of deep learning models in target detection tasks. Distinct from conventional approaches, our strategy incorporates advanced activation functions, relative discriminators, and residual connections, which collectively foster greater image diversity and mitigate mode collapse, all while maintaining a low computational overhead. Evaluations conducted on diverse datasets, including MNIST, Synthetic Aperture Radar (SAR), and medical blood cell images, demonstrate the method's superior augmentation capabilities compared to traditional Deep Convolutional GAN (DCGAN) techniques, underscoring its efficacy and potential utility in preprocessing for deep learning applications.

Developing an Improved Cycle Architecture for AI-Based Generation of New Structures Aimed at Drug Discovery.

Drug discovery involves a crucial step of optimizing molecules with the desired structural groups. In the domain of computer-aided drug discovery, deep learning has emerged as a prominent technique in molecular modeling. Deep generative models, based on deep learning, play a crucial role in generating novel molecules when optimizing molecules. However, many existing molecular generative models have limitations as they solely process input information in a forward way. To overcome this limitation, we propose an improved generative model called BD-CycleGAN, which incorporates BiLSTM (bidirectional long short-term memory) and Mol-CycleGAN (molecular cycle generative adversarial network) to preserve the information of molecular input. To evaluate the proposed model, we assess its performance by analyzing the structural distribution and evaluation matrices of generated molecules in the process of structural transformation. The results demonstrate that the BD-CycleGAN model achieves a higher success rate and exhibits increased diversity in molecular generation. Furthermore, we demonstrate its application in molecular docking, where it successfully increases the docking score for the generated molecules. The proposed BD-CycleGAN architecture harnesses the power of deep learning to facilitate the generation of molecules with desired structural features, thus offering promising advancements in the field of drug discovery processes.

Minimum imaging dose for deep learning-based pelvic synthetic computed tomography generation from cone beam images

Background and purposeDaily cone-beam computed tomography (CBCT) in image-guided radiotherapy administers radiation exposure and subjects patients to secondary cancer risk. Reducing imaging dose remains challenging as image quality deteriorates. We investigated three imaging dose levels by reducing projections and correcting images using two deep learning algorithms, aiming at identifying the lowest achievable imaging dose. Materials and methodsCBCTs were reconstructed with 100%, 25%, 15% and 10% projections. Models were trained (30), validated (3) and tested (8) with prostate cancer patient datasets. We optimized and compared the performance of 1) a cycle generative adversarial network (cycleGAN) with residual connection and 2) a contrastive unpaired translation network (CUT) to generate synthetic computed tomography (sCT) from reduced imaging dose CBCTs. Volumetric modulated arc therapy plans were optimized on a reference intensity-corrected full dose CBCTcor and recalculated on sCTs. Hounsfield unit (HU) and positioning accuracy were evaluated. Bladder and rectum were manually delineated to determine anatomical fidelity. ResultsAll sCTs achieved average mean absolute mean absolute error/structural similarity index measure/peak signal-to-noise ratio of ⩽59HU/⩾0.94/⩾33 dB. All dose-volume histogram parameter differences were within 2 Gy or 2%. Positioning differences were ⩽0.30 mm or 0.30°. cycleGAN with Dice similarity coefficients (DSC) for bladder/rectum of ⩾0.85/⩾0.81 performed better than CUT (⩾0.83/⩾0.76). A significantly lower DSC accuracy was observed for 15% and 10% sCTs. cycleGAN performed better than CUT for contouring, however both yielded comparable outcomes in other evaluations. ConclusionsCTs based on different CBCT doses using cycleGAN and CUT were investigated. Based on segmentation accuracy, 25% is the minimum imaging dose.