Hasil untuk "deep learning"

Hye Kyung Jeon, Gwang Ha Kim

Accurate lesion size measurement is essential in endoscopic practice as it influences treatment strategies, surveillance decisions, and clinical outcomes, especially in colorectal polyps. Traditional measurement techniques, including visual estimation and biopsy forceps, have significant interobserver variability and procedural inefficiencies. Recent advancements in digital measurement technologies, including virtual scale endoscopy (VSE) and artificial intelligence (AI)-assisted virtual rulers, have addressed these limitations. VSE projects a virtual scale onto endoscopic images, enhancing measurement precision and reducing variability. Several studies have demonstrated its superior accuracy compared with conventional methods; however, limitations such as increased procedure time and operator training requirements persist. AI-assisted virtual rulers utilize deep learning algorithms to automate lesion size estimation, significantly improving reproducibility and diagnostic reliability. Although these technologies offer promising improvements, challenges remain, including real-time integration, standardization, and regulatory approval. Future research should focus on refining AI models, expanding validation studies, and optimizing their usability in routine practice. A hybrid approach that combines AI automation with real-time digital tools may enhance the precision and efficiency of endoscopic lesion assessment, ultimately improving patient outcomes.

Zhejian Zhang, Nan Li, Shude Mao et al.

IntroductionGalaxy cluster-scale strong gravitational lensing systems are rare yet valuable tools for investigating dark matter and dark energy, as well as providing the opportunity to study the distant universe at flux levels and spatial resolutions that would otherwise be unavailable. Large-scale imaging surveys present unprecedented opportunities to expand the sample of cluster lenses.MethodsIn this study, we adopt a deep learning-based approach to identify cluster lenses from the DESI Legacy Imaging Surveys, utilizing the catalog of galaxy cluster candidates identified by Zou et al. (2021). Our lens-finder employs a ResNet-18 architecture, trained with mock images of cluster lenses as positives and observational images of cluster scale non-lenses as negatives. We do an iterative operation to increase the completeness of our work, namely adding the found true positive samples back to the training set and training again for several times. Human inspection is conducted to further refine the candidates, categorizing them into grades (A, B, C) according to the significance of the strongly lensed arcs.ResultsReviewing all 540,432 objects in Zou’s catalog, we discover 485 high-confidence cluster lens candidates with a cluster M500 range of 1013.67∼14.97M⊙ and a Brightest Central Galaxy (BCG) redshift range of 0.04∼0.89. After excluding the lens candidates listed in previous studies, we identify 247 newly discovered cluster lens candidates, including 16 grade A, 90 grade B, and 141 grade C.DiscussionThis catalog of cluster lens candidates is publicly available online, and follow-up observations are encouraged to confirm and conduct thorough investigations of these systems.

Peilin Zhao, Ziqiao Meng

Opeyemi Bamigbade, Mark Scanlon, John Sheppard

Embeddings remain the best way to represent image features, but do not always capture all latent information. This is still a problem in representation learning, and computer vision descriptors struggle with precision and accuracy. Improving image embedding with other features is necessary for tasks like image geolocation, especially for indoor scenes where descriptive cues can have less distinctive characteristics. This work proposes a model architecture that integrates image N-dominant colours and colour histogram vectors in different colour spaces with image embedding from deep metric learning and classification perspectives. The results indicate that the integration of colour features improves image embedding, surpassing the performance of using embedding alone. In addition, the classification approach yields higher accuracy compared to deep metric learning methods. Interestingly, different saturation points were observed for image colour-improved embedding features in models and colour spaces. These findings have implications for the design of more robust image geolocation systems, particularly in indoor environments.

Amir Bahmani, Kexin Cha, Arash Alavi et al.

Abstract Background Precision medicine promises significant health benefits but faces challenges such as complex data management and analytics, interdisciplinary collaboration, and education of researchers, healthcare professionals, and participants. Addressing these needs requires the integration of computational experts, engineers, designers, and healthcare professionals to develop user-friendly systems and shared terminologies. The widespread adoption of large language models (LLMs) such as Generative Pretrained Transformer (GPT) and Claude highlights the importance of making complex data accessible to non-specialists. Methods We evaluated the Stanford Data Ocean (SDO) precision medicine training program’s learning outcomes, AI Tutor performance, and learner satisfaction by assessing self-rated competency on key learning objectives through pre- and post-learning surveys, along with formative and summative assessment completion rates. We also analyzed AI Tutor accuracy and learners’ self-reported satisfaction, and post-program academic and career impacts. Additionally, we demonstrated the capabilities of the AI Data Visualization tool. Results SDO demonstrates the ability to improve learning outcomes for learners from broad educational and socioeconomic backgrounds with the support of the AI Tutor. The AI Data Visualization tool enables learners to interpret multi-omics and wearable data and replicate research findings. Conclusions SDO strives to mitigate challenges in precision medicine through a scalable, cloud-based platform that supports data management for various data types, advanced research, and personalized learning. SDO provides AI Tutors and AI-powered data visualization tools to enhance educational and research outcomes and make data analysis accessible to users from broad educational backgrounds. By extending engagement and cutting-edge research capabilities globally, SDO particularly benefits economically disadvantaged and historically marginalized communities, fostering interdisciplinary biomedical research and bridging the gap between education and practical application in the biomedical field.

Burcu Oltu, Selda Güney, Seniha Esen Yuksel et al.

Abstract Background Pulmonary diseases such as COVID-19 and pneumonia, are life-threatening conditions, that require prompt and accurate diagnosis for effective treatment. Chest X-ray (CXR) has become the most common alternative method for detecting pulmonary diseases such as COVID-19, pneumonia, and lung opacity due to their availability, cost-effectiveness, and ability to facilitate comparative analysis. However, the interpretation of CXRs is a challenging task. Methods This study presents an automated deep learning (DL) model that outperforms multiple state-of-the-art methods in diagnosing COVID-19, Lung Opacity, and Viral Pneumonia. Using a dataset of 21,165 CXRs, the proposed framework introduces a seamless combination of the Vision Transformer (ViT) for capturing long-range dependencies, DenseNet201 for powerful feature extraction, and global average pooling (GAP) for retaining critical spatial details. This combination results in a robust classification system, achieving remarkable accuracy. Results The proposed methodology delivers outstanding results across all categories: achieving 99.4% accuracy and an F1-score of 98.43% for COVID-19, 96.45% accuracy and an F1-score of 93.64% for Lung Opacity, 99.63% accuracy and an F1-score of 97.05% for Viral Pneumonia, and 95.97% accuracy with an F1-score of 95.87% for Normal subjects. Conclusion The proposed framework achieves a remarkable overall accuracy of 97.87%, surpassing several state-of-the-art methods with reproducible and objective outcomes. To ensure robustness and minimize variability in train-test splits, our study employs five-fold cross-validation, providing reliable and consistent performance evaluation. For transparency and to facilitate future comparisons, the specific training and testing splits have been made publicly accessible. Furthermore, Grad-CAM-based visualizations are integrated to enhance the interpretability of the model, offering valuable insights into its decision-making process. This innovative framework not only boosts classification accuracy but also sets a new benchmark in CXR-based disease diagnosis.

Yaoteng Zhang, Shuaipeng Wang, Yanlong Chen et al.

To maintain marine ecosystem health, effective algae monitoring is essential. Traditional threshold-based methods and standard machine learning techniques often fall short in accurately and automatically distinguishing algae types. This study presents Algae-Mamba, an advanced network for algae extraction that builds upon the visual state-space (VSS) model. The Algae-Mamba unified VSS model and the Kolmogorov–Arnold network proposed the Kolmogorov–Arnold visual state space (KVSS) model. KVSS block combines VSS for comprehensive global feature extraction with a small-kernel convolution module to capture local spatial and channel-specific information, supporting multiscale data processing and improving model generalization. The KVSS represents high-dimensional features using orthogonal polynomial combinations through Gram polynomials and leverages an attention mechanism to index interactions between target algae and their features, enabling the model to learn distinct characteristics of sargassum and ulva effectively and enhance extraction precision. To address the common misclassification between sargassum and ulva under limited spectral data, Algae-Mamba incorporates the normalized difference water index (NDWI) to enhance semantic richness. Furthermore, the model addresses class imbalances by employing a hybrid cross-entropy and Lovász-Softmax loss function, ensuring balanced and robust training. Unlike other methods that depend on extensive spectral information, Algae-Mamba achieves precise differentiation of sargassum and ulva with just 4-band spectral imagery, offering a powerful tool for monitoring marine ecological security. Testing on the GF-1 algae dataset demonstrates that Algae-Mamba surpasses other deep learning approaches in accurately extracting sargassum and ulva.

Turki Turki, Y-h. Taguchi

Understanding breast cancer drug response mechanisms can play a crucial role in improving treatment outcomes and survival rates. Existing bioinformatics-based approaches are far from perfect and do not adopt computational methods based on advanced artificial intelligence concepts. Therefore, we introduce a novel computational framework based on an efficient support vector machine (esvm) working as follows: First, we downloaded and processed three gene expression datasets related to breast cancer responding and non-responding to treatments from the gene expression omnibus (GEO) according to the following GEO accession numbers: GSE130787, GSE140494, and GSE196093. Our method esvm is formulated as a constrained optimization problem in its dual form as a function of <b>λ</b>. We recover the importance of each gene as a function of <b>λ</b>, <b>y</b>, and <b>x</b>. Then, we select <i>p</i> genes out of <i>n</i>, which are provided as input to enrichment analysis tools, Enrichr and Metascape. Compared to existing baseline methods, including deep learning, results demonstrate the superiority and efficiency of esvm, achieving high-performance results and having more expressed genes in well-established breast cancer cell lines, including MD-MB231, MCF7, and HS578T. Moreover, esvm is able to identify (1) various drugs, including clinically approved ones (e.g., tamoxifen and erlotinib); (2) seventy-four unique genes (including tumor suppression genes such as TP53 and BRCA1); and (3) thirty-six unique TFs (including SP1 and RELA). These results have been reported to be linked to breast cancer drug response mechanisms, progression, and metastasizing. Our method is available publicly on the maGENEgerZ web server.

Maria Frasca, Davide La Torre, Gabriella Pravettoni et al.

Abstract This review aims to explore the growing impact of machine learning and deep learning algorithms in the medical field, with a specific focus on the critical issues of explainability and interpretability associated with black-box algorithms. While machine learning algorithms are increasingly employed for medical analysis and diagnosis, their complexity underscores the importance of understanding how these algorithms explain and interpret data to take informed decisions. This review comprehensively analyzes challenges and solutions presented in the literature, offering an overview of the most recent techniques utilized in this field. It also provides precise definitions of interpretability and explainability, aiming to clarify the distinctions between these concepts and their implications for the decision-making process. Our analysis, based on 448 articles and addressing seven research questions, reveals an exponential growth in this field over the last decade. The psychological dimensions of public perception underscore the necessity for effective communication regarding the capabilities and limitations of artificial intelligence. Researchers are actively developing techniques to enhance interpretability, employing visualization methods and reducing model complexity. However, the persistent challenge lies in finding the delicate balance between achieving high performance and maintaining interpretability. Acknowledging the growing significance of artificial intelligence in aiding medical diagnosis and therapy, and the creation of interpretable artificial intelligence models is considered essential. In this dynamic context, an unwavering commitment to transparency, ethical considerations, and interdisciplinary collaboration is imperative to ensure the responsible use of artificial intelligence. This collective commitment is vital for establishing enduring trust between clinicians and patients, addressing emerging challenges, and facilitating the informed adoption of these advanced technologies in medicine.

Manasa Kesapragada, Yao-Hui Sun, Ksenia Zlobina et al.

Macrophages can exhibit pro-inflammatory or pro-reparatory functions, contingent upon their specific activation state. This dynamic behavior empowers macrophages to engage in immune reactions and contribute to tissue homeostasis. Understanding the intricate interplay between macrophage motility and activation status provides valuable insights into the complex mechanisms that govern their diverse functions. In a recent study, we developed a classification method based on morphology, which demonstrated that movement characteristics, including speed and displacement, can serve as distinguishing factors for macrophage subtypes. In this study, we develop a deep learning model to explore the potential of classifying macrophage subtypes based solely on raw trajectory patterns. The classification model relies on the time series of x-y coordinates, as well as the distance traveled and net displacement. We begin by investigating the migratory patterns of macrophages to gain a deeper understanding of their behavior. Although this analysis does not directly inform the deep learning model, it serves to highlight the intricate and distinct dynamics exhibited by different macrophage subtypes, which cannot be easily captured by a finite set of motility metrics. Our study uses cell trajectories to classify three macrophage subtypes: M0, M1, and M2. This advancement holds promising implications for the future, as it suggests the possibility of identifying macrophage subtypes without relying on shape analysis. Consequently, it could potentially eliminate the necessity for high-quality imaging techniques and provide more robust methods for analyzing inherently blurry images.

TAN Ruoqi, DONG Minggang, ZHAO Weixiao, WU Tianhao

Holding non-motorized vehicles accountable for legal violations effectively enhances urban traffic safety. Non-motorized vehicle license plates are characterized by small size, dense distribution, and ease of being obscured, which leads to significant feature information loss during the detection process in traditional deep learning-based methods. A non-motorized vehicle license plate recognition and localization method based on semantic alignment and hierarchical optimization is proposed. In this method, a semantic alignment module is designed for the underlying information fusion. During the upsampling process, low-level target information is used to guide the fusion of high-level semantics downwards, addressing the loss of small target features caused by conflicts between high- and low-level semantics. Subsequently, a hierarchical optimization module is constructed within the CSP structure to replace the deep ELAN module. This module uses a stack of a few convolutional kernel modules to extract the target information, reducing the number of network layers and preventing the loss of feature information at deeper levels. In the final stage, the K-Means++ algorithm is employed to cluster and obtain the initial anchor boxes suitable for non-motorized license plates to reduce the matching error during the training process. This approach aims to improve the accuracy of small-object recognition and localization. The experimental results demonstrate that the proposed method achieves a recognition and localization accuracy of 90.95% on a non-motorized vehicle license plate dataset. Compared with representative methods such as YOLOv7 and YOLOv8, it improves the accuracy by at least 3.58%. The proposed approach is effective for non-motorized vehicle license plate recognition and localization.

Syed Rizvi, Akash Awasthi, Maria J. Peláez et al.

Abstract The COVID-19 pandemic affected countries across the globe, demanding drastic public health policies to mitigate the spread of infection, which led to economic crises as a collateral damage. In this work, we investigate the impact of human mobility, described via international commercial flights, on COVID-19 infection dynamics on a global scale. We developed a graph neural network (GNN)-based framework called Dynamic Weighted GraphSAGE (DWSAGE), which operates over spatiotemporal graphs and is well-suited for dynamically changing flight information updated daily. This architecture is designed to be structurally sensitive, capable of learning the relationships between edge features and node features. To gain insights into the influence of air traffic on infection spread, we conducted local sensitivity analysis on our model through perturbation experiments. Our analyses identified Western Europe, the Middle East, and North America as leading regions in fueling the pandemic due to the high volume of air traffic originating or transiting through these areas. We used these observations to propose air traffic reduction strategies that can significantly impact controlling the pandemic with minimal disruption to human mobility. Our work provides a robust deep learning-based tool to study global pandemics and is of key relevance to policymakers for making informed decisions regarding air traffic restrictions during future outbreaks.

Avraam Tsantekidis, Nikolaos Passalis, Anastasios Tefas

Yiping Guo, Guyu Hu, Dongsheng Shao

Abstract According to the research status of Software Defined Network (SDN) control layer traffic scheduling, we find the current common problems, including single path, easy congestion, Quality of Service (QoS) requirements and high delay. To solve these four problems, we design and implement a QoS-oriented global multi-path traffic scheduling algorithm for SDN, referred to as QOGMP. First, we propose a link weight calculation algorithm based on the idea of traction links and deep reinforcement learning, and conduct experimental verifications related to traction links. The algorithm considers QoS requirements and alleviates the problems of easy congestion and high delay. Then, we propose a traffic scheduling algorithm based on link weight and multi-path scheme, which also considers QoS requirements and solves the problem of single path. Finally, we combined the link weight calculation algorithm and the traffic scheduling algorithm to implement QOGMP, and carried out comparative experiments in the built simulation environment. The experimental results show that QOGMP is better than the two comparison algorithms in terms of delay and rescheduling rate.

Simon Batzner, Albert Musaelian, Lixin Sun et al.

An E(3)-equivariant deep learning interatomic potential is introduced for accelerating molecular dynamics simulations. The method obtains state-of-the-art accuracy, can faithfully describe dynamics of complex systems with remarkable sample efficiency.

Chun-Hui Lin, Cheng-Jian Lin, Yu-Chi Li et al.

Cancer is the leading cause of death worldwide. Lung cancer, especially, caused the most death in 2018 according to the World Health Organization. Early diagnosis and treatment can considerably reduce mortality. To provide an efficient diagnosis, deep learning is overtaking conventional machine learning techniques and is increasingly being used in computer-aided design systems. However, a sparse medical data set and network parameter tuning process cause network training difficulty and cost longer experimental time. In the present study, the generative adversarial network was proposed to generate computed tomography images of lung tumors for alleviating the problem of sparse data. Furthermore, a parameter optimization method was proposed not only to improve the accuracy of lung tumor classification, but also reduce the experimental time. The experimental results revealed that the average accuracy can reach 99.86% after image augmentation and parameter optimization.

QIAO Tong, YAO Hongwei, PAN Binmin, XU Ming, CHEN Yanli

In the new era of rapid development of internet, where massive forgery images with updated tampering techniques flood into, traditional algorithms are no longer able to deal with the latest multimedia tampering techniques, especially those caused by Deepfake and deep learning techniques. Thus, a universal framework for image forensics including image pre-processing module, feature extraction module and post-processing module designed for specific classification were proposed creatively. Accordingly, the state-of-the-art algorithms were reviewed，and meanwhile the main strength and limitations of current algorithms were generalized. More importantly, the future studies were also listed for advancing the development of digital image forensics.