RESUMO
BACKGROUND: Ultrasound imaging is the most frequently performed for the patients with chronic hepatitis or liver cirrhosis. However, ultrasound imaging is highly operator dependent and interpretation of ultrasound images is subjective, thus well-trained radiologist is required for evaluation. Automated classification of liver fibrosis could alleviate the shortage of skilled radiologist especially in low-to-middle income countries. The purposed of this study is to evaluate deep convolutional neural networks (DCNNs) for classifying the degree of liver fibrosis according to the METAVIR score using US images. METHODS: We used ultrasound (US) images from two tertiary university hospitals. A total of 7920 US images from 933 patients were used for training/validation of DCNNs. All patient were underwent liver biopsy or hepatectomy, and liver fibrosis was categorized based on pathology results using the METAVIR score. Five well-established DCNNs (VGGNet, ResNet, DenseNet, EfficientNet and ViT) was implemented to predict the METAVIR score. The performance of DCNNs for five-level (F0/F1/F2/F3/F4) classification was evaluated through area under the receiver operating characteristic curve (AUC) with 95% confidential interval, accuracy, sensitivity, specificity, positive and negative likelihood ratio. RESULTS: Similar mean AUC values were achieved for five models; VGGNet (0.96), ResNet (0.96), DenseNet (0.95), EfficientNet (0.96), and ViT (0.95). The same mean accuracy (0.94) and specificity values (0.96) were yielded for all models. In terms of sensitivity, EffcientNet achieved highest mean value (0.85) while the other models produced slightly lower values range from 0.82 to 0.84. CONCLUSION: In this study, we demonstrated that DCNNs can classify the staging of liver fibrosis according to METAVIR score with high performance using conventional B-mode images. Among them, EfficientNET that have fewer parameters and computation cost produced highest performance. From the results, we believe that DCNNs based classification of liver fibrosis may allow fast and accurate diagnosis of liver fibrosis without needs of additional equipment for add-on test and may be powerful tool for supporting radiologists in clinical practice.
Assuntos
Técnicas de Imagem por Elasticidade , Humanos , Técnicas de Imagem por Elasticidade/métodos , Cirrose Hepática/patologia , Ultrassonografia , Curva ROC , Redes Neurais de Computação , Fígado/diagnóstico por imagemRESUMO
Graphs are data structures that effectively represent relational data in the real world. Graph representation learning is a significant task since it could facilitate various downstream tasks, such as node classification, link prediction, etc. Graph representation learning aims to map graph entities to low-dimensional vectors while preserving graph structure and entity relationships. Over the decades, many models have been proposed for graph representation learning. This paper aims to show a comprehensive picture of graph representation learning models, including traditional and state-of-the-art models on various graphs in different geometric spaces. First, we begin with five types of graph embedding models: graph kernels, matrix factorization models, shallow models, deep-learning models, and non-Euclidean models. In addition, we also discuss graph transformer models and Gaussian embedding models. Second, we present practical applications of graph embedding models, from constructing graphs for specific domains to applying models to solve tasks. Finally, we discuss challenges for existing models and future research directions in detail. As a result, this paper provides a structured overview of the diversity of graph embedding models.
RESUMO
This study aimed to automatically classify live cells based on their cell type by analyzing the patterns of backscattered signals of cells with minimal effect on normal cell physiology and activity. Our previous studies have demonstrated that label-free acoustic sensing using high-frequency ultrasound at a high pulse repetition frequency (PRF) can capture and analyze a single object from a heterogeneous sample. However, eliminating possible errors in the manual setting and time-consuming processes when postprocessing integrated backscattering (IB) coefficients of backscattered signals is crucial. In this study, an automated cell-type classification system that combines a label-free acoustic sensing technique with deep learning-empowered artificial intelligence models is proposed. We applied an one-dimensional (1D) convolutional autoencoder to denoise the signals and conducted data augmentation based on Gaussian noise injection to enhance the robustness of the proposed classification system to noise. Subsequently, denoised backscattered signals were classified into specific cell types using convolutional neural network (CNN) models for three types of signal data representations, including 1D CNN models for waveform and frequency spectrum analysis and two-dimensional (2D) CNN models for spectrogram analysis. We evaluated the proposed system by classifying two types of cells (e.g., RBC and PNT1A) and two types of polystyrene microspheres by analyzing their backscattered signal patterns. We attempted to discover cell physical properties reflected on backscattered signals by controlling experimental variables, such as diameter and structure material. We further evaluated the effectiveness of the neural network models and efficacy of data representations by comparing their accuracy with that of baseline methods. Therefore, the proposed system can be used to classify reliably and precisely several cell types with different intrinsic physical properties for personalized cancer medicine development.
Assuntos
Inteligência Artificial , Redes Neurais de Computação , Acústica , Frequência Cardíaca , UltrassonografiaRESUMO
Solar irradiance forecasting is fundamental and essential for commercializing solar energy generation by overcoming output variability. Accurate forecasting depends on historical solar irradiance data, correlations between various meteorological variables (e.g., wind speed, humidity, and cloudiness), and influences between the weather contexts of spatially adjacent regions. However, existing studies have been limited to spatiotemporal analysis of a few variables, which have clear correlations with solar irradiance (e.g., sunshine duration), and do not attempt to establish atmospheric contextual information from a variety of meteorological variables. Therefore, this study proposes a novel solar irradiance forecasting model that represents atmospheric parameters observed from multiple stations as an attributed dynamic network and analyzes temporal changes in the network by extending existing spatio-temporal graph convolutional network (ST-GCN) models. By comparing the proposed model with existing models, we also investigated the contributions of (i) the spatial adjacency of the stations, (ii) temporal changes in the meteorological variables, and (iii) the variety of variables to the forecasting performance. We evaluated the performance of the proposed and existing models by predicting the hourly solar irradiance at observation stations in the Korean Peninsula. The experimental results showed that the three features are synergistic and have correlations that are difficult to establish using single-aspect analysis.
RESUMO
Cell deformability is a useful feature for diagnosing various diseases (e.g., the invasiveness of cancer cells). Existing methods commonly inflict pressure on cells and observe changes in cell areas, diameters, or thickness according to the degree of pressure. Then, the Young's moduli (i.e., a measure of deformability) of cells are estimated based on the assumption that the degrees of the changes are inversely proportional to Young's moduli. However, manual measurements of the physical changes in cells are labor-intensive, and the subjectivity of the operators can intervene during this step, thereby causing considerable uncertainty. Further, because the shapes of cells are nonuniform, we cannot ensure the assumption for linear correlations of physical changes in cells with their deformability. Therefore, this study aims at measuring non-linear elastic moduli of live cells (degrees of cell deformability) automatically by employing conventional neural networks (CNN) and multilayer perceptrons (MLP) while preserving (or enhancing) the accuracy of the manual methods. First, we obtain photomicrographs of cells on multiple pressure levels using single-beam acoustic tweezers, and then, we suggest an image preprocessing method for emphasizing changes in cell areas on the photomicrographs. The CNN model is trained to measure the ratios of the cell area change at each pressure level. Then, we apply the multilayer perceptron (MLP) to learn the correlations of the cell area change ratios according to the pressure levels with cell deformability. The accuracy of the CNN was evaluated using two types of breast cancer cells: MDA-MB-231 (invasive) and MCF-7 (noninvasive). The MLP was assessed using five different beads (Young's moduli from 0.214 to 9.235 kPa), which provides standardized reference data of the non-linear elastic moduli of live cells. Finally, we validated the practicality of the proposed system by examining whether the non-linear elastic moduli estimated by the proposed system can distinguish invasive breast cancer cells from noninvasive ones.
Assuntos
Neoplasias da Mama , Aprendizado de Máquina , Acústica , Neoplasias da Mama/diagnóstico , Módulo de Elasticidade , Feminino , Humanos , Redes Neurais de ComputaçãoRESUMO
Single-beam acoustic tweezers (SBAT) is a widely used trapping technique to manipulate microscopic particles or cells. Recently, the characterization of a single cancer cell using high-frequency (>30 MHz) SBAT has been reported to determine its invasiveness and metastatic potential. Investigation of cell elasticity and invasiveness is based on the deformability of cells under SBAT's radiation forces, and in general, more physically deformed cells exhibit higher levels of invasiveness and therefore higher metastatic potential. However, previous imaging analysis to determine substantial differences in cell deformation, where the SBAT is turned ON or OFF, relies on the subjective observation that may vary and requires follow-up evaluations from experts. In this study, we propose an automatic and reliable cancer cell classification method based on SBAT and a convolutional neural network (CNN), which provides objective and accurate quantitative measurement results. We used a custom-designed 50 MHz SBAT transducer to obtain a series of images of deformed human breast cancer cells. CNN-based classification methods with data augmentation applied to collected images determined and validated the metastatic potential of cancer cells. As a result, with the selected optimizers, precision, and recall of the model were found to be greater than 0.95, which highly validates the classification performance of our integrated method. CNN-guided cancer cell deformation analysis using SBAT may be a promising alternative to current histological image analysis, and this pretrained model will significantly reduce the evaluation time for a larger population of cells.
RESUMO
Narrative works (e.g., novels and movies) consist of various utterances (e.g., scenes and episodes) with multi-layered structures. However, the existing studies aimed to embed only stories in a narrative work. By covering other granularity levels, we can easily compare narrative utterances that are coarser (e.g., movie series) or finer (e.g., scenes) than a narrative work. We apply the multi-layered structures on learning hierarchical representations of the narrative utterances. To represent coarser utterances, we consider adjacency and appearance of finer utterances in the coarser ones. For the movies, we suppose a four-layered structure (character roles ∈ characters ∈ scenes ∈ movies) and propose three learning methods bridging the layers: Char2Vec, Scene2Vec, and Hierarchical Story2Vec. Char2Vec represents a character by using dynamic changes in the character's roles. To find the character roles, we use substructures of character networks (i.e., dynamic social networks of characters). A scene describes an event. Interactions between characters in the scene are designed to describe the event. Scene2Vec learns representations of a scene from interactions between characters in the scene. A story is a series of events. Meanings of the story are affected by order of the events as well as their content. Hierarchical Story2Vec uses sequential order of scenes to represent stories. The proposed model has been evaluated by estimating the similarity between narrative utterances in real movies.
RESUMO
This study aims to validate whether the research performance of scholars correlates with how the scholars work together. Although the most straightforward approaches are centrality measurements or community detection, scholars mostly participate in multiple research groups and have different roles in each group. Thus, we concentrate on the subgraphs of co-authorship networks rooted in each scholar that cover (i) overlapping of the research groups on the scholar and (ii) roles of the scholar in the groups. This study calls the subgraphs "collaboration patterns" and applies subgraph embedding methods to discover and represent the collaboration patterns. Based on embedding the collaboration patterns, we have clustered scholars according to their collaboration styles. Then, we have examined whether scholars in each cluster have similar research performance, using the quantitative indicators. The coherence of the indicators cannot be solid proofs for validating the correlation between collaboration and performance. Nevertheless, the examination for clusters has exhibited that the collaboration patterns can reflect research styles of scholars. This information will enable us to predict the research performance more accurately since the research styles are more consistent and sustainable features of scholars than a few high-impact publications.