Introduction: Vision impairment caused by retinal diseases stands out as one of the significant sources of irreversible blindness, highlighting the importance of developing robust automated screening tools to detect eye diseases at an early stage. Despite the promising results of recent deep learning approaches to the automatic classification of retinal images, most existing solutions focus on image-based representations and do not consider the importance of vascular topology. To overcome the limitations, this study proposes VG-RETFound - a hybrid framework based on a retinal foundation model and graph-based vascular-topology modeling that can be used to classify multiple retinal categories and detect vision impairment. Methods: The authors use RETFound-a retinal foundation model-pretrained on a vast number of retinal images, as well as the Graph Attention Network (GAT), to learn vascular topology features by representing retinal vessels as graphs. The presented approach employs a cross-modal multi-head attention fusion method to integrate features from both modalities in real time. To evaluate the efficiency of VG-RETFound, experiments were conducted using the RFMiD and ODIR-5 K datasets for model training and testing across six retinal categories: Normal, Diabetic Retinopathy, Pathological Myopia, Age-related Macular Degeneration, Hypertensive Retinopathy, and Other Retinal Diseases. Results: The accuracy of 96.39%, F1-score of 95.84%, MCC of 94.87%, and Kappa of 94.71% were attained on an internal test dataset, while cross-dataset external validation on the BRSET dataset revealed accuracy of 94.81%, F1-score of 94.25%, MCC of 93.09%, and Kappa of 92.94%. Discussion: These findings demonstrate that integrating retinal foundation models with anatomically informed vessel-topology learning significantly improves classification accuracy, interpretability, and generalization, highlighting the potential of VG-RETFound as a tool for large-scale vision impairment. screening and retinal disease diagnosis.
Expert Systems with Applications · مجلد 255 · 124650-124650
Most existing methods for detecting fraudulent news or other forms of disinformation primarily rely on user profiling or content analysis, determining whether a given article aligns with a user’s stylistic or content-related preferences. This research departs from conventional approaches by concentrating not only on the characteristics of shared content but also on user interactions and the resultant topologies of content propagation trees. In our study, we have introduced a deep learning model based on Graph Convolutional Neural Networks (GCNN) and enhanced with multi-head attention mechanisms. This model leverages a wide range of psycholinguistic attributes extracted from users’ posts, including sentiment, emotional content, linguistic features, personality traits, readability, and communication style. We further enrich these attributes with BERT embeddings to improve textual representation. Our research framework involves creating two distinct graph networks: the user interaction graph and the semantic propagation graph. These graphical representations are essential for visualizing dynamic interactions and patterns of information dissemination among users. For each user, we generate a set of carefully crafted features that capture their unique characteristics and behaviors. These user-specific features are then integrated into the User Interaction Graph, enabling a more nuanced understanding and representation of user interactions within our proposed model. The proposed framework has demonstrated superior performance on benchmark datasets, achieving accuracies and precisions of 91.07% and 91.24% on the FakeNewsNet dataset, and 94.20% and 94.92% on the FibVid dataset, respectively. The results obtained from the experimental evaluation have also revealed the effect of various parameters significant to profile disinformers, signifying a substantial improvement in the field of fake news detection and profiling.
Computer Systems Science and Engineering · مجلد 44 (2) · 1219-1231
A cyber physical energy system (CPES) involves a combination of processing, network, and physical processes. The smart grid plays a vital role in the CPES model where information technology (IT) can be related to the physical system. At the same time, the machine learning (ML) models find useful for the smart grids integrated into the CPES for effective decision making. Also, the smart grids using ML and deep learning (DL) models are anticipated to lessen the requirement of placing many power plants for electricity utilization. In this aspect, this study designs optimal multi-head attention based bidirectional long short term memory (OMHA-MBLSTM) technique for smart grid stability prediction in CPES. The proposed OMHA-MBLSTM technique involves three subprocesses such as pre-processing, prediction, and hyperparameter optimization. The OMHA-MBLSTM technique employs min-max normalization as a pre-processing step. Besides, the MBLSTM model is applied for the prediction of stability level of the smart grids in CPES. At the same time, the moth swarm algorithm (MHA) is utilized for optimally modifying the hyperparameters involved in the MBLSTM model. To ensure the enhanced outcomes of the OMHA-MBLSTM technique, a series of simulations were carried out and the results are inspected under several aspects. The experimental results pointed out the better outcomes of the OMHA-MBLSTM technique over the recent models.
cyber-physical energy systemsmulti-head attentionbidirectional LSTMhyperparameter optimization