Objective. Stroke is a highly lethal condition, with intracranial vessel occlusion being one of its primary causes. Intracranial vessel occlusion can typically be categorized into four types, each requiring different intervention measures. Therefore, the automatic and accurate classification of intracranial vessel occlusions holds significant clinical importance for assessing vessel occlusion conditions. However, due to the visual similarities in shape and size among different vessels and variations in the degree of vessel occlusion, the automated classification of intracranial vessel occlusions remains a challenging task. Our study proposes an automatic classification model for large vessel occlusion (LVO) based on the difference information between the left and right hemispheres. Approach. Our approach is as follows. We first introduce a dual-branch attention module to learn long-range dependencies through spatial and channel attention, guiding the model to focus on vessel-specific features. Subsequently, based on the symmetry of vessel distribution, we design a differential information classification module to dynamically learn and fuse the differential information of vessel features between the two hemispheres, enhancing the sensitivity of the classification model to occluded vessels. To optimize the feature differential information among similar vessels, we further propose a novel cooperative learning loss function to minimize changes within classes and similarities between classes. Main results. We evaluate our proposed model on an intracranial LVO data set. Compared to state-of-the-art deep learning models, our model performs optimally, achieving a classification sensitivity of 93.73%, precision of 83.33%, accuracy of 89.91% and Macro-F1 score of 87.13%. Significance. This method can adaptively focus on occluded vessel regions and effectively train in scenarios with high inter-class similarity and intra-class variability, thereby improving the performance of LVO classification.
18 December 2025 in Journal on Computing and Cultural Heritage
Dongba script is a unique primitive pictographic writing system. Due to its complex layout structure and highly variable glyph shapes, detecting Dongba characters poses a significant challenge. To address this, we have established a high-quality Dongba script dataset, Dongba1800, which comprehensively covers various layout structures and spatial features found in ancient Dongba manuscripts. We also propose a character detection model named DongbaBPN, capable of precisely locating and detecting Dongba characters at the character level in complex pages. Specifically, our method directly models the boundaries of Dongba characters and consists of a feature extraction backbone similar to an FPN, an initial boundary proposal module, and a GCN-based boundary iterative refinement module. The boundary proposal module generates initial Dongba character boundary proposals by extracting feature semantic information, and then, the boundary iterative refinement module progressively refines these proposals until the boundaries can accurately locate and cover individual Dongba characters. Additionally, we conducted comparative experiments with other state-of-the-art text detection models, and our method achieved the best performance in the Dongba character detection task.
26 November 2025 in Journal of King Saud University Computer and Information Sciences
Ancient scripts embody valuable historical and cultural heritage, emphasizing the importance of their digital preservation. However, open-set recognition of such scripts remains challenging due to limited samples for novel classes, high morphological similarity among characters, and the inadequacy of traditional closed-set methods to accommodate emerging categories. To address these challenges, particularly the large intra-class variance and small inter-class differences caused by visually similar or variant characters, we propose FFSSC, a novel framework for discovering new classes of ancient texts via feature fusion and semi-supervised clustering. FFSSC consists of three main components: (i) a multi-level feature fusion Vision Transformer (FF-ViT) enhanced with an Isolation Score mechanism for key token selection, (ii) a Self-Supervised Ranking Contrastive Loss (SRC) that improves feature discrimination by training the model to group similar characters together while separating dissimilar ones, and (iii) a Semi-Supervised Gaussian Mixture Model (Semi-GMM) that clusters features to discover new classes. Experiments on multiple open-set ancient script datasets show that FFSSC achieves superior performance in novel class detection and incremental category expansion, while maintaining stable accuracy on known classes. This work lays a foundation for intelligent recognition and digital preservation of ancient scripts, supporting practical applications of open-set recognition in cultural heritage studies.