Multimodal medical image fusion is crucial for clinical diagnosis and treatment, as it integrates complementary information from multiple imaging modalities into a single image. However, existing fusion methods still struggle to preserve both fine lesion details and global anatomical structures across modalities. They often over-smooth small structures, blur organ boundaries, and introduce inconsistencies between structural and functional information. To address this problem, we introduce a hybrid kernel attention mechanism that couples small-kernel convolutions with large-kernel attention in a unified block, rather than relying on a fixed receptive field. It leverages scale-aware relationships between lesion-level textures, organ boundaries, and global anatomy across modalities, embedding these relationships into the fused representation. Concretely, A U-shaped encoder-decoder backbone is adopted with Restormerbased shallow feature extraction and several scale-aware hybrid kernel attention blocks are stacked at multiple resolutions. The resulting modality-specific features are then fed into a multimodal fusion module, which adaptively couples them via cross-modal interaction and reconstruction branches to enforce structural and semantic consistency. We adopt a comprehensive evaluation protocol spanning eight widely used fusion metrics on MRI-CT, MRI-PET, and MRI-SPECT benchmarks. Our proposed method achieves consistent improvements over recent state-of-the-art methods, yielding up to 41.9 %\) relative gain in mutual information compared with CDDFusion on the MRI-PET dataset, demonstrating its superiority for multimodal medical image fusion. Moreover, qualitative visual comparisons show sharper anatomical boundaries and more distinguishable lesion regions, indicating that HKAFusion preserves clinically relevant details more faithfully.

Electroencephalogram (EEG) signal classification faces significant challenges due to data distribution shifts caused by heterogeneous electrode configurations, acquisition protocols, and hardware discrepancies across domains. This paper introduces IMAC, a novel channel-dependent mask and imputation self-supervised framework that formulates the alignment of cross-domain EEG data shifts as a spatial time series imputation task. To address heterogeneous electrode configurations in cross-domain scenarios, IMAC first standardizes different electrode layouts using a 3D-to-2D positional unification mapping strategy, establishing unified spatial representations. Unlike previous mask-based self-supervised representation learning methods, IMAC introduces spatio-temporal signal alignment. This involves constructing a channel-dependent mask and reconstruction task framed as a low-to-high resolution EEG spatial imputation problem. Consequently, this approach simulates cross-domain variations such as channel omissions and temporal instabilities, thus enabling the model to leverage the proposed imputer for robust signal alignment during inference. Furthermore, IMAC incorporates a disentangled structure that separately models the temporal and spatial information of the EEG signals separately, reducing computational complexity while enhancing flexibility and adaptability. Comprehensive evaluations across 10 publicly available EEG datasets demonstrate IMAC's superior performance, achieving state-of-the-art classification accuracy in both cross-subject and cross-center validation scenarios. Notably, IMAC shows strong robustness under both simulated and real-world distribution shifts, surpassing baseline methods by up to 35% in integrity scores while maintaining consistent classification accuracy.

Abstract Online education has become more popular and effective due to the availability of high‐speed internet and technological innovations, which allow people from different locations to access educational resources and opportunities. However, online classes often face challenges such as limited interactivity and display options, which can affect the quality and effectiveness of the online learning experience. In this article, we propose GestureTeach, a new pedagogical paradigm that enables free handwriting interaction and animation generation for online teaching. GestureTeach uses gestures as a natural and intuitive way of interaction, which enhances the teacher's intention expression and the student's engagement. GestureTeach also generates animations from handwritten sketches, which improves the display effects of the interaction and the student's knowledge comprehension. We conducted a two‐stage study with 15 teachers and 90 students to evaluate the effectiveness of GestureTeach in facilitating classroom interaction. The results show that GestureTeach is preferred by both teachers and students over traditional online teaching methods and has the potential to transform the online teaching landscape by providing a seamless and interactive experience.