Traditional Chinese Medicine (TCM) diagnosis, particularly methods like tongue diagnosis, faces significant challenges in subjectivity and scalability. The application of Large Language Models (LLMs) to fundamental TCM tasks, such as syndrome differentiation and prescription generation, is significantly hampered by the difficulty of integrating visual tongue data with clinical text, and by the scarcity of suitable public datasets. To overcome these barriers, we introduce ViTCM-LLM, a novel framework that emulates an expert's diagnostic process by integrating multimodal language modeling with Retrieval-Augmented Generation (RAG). Employing a dual-component architecture, ViTCM-LLM integrates a fine-tuned Qwen2.5-VL model for visual analysis with a Qwen3-based RAG for clinical reasoning. The framework was developed and validated using MedTCM, a new large-scale multimodal dataset that we introduce specifically for advanced TCM research. To properly evaluate our framework's clinical accuracy, which existing metrics fail to capture, we also developed TDEU, a domain-specific evaluation metric. Evaluated using this metric alongside standard benchmarks, our comprehensive experiments on MedTCM demonstrate that ViTCM-LLM significantly outperforms leading models, including GPT-4o and Gemini 2.5 Flash. These findings not only establish the feasibility of a generalizable tongue diagnosis model but also validate the critical role of integrating visual data for advanced TCM diagnostics. The code and data can be found at https://github.com/jw-chae/ViTCM_LLM.

Cerebral small vessel disease (CSVD) is a major contributor to cognitive impairment and dementia in aging populations, necessitating early and precise diagnosis for effective clinical intervention. While existing multimodal imaging approaches address CSVD-related cognitive deficits, they often lack deep semantic integration of complementary modalities and struggle with class imbalance in heterogeneous data. We present the Enhancing Dual-Stream Attention Network (EDSAN), a novel framework that synergistically combines T1-weighted and diffusion tensor imaging (DTI) through multi-stage preprocessing for modality consistency. Our architecture introduces bidirectional cross-attention fusion and heterogeneous encoder optimization to achieve dynamically aligned multimodal feature learning. A unified optimization strategy employing gated weight learning and knowledge distillation streamlines cross-modal interactions while preserving complementary information under modality-specific constraints. The framework incorporates diffusion models and GAN-based auxiliary modules to enhance feature robustness and cross-modal distribution modeling. By integrating Focal Loss, cross-entropy, WGAN-GP, and diffusion reconstruction losses into a hybrid objective function, EDSAN demonstrates state-of-the-art performance in classifying CSVD-induced cognitive impairment. This modular paradigm establishes an extensible foundation for multimodal medical imaging analysis, adaptable to early diagnostic applications across neurological disorders.

With the great popularity of Sora, video-based audio generation has become indispensable. While numerous video-to-audio generation models have emerged, they frequently face difficulties including semantic incompatibilities and synchronization problems, especially in situations with multiple objects. To address these difficulties, we introduce TAGMO, a novel training-free audio generation method that offers precise time control for multi-object video scenarios. Our approach first employs object detection to obtain the class labels and temporal labels of each object, which are then structured and utilized as control conditions within a latent diffusion model (LDM) to generate multi-object audio. Additionally, we innovatively design a time mask based on the corresponding temporal labels and integrate it into the denoising process of the pre-trained audio generation model to achieve accurate temporal control. Experimental results demonstrate that our method enhances temporal alignment accuracy and semantic consistency. Audio demonstrations are available at https://coco-create.github.io/.

Breast cancer lesion segmentation in DCE-MRI remains challenging due to heterogeneous tumor morphology and indistinct boundaries. To address these challenges, this study proposes a novel hybrid segmentation network, HCMA-UNet, for lesion segmentation of breast cancer. Our network consists of a lightweight CNN backbone and a Multi-view Axial Self-Attention Mamba (MISM) module. The MISM module integrates Visual State Space Block (VSSB) and Axial Self-Attention (ASA) mechanism, effectively reducing parameters through Asymmetric Split Channel (ASC) strategy to achieve efficient tri-directional feature extraction. Our lightweight model achieves superior performance with 2.87M parameters and 126.44 GFLOPs. A Feature-guided Region-aware loss function (FRLoss) is proposed to enhance segmentation accuracy. Extensive experiments on one private and two public DCE-MRI breast cancer datasets demonstrate that our approach achieves state-of-the-art performance while maintaining computational efficiency. FRLoss also exhibits good cross-architecture generalization capabilities. The source code is available at https://github.com/Haoxuanli-Thu/HCMA-UNet.

Medical image analysis faces significant challenges due to limited annotation data, particularly in three-dimensional carotid artery segmentation tasks, where existing datasets exhibit spatially discontinuous slice annotations with only a small portion of expert-labeled slices in complete 3D volumetric data. To address this challenge, we propose a two-stage segmentation framework. First, we construct continuous vessel centerlines by interpolating between annotated slice centroids and propagate labels along these centerlines to generate interpolated annotations for unlabeled slices. The slices with expert annotations are used for fine-tuning SAM-Med2D, while the interpolated labels on unlabeled slices serve as prompts to guide segmentation during inference. In the second stage, we propose a novel Dense Bidirectional Feature Fusion UNet (DBF-UNet). This lightweight architecture achieves precise segmentation of complete 3D vascular structures. The network incorporates bidirectional feature fusion in the encoder and integrates multi-scale feature aggregation with dense connectivity for effective feature reuse. Experimental validation on public datasets demonstrates that our proposed method effectively addresses the sparse annotation challenge in carotid artery segmentation while achieving superior performance compared to existing approaches. The source code is available at https://github.com/Haoxuanli-Thu/DBF-UNet.

Document Visual Question Answering (Document VQA) faces significant challenges when processing long documents in low-resource environments due to context limitations and insufficient training data. This paper presents AdaDocVQA, a unified adaptive framework addressing these challenges through three core innovations: a hybrid text retrieval architecture for effective document segmentation, an intelligent data augmentation pipeline that automatically generates high-quality reasoning question-answer pairs with multi-level verification, and adaptive ensemble inference with dynamic configuration generation and early stopping mechanisms. Experiments on Japanese document VQA benchmarks demonstrate substantial improvements with 83.04% accuracy on Yes/No questions, 52.66% on factual questions, and 44.12% on numerical questions in JDocQA, and 59% accuracy on LAVA dataset. Ablation studies confirm meaningful contributions from each component, and our framework establishes new state-of-the-art results for Japanese document VQA while providing a scalable foundation for other low-resource languages and specialized domains. Our code available at: https://github.com/Haoxuanli-Thu/AdaDocVQA.

Abstract Dental caries, a common oral disease, poses serious risks if untreated, necessitating effective preventive measures like pit and fissure sealing. However, the reliance on experienced dentists for pit and fissures or caries detection limits accessibility, potentially leading to missed treatment opportunities, especially among children. To bridge this gap, we leverage deep learning in object detection to develop a method for autonomously identifying caries and determining pit and fissure sealing requirements using smartphone oral photos. We test several detection models and adopt a tiling strategy to reduce information loss during image pre‐processing. Our implementation achieves 72.3 mAP.5 with the YOLOXs model and tiling strategy. We enhance accessibility by deploying the pre‐trained network as a WeChat applet on mobile devices, enabling in‐home detection by parents or guardians. In addition, our data set of children's first permanent molars will also aid in the broader study of pediatric oral disease.

Brain tumor segmentation (BTS) has been studied from handcrafted engineered features to conventional machine learning (ML) methods, followed by the cutting-edge deep learning approaches. Each recent approach has attempted to overcome the challenges of previous methods and brought conveniences in efficacy, throughput, computation, explainability, investigation, and interpretability. Recently, deep learning (DL) algorithms show excellent performance regarding diverse fields, including image process, computer vision, health analytics, autonomous vehicles, and natural language processes; however, ultimately impediment in making the artificial intelligence explainable and interpretable to clinicians while dealing with critical health informatics and radiomics. Besides the sophisticated deep learning models for brain tumor segmentation, notorious notions like explainability, investigation, trust, and interpretability of DL raised significant concerns for clinicians in their domains. Among many DL methods, the neuro-symbolic learning (NSL) concept has gained more attention as it can contribute to explainable and interpretable AI. In the current study, we survey the prominent approaches, from handcrafted engineering conventional ML to deep learning algorithms, highlight the challenges in DL algorithms, and propose NSL architectures for BTS. Compared to existing surveys, our study not only outlines handcrafted to DL methods for BTS but also proposed explainable and interpretable pipelines appropriate for clinical practices. Our study can better facilitate novice learners in explainable AI and propose efficient, robust, interpretable DL models to facilitate the diagnosis, prognosis, and treatment of BTS.

Hypertension is a prevalent disease that is known to affect the vascular system especially to the people with poor living habits and lifestyles. Virtual reality (VR) is effective to interact with people to release their pressure and cheer them up, which however is less conducted towards manipulating blood pressure and hypertension. In this paper, we consider how hypertension can be treated with VR devices and design virtual reality river bathing therapy (VRBT) with respect to a combination of traditional methods through sensory stimulation, audio interventions, and motor training.

Frame rate up-conversion (FRUC) usually suffers from unreliable motion vectors due to the absence of the current frame to be interpolated. In addition, since the majority of video sequences are usually compressed by various coding standards to reduce the data volume, the quality of the generated frames in the FRUC will be further impaired. To address this problem, we proposed two FRUC algorithms based on deep residual network. We first present a deep residual network for the FRUC (DRNFRUC), which consists of feature extraction, feature recursive analysis, and image restoration parts with a skip connection between the input and the output of the network. The proposed DRNFRUC takes the result of an arbitrary existing FRUC method as the input and is able to significantly reduce the edge blurring and blocking artifacts when the motion of the block is violent. In addition, we proposed a deep residual network with weighted convolutional motion compensation (DRNWCMC) for the FRUC, where the convolution operations can be embedded into the motion compensation interpolation (MCI) in any existing MCI-based FRUC method. In DRNWCMC, we first devise two convolutional neural networks corresponding to the forward and backward motion compensated frames, respectively. And then, the adaptive interpolation coefficients for motion compensation are designed as two$1\times1$convolutional kernels. Finally, the interpolation result of WCMC is fed into another convolutional neural network to further improve the performance. All the parameters involved in the DRNWCMC are trained simultaneously under the same cost function. The experimental results show that the two proposed algorithms can remarkably improve both the objective and subjective quality of the interpolated frames.