Test-time adaptation (TTA) has emerged as a promising solution to improve the robustness of deep learning models under domain shifts, particularly in real-world scenarios where source data or true labels are unavailable. In this paper, we propose a novel TTA framework tailored for medical vision-language models (VLMs) that leverages a Mixture-of-Experts (MoE) mechanism. Building upon a frozen, pre-trained BiomedCLIP backbone, our method integrates parallel MoE adapters of different medical imaging modalities in each vision MLP block, enabling expert-specific adaptation without disrupting the core model representation. At inference time, only the MoE router is optimized through an entropy-regularized objective, which is further augmented by pseudo-label guidance and adaptive scaling strategies. Additionally, we propose an entropy-aware MoE scaling policy that dynamically adjusts expert influence based on prediction uncertainty, improving model adaptability. Extensive experiments on multiple medical imaging benchmarks demonstrate that our approach achieves substantial performance improvements over existing TTA baselines, while maintaining high efficiency and parameter sparsity. Our results highlight the potential of MoE-enhanced TTA to achieve robust and generalizable medical VLMs in unseen domains without access to source data. Code is available at https://openi.pcl.ac.cn/OpenMedIA/MoME-TTA.git

Medical report generation (MRG) is a challenging yet highly demanding task in the application of multi-modal artificial intelligence in medicine. Typically, training an MRG model requires tens of thousands of labelled radiology images and reports datasets, which could be impractical for most clinical research groups. In this study, we present C2RG, a novel 3D vision and language foundation model tailored for Coronary Computed Tomography Angiography (CTA) Report Generation. Inspired by BLIP-2’s architecture, our method integrates a self-supervised pre-trained 3D cardiac vision model (ViT-B) and a general-purpose bilingual foundation model (ChatGLM-6B), with a lightweight querying Transformer (Q-Former). We also introduce a parallel high-resolution feature extractor module and a coronary calcification evaluation loss to simultaneously encode fine-grained 3D features and constrain the accuracy of report generation. We compared our model with six state-of-the-art MRG methods on a clinical dataset with 118 subjects, comprising 453 paired 3D CTA images and radiology reports. Experimental results with extensive ablations show the efficacy of our C2RG. Codes will be open-sourced after the conference.

Urban greening trees play a significant role in optimizing the environment, and the segmentation of branches can effectively help researchers assess the growth status of trees. This paper proposed an innovative RGB image-based model to segment tree branches within urban landscapes. The proposed model employed a dual-encoder architecture, including a basic encoder and an edge encoder. These components were designed with specialized blocks adept at extracting critical semantic features and edge information, enhancing the model's ability to segment fine branches. A graph reasoning decoder block with attention-based feature fusion was proposed to capture the semantic associations between regions incorporating edge information. Moreover, the elastic interaction-based loss function, a groundbreaking loss function, was introduced to ensure that the segmentation of the fine branches should be achieved smoothly and consistently. Upon evaluation against a public urban street tree dataset, the precision, recall, IoU, and accuracy of tree branch segmentation are 94.39%, 93.16%, 88.27%, and 98.78%, respectively, achieving the best result among all the tested models. This performance demonstrates the impact of deep learning on enhancing urban greening and sustainable development in effective tree management.

In order to improve the working efficiency and reliability of energy-harvesting low-power wide area networks (EH-LPWANs) applied in smart forest monitoring, a cross-layer collaboration control model is proposed. Taking the influencing factors of EH-LPWAN as concept nodes, a fuzzy cognitive map (FCM) was devised, and the relationship between each concept was utilized to establish the cross-layer model for optimally satisfying multiple objectives and conflicting constraints. Here, a dynamic FCM scheme based on adaptive glowworm swarm optimization (AGSO) is given to determine the concept weights and sustain online updates. The results show that the energy neutrality of EH nodes and the data throughput of the entire network completely meet the real-time and stability requirements for precise forest monitoring.