Accurate multi-organ segmentation across heterogeneous medical images is pivotal for real-world surgical navigation. The scarcity of annotation constitutes a well-established consensus in the field, prompting semi-supervised learning to emerge as a prominent solution. However, two critical bottlenecks persist in clinical translation: (1) inter-class feature ambiguity, and (2) high multi-source sample heterogeneity. To tackle these bottlenecks, we propose TP-Net, a semi-supervised framework for multi-organ segmentation in heterogeneous medical images, which innovatively integrates reliable multi-prototype contrastive learning. Specifically, we propose a heterogeneous prototype dynamic evolution mechanism that self-adaptively models fragmented intra-class distributions in multi-source data. Then, to mitigate prototype shift, we first devise an uncertainty-aware cross-domain alignment strategy grounded in the smoothness assumption, which constructs reliable prototypes by propagating reliable pixel prediction distributions from labeled to unlabeled domains. Furthermore, this is synergized with contrastive separation that enforces feature proximity to class-matched prototypes in the embedding space, effectively resolving inter-class ambiguity by minimizing overlap between adjacent organ clusters via prototype repulsion. Experimental results on two public datasets and one in-house dataset prove that the proposed method achieves state-of-the-art performance. Further, clinical validation with our self-developed surgical navigation system demonstrated the clinical viability of the proposed method. The code associated with this work will be made publicly available at https://github.com/JIESHUREN330/TP-Net/tree/main.

Automatic segmentation of brain tumors in magnetic resonance imaging (MRI) is essential for clinical decision support, yet it remains a challenging task due to the pronounced heterogeneity of gliomas and the often indistinct boundaries between their subregions. Moreover, the opaque, “black-box” nature of conventional deep learning models limits their adoption in clinical workflows, where interpretability is critical. To address these challenges, we propose EProtoSeg, a novel 3D segmentation network that synergistically integrates explainable prototypebased feature learning with adaptive multi-scale context aggregation. Specifically, EProtoSeg incorporates an Explainable Prototype Fusion (EPF) module in the decoder, which learns class-specific prototypes to guide voxel-wise classification and improve boundary precision. In parallel, an Adaptive Multi-Scale Context (AMSC) module is embedded in the skip connections to dynamically fuse fine spatial details and high-level semantic information across scales. A deep supervision strategy is further employed to enhance discriminability in ambiguous regions and ensure stable optimization. Extensive experiments on the BraTS 2020 and 2021 benchmarks demonstrate that EProtoSeg achieves state-of-the-art segmentation performance while offering interpretable feature representations, thereby improving both accuracy and clinical trust. The code is available at https://github.com/chenbn266/EProtoSeg

Few-shot learning offers a promising solution for computational pathology by alleviating the reliance on large an-notated datasets, but faces challenges from the high redundancy in whole slide images and underutilized cross-modal knowledge. Existing methods typically use foundation models only for pre-liminary feature extraction while employing fixed or single-level prompts that lack multi-scale pathological representation. To address these limitations, we propose a Hierarchical Vision-Text Prompt (H- VTP) framework that enables multi-level cross-modal interaction through GPT-4 generated Local Instance Prompts for patch-level morphological details and Global Semantic Prompts for slide-level diagnostic context. A dual-branch decoding mech-anism with Text-Guided-Patch Decoder and Patch-Augmented-Text Decoder facilitates closed-loop vision-text fusion, while a parameter-efficient adaptation strategy trains only lightweight prompts and adapters. Extensive experiments on three cancer subtype datasets demonstrate the superiority of H-VTP in few-shot WSI classification, confirming its effectiveness for clinical applications.

Accurate MRI brain tumor segmentation is significant for disease diagnosis and treatment. U-Nets are widely explored in automatic segmentation due to high accuracy and efficiency. Additionally, global semantic information establishes effectiveness in improving MRI brain tumor segmentation accuracy, and corresponding models have gained increasing attention. Inspired by state-space models excelling in long-range dependency modeling, we propose the Volumetric Axial Shift Mamba network (VASMambaU-Net) - a novel brain tumor segmentation method more suitable for 3D images. It integrates an improved Mamba module with large kernel convolution in U-Net to capture global tumor features. In VASMambaU-Net, we propose VASMamba that using a volumetric axial shift mechanism as the bottleneck to capture long-range dependencies of 3D brain tumor images. In addition, along with the conventional convolutional encoder, a large kernel convolutional encoder is supplemented to enhance the image receptive field, further improving the capability to capture global information, thereby enhancing the brain tumor segmentation performance. VASMambaU-Net achieves mean DSC values of$84.37 \%, 84.31 \%$and 91.07 % in the BraTS2019-2021 datasets, respectively. The corresponding mean HD95 values obtained in these three datasets are 4.67 mm, 13.51 mm, and 5.77 mm, respectively. These results demonstrate the competitiveness and effectiveness of VASMambaU-Net compared to state-of-the-art methods.

Automatic colorization of anime line art videos aims to produce color frames given line art frames and reference color images, which is challenging due to various motions and geometric transformations across frame sequences. Existing methods usually utilize the feature maps of reference images directly and treat all the regions in an image equally. However, this may overlook the details of the regions undergoing geometric transformations . To emphasize the regions with significant transformations between the reference and target frames, we propose a Transformation Region Enhancement Network (TRE-Net) to exploit useful reference information and enhance the colorization of key transformation regions with Region Localization Module (RLM) and Feature Enhancement Module (FEM). Specifically, we propose Multi-scale Euclidean Distance Difference (Multi-scale EDD) Maps in RLM which effectively locate geometric transformation regions by contrasting the Euclidean Distance Maps of two line arts and aggregating representations at multiple scales of the network. In addition, FEM is devised to enhance feature learning in the regions with geometric transformation and to ensure proper color alignment. FEM learns locally enhanced features through an attention-gating operation at a low computational cost. With the well-represented key geometric transformation regions, our method exploits the multi-scale reference information well for color alignment, thus produces perceptually pleasing frames. Comprehensive experimental results show that our proposed method is superior to existing methods in terms of the overall quality of colorized anime line art videos.

Automatic sketch colorization is a challenging task that aims to generate a color image from a sketch, primarily due to its inherently ill-posed nature. While many approaches have shown promising results, two significant challenges remain: limited color patterns and a wide range of artifacts such as color bleeding and semantic inconsistencies among relevant regions. These issues stem from the operation of traditional convolutional structures, which capture structural features in a pixel-wise manner, resulting in inadequate utilization of regional information within the sketch. Therefore, we propose the Region-Assisted Sketch Coloring (RASC) method, which introduces an intermediate representation called the 'Region Map' to explicitly characterize the regional information of the sketch. This Region Map is derived from the input sketch and is effectively formulated by our RASC architecture, enhancing the perception of region-wise features beyond the original pixel-wise features. Specifically, we start by employing the sketch encoder to extract hierarchical feature maps from the input sketches. Subsequently, we introduce a coarse-to-fine decoder comprising a series of Region-based Modulation (RM) blocks. This decoder modulates features that combine the modulation results of its previous block and the sketch features of the corresponding encoder block with our Region Formulation module. Each module explicitly formulates the sketch features in a region-wise manner. This accurately captures both the inner-region local style and inter-region global context dependency, resulting in various color patterns and fewer synthesis artifacts. Our experimental results show that our proposed method surpasses state-of-the-art methods in both synthetic and real sketch datasets.

Researches on breast cancer histopathological image classification have achieved a great breakthrough using deep backbones of Convolutional Neural Networks (CNNs) in recent years. However, due to the inductive bias of locality, CNNs are unable to effectively extract the global feature information of breast cancer histopathological images, limiting the improvement of the classification results. To overcome this shortcoming, this paper reasonably introduces an extra backbone stream of a pure transformer, which consists of a self-attention mechanism to capture global receptive fields of histopathological images, thereby compensating the locality characteristic of CNNs backbone. Based on two backbone streams of CNN and transformer, a dual-stream network called DCET-Net is proposed, which considers local features and global ones simultaneously, and progressively combines them from these two streams to form the final representations for classification. DCET-Net is extensively evaluated on the representative BreakHis histopathological image dataset, and experimental results demonstrate that it is highly competitive with the state-of-the-art CNN methods in breast cancer histopathological image classification task.

Automatic sketch colorization is a challenging task in both computer graphics and computer vision since all the color, texture, shading generation have to be created based on the abstract sketch. Besides, it is a subjective task in painting process, which needs illustrators to comprehend drawing priori (DP), such as hue variation, saturation contrast and gray contrast and utilize them in the HSV color space which is closer to human visual cognition system. As such, incorporating supplementary supervision in the HSV color space may be beneficial to sketch colorization. However, previous methods improve the colorization quality only in the RGB color space without considering the HSV color space, often causing results with dull color, inappropriate saturation contrast, and artifacts. To address this issue, we propose a novel sketch colorization method, dual color space guided generative adversarial network (DCSGAN), that considers the complementary information contained in both the RGB and HSV color space. Specifically, we incorporate the HSV color space to construct dual color spaces for supervising our method with a color space transformation (CST) network that learns transformation from the RGB to HSV color space. Then, we propose a DP loss that enables the DCSGAN to generate vivid color images with pixel level supervision. Additionally, a novel dual color space adversarial (DCSA) loss is designed to guide the generator at global level to reduce the artifacts to meet audiences' aesthetic expectations. Extensive experiments and ablation studies demonstrate the superiority of the proposed method over previous state-of-the-art (SOTA) methods.

Recently, modeling deep convolutional activations by the global second-order pooling has shown great advance on visual recognition tasks. However, most of the existing deep second-order statistical models mainly compute second-order statistics of activations of the last convolutional layer as image representations, and they seldom introduce second-order statistics into earlier layers to better fit network topology, thus limiting the representational ability to a certain extent. Motivated by the flexibility of attention blocks that are commonly plugged into intermediate layers of deep convolutional networks (ConvNets), this work makes an attempt to combine deep second-order statistics with attention mechanisms in ConvNets, and further proposes a novel Second-order Attention Guided Network (SoAG-Net) for visual recognition. More specifically, SoAG-Net involves several SoAG modules seemingly inserted into intermediate layers of the network, in which SoAG collects second-order statistics of convolutional activations by polynomial kernel approximation to predict channel-wise attention maps utilized for guiding the learning of convolutional activations through tensor scaling along channel dimension. SoAG improves the nonlinearity of ConvNets and enables ConvNets to fit more complicated distribution of convolutional activations. Experiment results on three commonly used datasets illuminate that SoAG-Net outperforms its counterparts and achieves competitive performance with state-of-the-art models under the same backbone.

Deep hashing has recently been attracting more and more attentions for large-scale image retrieval task owing to its superior performance of search efficiency and less storage space requirements. Among deep hashing models, asymmetric deep hashing performs feature learning on query dataset and directly generates hash code on database images, significantly improving the retrieval performance of deep hashing models. Meanwhile, recently works also establish that high-order statistic of deep features are helpful to obtain more discriminant representations of images. Therefore, to boost the retrieval capability of deep hashing, this work tries to integrate merits of the high-order statistic module and the asymmetric deep hashing architecture, and it further proposes a novel deep high-order asymmetric supervised hashing (DHoASH) for image retrieval. More specifically, we utilize a powerful global covariance pooling module based on matrix power normalization to compute the second-order statistic features of input images, which is fluently embedded into an asymmetric hashing architecture in an end-to-end manner, leading to the generation of more discriminant binary hashing code. Experiment results on two benchmarks illuminates the effectiveness of the proposed DHoASH, which also achieves very competitive retrieval accuracy compared to the state-of-the-art methods.

Recently, combination of advanced convolutional neural networks and efficient hashing, deep hashing have achieved impressive performance for image retrieval. However, state-of-the-art deep hashing methods mainly focus on constructing hash function, loss function and training strategies to preserve semantic similarity. For the fundamental image characteristics, they depend heavily on the first-order convolutional feature statistics, failing to take their global structure into consideration. To address this problem, we present a deep covariance estimation hashing (DCEH) method with robust covariance form to improve hash code quality. The core of DCEH involves covariance pooling as deep hashing representation performing global pairwise feature interactions. Due to convolutional features are usually high dimension and small sample size, we estimate robust covariance with matrix power normalization and then insert it into deep hashing paradigm in an end-to-end learning manner. Extensive experiments on three benchmarks show that the proposed DCEH outperforms its counterparts and achieves superior performance.