Yining Xie

Different imaging devices introduce substantial domain shifts. These shifts severely degrade the performance of traditional Chinese medicine (TCM) microscopic image classification models. Textual descriptions of microscopic identification characteristics provide domain-agnostic semantic anchors for cross-domain classification. However, existing cross-modal classification methods suffer from insufficient discriminability of text prototypes and a bias of visual representations toward seen domains. To address these issues, we propose the Target-Aware Prototype Rectification and Memory Learning Network (TAPM-Net) for cross-domain TCM microscopic image classification. TAPM-Net consists of three stages: text prototype construction, feature adaptation, and decision calibration. In the text prototype construction stage, low-bias text anchors are selected through multi-criteria joint scoring by the Evidence-Driven Cross-Domain Prototype Grounding (ECPG) module. In the feature adaptation stage, source-domain samples are aligned with text prototypes, while target-domain samples are aligned with dynamically updated visual prototypes through dual-signal constraints. A confidence-filtered memory mechanism is further introduced to stabilize prototype estimation. In the decision calibration stage, decision-level correction is performed on low-confidence and high-entropy samples by the Uncertainty-Aware Domain Bias Decoupling (UDBD) module without requiring domain labels. Compared with a non-adaptive baseline, TAPM-Net improves accuracy by up to 11.49 percentage points and weighted F1 by up to 14.31 percentage points on fully unseen target domains. Moreover, TAPM-Net outperforms all competing methods, including test-time adaptation approaches, on unseen domains, while maintaining an accuracy of 0.7622 on the seen target domain.

In recent years, researchers have found that the diagnosis of cardiovascular disease (CVD) is closely related to retinal fundus images and specific clinical indicators. However, current diagnostic methods are still limited to analyzing single-modal data, lacking analysis of multi-modal diagnostic information, and unable to comprehensively predict the progression of CVD. To this end, we design a non-invasive multimodal CVD detection method based on comprehensive view analysis. This method uses retinal fundus images and non-invasive clinical indicators as analysis data. The method is divided into two branches; one branch is used to extract retinal fundus features, and the other branch is used to extract clinical indicator features. In the retinal fundus feature extraction branch, we propose a method for comprehensive view analysis, which assists in learning retinal fundus images by incorporating vascular segmentation maps. Additionally, we introduce a multi-order belief interaction feature fusion method, through the basic belief assignment and Dempster-Shafer theory, the extracted features of the two modalities are fused in two stages, thereby achieving complementarity between multimodal information. The model in this paper was developed and verified using 1870 UK Biobank data. The results show that our method achieves an area under the curve (AUC) of 0.873 (95% CI, 0.855–0.892) and a PR score of 0.896 (95% CI, 0.882–0.91), which were higher than those of the method using single modality data and traditional fusion methods.

Aero-engine defects often exhibit micro-scale and high-frequency characteristics under complex metallic textures, which makes precise segmentation difficult. Most existing pixel-level methods rely on spatial-domain modeling and lack frequency-domain decoupling. As a result, high-frequency details are easily hidden by low-frequency background information. In addition, repeated downsampling weakens the representation of fine-grained structures, leading to inaccurate boundary localization and limited robustness. To address these issues, a spectrum-driven hierarchical learning network is proposed for aero-engine defect segmentation. First, a dual-band spectral module is constructed using the discrete cosine transform to separate high-frequency and low-frequency components, providing stable and physically meaningful frequency-domain priors for the network. Second, a detail-guided module is designed where high-frequency features adaptively guide skip connections, compensating information loss during encoding and improving boundary recovery. Furthermore, a low-frequency-driven region-aware modeling module is developed. The internal defect regions, boundary areas, and background regions are modeled hierarchically. A dynamic hyper-kernel generation mechanism performs region-sensitive convolutional modeling, improving adaptation to complex structural variations. Extensive experiments on the Turbo19 and NEU-Seg datasets demonstrate that the proposed method produces accurate defect boundaries and achieves mIoU scores of 89.82% and 91.44%, improving over the second-best method by 5.22% and 4.42%, respectively.

Immunohistochemistry (IHC) staining is crucial for determining tumor subtypes, obtaining protein expression information, and developing personalized treatment plans. But compared with hematoxylin and eosin (H&E) staining, IHC staining is more complex and expensive. With the advancement of deep learning, converting H&E stained images into IHC stained images has gradually emerged as a solution for obtaining IHC staining. However, current virtual staining processes suffer from difficulties in aligning pathological semantic features, posing significant challenges for network training, which poses significant challenges for network training. To solve these issues, we propose a multi-view multi-level pathology semantic perception learning method for H&E-to-IHC virtual staining (M2PL-GAN). Unlike prior approaches, M2PL-GAN introduces a comprehensive semantic learning paradigm from three views: structural contextual relations, feature distribution, and topology-aware fine-grained semantics. These correspond to the Context-aware Correlation Mechanism (CACM), the Local-aware Distribution Alignment Mechanism (LDAM), and the Graph-aware Bidirectional Contrastive Learning Mechanism (GBCLM) respectively. Among them, CACM enhances contextual consistency by establishing semantic correlations between virtual and real IHC images at local scales. LDAM ensures alignment of semantic feature distributions between virtual and real IHC images, mitigating semantic shifts caused by HE-IHC staining. GBCLM leverages graph neural network to capture topology-aware semantic representations and optimizes semantic feature alignment through bidirectional contrastive learning. Extensive experiments on both public and private datasets demonstrate that our method outperforms state-of-the-art approaches in both quantitative metrics and qualitative evaluations. Our code is available in https://github.com/Pikachu-one/M2PL-GAN.