Jianhua Yin

Optical imaging can acquire various types of information from different skin levels, supporting non‐invasive diagnosis and dynamic monitoring. This article analyzes typical functional and biochemical optical imaging modalities providing clear interpretable information and reviews their applications in skin disease detection. It compares their information characteristics, unique diagnostic capabilities, clinical status, current limitations, and translational requirements. However, no single modality can fully characterize the complex pathological changes and multidimensional features of skin lesions because imaging depth, spatial resolution, and information specificity cannot be optimized simultaneously. Multimodal imaging is therefore essential for integrating complementary structural, functional, and molecular information and providing a more comprehensive representation of lesion heterogeneity. This review further examines major multimodal fusion strategies and discusses challenges including cross‐modal registration, information redundancy, and unified representation. Developing adaptive multimodal imaging and fusion frameworks that effectively capture and integrate multidimensional diagnostic information represents an important direction toward more accurate skin disease assessment.

Osteoarthritis (OA) is a common degenerative joint disease with limited early diagnostic options, highlighting the need for molecular‐level detection methods. This study systematically compared the performance of Raman scattering (RS), fiber‐enhanced Raman scattering (FERS), and surface‐enhanced Raman scattering (SERS) for synovial fluid (SF) analysis and OA staging, and further investigated the effect of SF phase transition on spectral responses. Results showed that, compared with RS, liquid‐phase FERS reduced drying‐induced molecular redistribution and conformational changes, thereby improving reproducibility and quantitative reliability. All three techniques consistently captured OA‐related molecular alterations, including progressive extracellular matrix degradation and accumulation of collagen degradation products in SF. Machine learning analysis demonstrated that SERS achieved the best staging performance, with the highest classification accuracy (77%) and AUC across all models. Importantly, fusion of FERS and SERS further improved accuracy to 83%, highlighting the advantage of complementary information integration. These findings emphasize the importance of detection state in spectral analysis and support Raman techniques as promising tools for minimally invasive OA diagnosis and staging.

Intraoperative pathological assessment requires rapid, high-quality stained sections to support surgical decisions. Currently, however, this area lacks cost-effective and rapid staining methods for intraoperative procedures. This study integrated a low-cost phase contrast microscope (PCM) imaging and Pix2Pix Generative Adversarial Network (GAN) to take the microimages and create a virtual staining framework of intraoperative sample-frozen sections for converting label-free phase contrast images (PCIs) into H&E-stained equivalents for the first time. PCIs and H&E-stained whole-slide images were captured from sample-frozen sections of 15 registered patients with breast cancer and compared with their autofluorescence image results. Pix2Pix and CycleGAN models were trained on these images. It’s found that the Pix2Pix GAN excels in PCI and generates a clear nuclear boundary and accurate tissue structure without staining artifacts. Quantitatively, Pix2Pix model provides a higher structural similarity index (SSIM [ ] 0.57) and peak signal-to-noise ratio (PSNR [ ] 21.65) in phase contrast mode than CycleGAN in the comparison study. The virtual staining results of PCM integrated with Pix2Pix are clearly superior to those of autofluorescence with Pix2Pix, as well as with CycleGAN. Additionally, the approach offers fast processing (1.5[ ]s/mm[ ] and can generate a standard WSI virtual stained section in 3 min. This low-cost virtual staining approach bypasses the cumbersome H&E staining process in conventional pathology, significantly accelerating diagnostic procedures. It’s very possible and of significance for this intelligent intraoperative technology to be extended to other intraoperative pathological assessment scenarios requiring rapid histological assessment.

Whole‐slide imaging assisted by deep learning has been employed to help the digital pathology, while limited by the scarcity of paired label data. To address this issue, a novel self‐supervised image modeling framework, PathMAE, is proposed to effectively enlarge the labeled dataset in a cross‐domain way, where cross‐dataset and even cross‐disease histopathological images can be used for model training. PathMAE integrates masked image modeling and contrastive learning to effectively learn transferable visual representations from unlabeled WSIs. The framework comprises two key components: a Swin‐Transformer‐based encoder‐decoder (SMED) with a window‐masking strategy for local feature reconstruction, and a Dynamic Memory Contrastive Learning (DMCL) module for enhancing global semantic alignment via memory‐guided feature comparison. Experimental results on three public histopathology datasets demonstrate the robustness and generalizability of the proposed method. In cross‐disease transfer (BreakHis → Osteosarcoma), PathMAE achieved 97.15% accuracy and 99.03% AUC; in cross‐dataset transfer (BreakHis → Camelyon16), it obtained 84.67% accuracy and 88.04% AUC. These findings validate the capability of PathMAE as a scalable and domain‐adaptive image analysis framework, offering new potential for building reliable computational pathology systems under limited supervision.