Shengzhu Yang

Shengzhu Yang, Jiawei Du, Jia Guo et al.

Subtle semantic differences in retinal image and text data present great challenges for pre-training visual-language models. Moreover, false negative samples, i.e., image-text pairs having the same semantics but incorrectly regarded as negatives, disrupt the visual-language pre-training process and affect the model's learning ability. This work aims to develop a retinal foundation model, called ViLReF, by pre-training on a paired dataset comprising 451,956 retinal images and corresponding diagnostic text reports. In our vision-language pre-training strategy, we leverage expert knowledge to facilitate the extraction of labels and propose a novel constraint, the Weighted Similarity Coupling Loss, to adjust the speed of pushing sample pairs further apart dynamically within the feature space. Furthermore, we employ a batch expansion module with dynamic memory queues, maintained by momentum encoders, to supply extra samples and compensate for the vacancies caused by eliminating false negatives. Extensive experiments are conducted on multiple datasets for downstream classification and segmentation tasks. The experimental results demonstrate the powerful zero-shot and transfer learning capabilities of ViLReF, verifying the effectiveness of our pre-training strategy. Our ViLReF model is available at: https://github.com/T6Yang/ViLReF.

Jia Guo, Jiawei Du, Shengzhu Yang et al.

Current retinal foundation models remain constrained by curated research datasets that lack authentic clinical context, and require extensive task-specific optimization for each application, limiting their deployment efficiency in low-resource settings. Here, we show that these barriers can be overcome by building clinical native intelligence directly from real-world medical practice. Our key insight is that large-scale telemedicine programs, where expert centers provide remote consultations across distributed facilities, represent a natural reservoir for learning clinical image interpretation. We present ReVision, a retinal foundation model that learns from the natural alignment between 485,980 color fundus photographs and their corresponding diagnostic reports, accumulated through a decade-long telemedicine program spanning 162 medical institutions across China. Through extensive evaluation across 27 ophthalmic benchmarks, we demonstrate that ReVison enables deployment efficiency with minimal local resources. Without any task-specific training, ReVision achieves zero-shot disease detection with an average AUROC of 0.946 across 12 public benchmarks and 0.952 on 3 independent clinical cohorts. When minimal adaptation is feasible, ReVision matches extensively fine-tuned alternatives while requiring orders of magnitude fewer trainable parameters and labeled examples. The learned representations also transfer effectively to new clinical sites, imaging domains, imaging modalities, and systemic health prediction tasks. In a prospective reader study with 33 ophthalmologists, ReVision's zero-shot assistance improved diagnostic accuracy by 14.8% across all experience levels. These results demonstrate that clinical native intelligence can be directly extracted from clinical archives without any further annotation to build medical AI systems suited to various low-resource settings.

Jiawei Du, Jia Guo, Weihang Zhang et al.

The Vision-Language Foundation model is increasingly investigated in the fields of computer vision and natural language processing, yet its exploration in ophthalmology and broader medical applications remains limited. The challenge is the lack of labeled data for the training of foundation model. To handle this issue, a CLIP-style retinal image foundation model is developed in this paper. Our foundation model, RET-CLIP, is specifically trained on a dataset of 193,865 patients to extract general features of color fundus photographs (CFPs), employing a tripartite optimization strategy to focus on left eye, right eye, and patient level to reflect real-world clinical scenarios. Extensive experiments demonstrate that RET-CLIP outperforms existing benchmarks across eight diverse datasets spanning four critical diagnostic categories: diabetic retinopathy, glaucoma, multiple disease diagnosis, and multi-label classification of multiple diseases, which demonstrate the performance and generality of our foundation model. The sourse code and pre-trained model are available at https://github.com/sStonemason/RET-CLIP.

Shengzhu Yang, Jiawei Du, Shuai Lu et al.

Large-scale natural image-text datasets, especially those automatically collected from the web, often suffer from loose semantic alignment due to weak supervision, while medical datasets tend to have high cross-modal correlation but low content diversity. These properties pose a common challenge for contrastive language-image pretraining (CLIP): they hinder the model's ability to learn robust and generalizable representations. In this work, we propose CLIPin, a unified non-contrastive plug-in that can be seamlessly integrated into CLIP-style architectures to improve multimodal semantic alignment, providing stronger supervision and enhancing alignment robustness. Furthermore, two shared pre-projectors are designed for image and text modalities respectively to facilitate the integration of contrastive and non-contrastive learning in a parameter-compromise manner. Extensive experiments on diverse downstream tasks demonstrate the effectiveness and generality of CLIPin as a plug-and-play component compatible with various contrastive frameworks. Code is available at https://github.com/T6Yang/CLIPin.

Shuai Lu, Meng Wang, Jia Guo et al.

Large Vision Language Models (LVLMs) show immense potential for automated ophthalmic diagnosis. However, their clinical deployment is severely hindered by lacking domain-specific knowledge. In this work, we identify two structural deficiencies hindering reliable medical reasoning: 1) the Perception Gap, where general-purpose visual encoders fail to resolve fine-grained pathological cues (e.g., microaneurysms); and 2) the Reasoning Gap, where sparse visual evidence is progressively overridden by massive language priors in deeper transformer layers, leading to ungrounded hallucinations. To bridge these gaps, we propose EyExIn, a data-efficient framework designed to anchor retinal VLMs with expert knowledge via a Deep Expert Injection mechanism. Our architecture employs an Expert-Aware Dual-Stream encoding strategy that decouples visual representation into a general stream for anatomical context and a specialized expert stream for pathological semantics. To ensure high-fidelity integration, we design a Semantic-Adaptive Gated Fusion module, which dynamically amplifies subtle lesion signals while filtering irrelevant background noise. Furthermore, we introduce Adaptive Deep Expert Injection to embed persistent "Vision Anchors" by integrating fused visual features as residual biases directly into intermediate LLM layers. This mechanism creates a visual shortcut that forces the reasoning stack to remain strictly grounded in visual evidence. Extensive experiments across four benchmarks demonstrate that our model consistently outperforms massive proprietary systems. EyExIn significantly enhances domain-specific knowledge embedding and achieves state-of-the-art precision in ophthalmic visual question answering, advancing the development of trustworthy ophthalmic AI.