Chief Supervisor

Dr SHI Danli

Project Title

Synopsis

This project aims to develop and clinically evaluate generative AI models for multimodal ophthalmic imaging, with a particular focus on cross-modal image generation and its translation into real-world clinical practice. Building on existing cross-modal generation approaches, the project will improve model performance by incorporating multimodal representation learning, diffusion-based generative modelling, ophthalmic foundation models, and structure- and pathology-aware constraints. The goal is not only to generate visually realistic images, but also to preserve clinically relevant anatomical structures, vascular patterns, lesions, and disease-specific features across imaging modalities such as fundus photography, OCT, OCTA, and fluorescein angiography.
The generated images will first undergo systematic technical and clinical validation using image-quality metrics, biomarker consistency, lesion-level assessment, and downstream clinical tasks such as disease classification, segmentation, and diagnostic decision support. Uncertainty estimation and quality-control mechanisms will also be developed to identify unreliable synthetic outputs.
Following retrospective validation, the project will progress towards prospective clinical evaluation to investigate whether generative AI-assisted multimodal imaging can improve diagnostic efficiency, reduce the need for additional imaging examinations, and support clinical decision-making. The study will assess diagnostic performance, clinician confidence, workflow efficiency, safety, and potential failure modes.
Ultimately, the project aims to establish a clinically reliable and deployable generative AI framework for ophthalmology, integrating model development, clinical evaluation, and implementation into routine ophthalmic workflows.