Description

Uveitis is an inflammatory condition of the uveal tract that represents a leading cause of preventable blindness worldwide. Establishing the correct etiology is notoriously challenging: more than 40% of uveitis cases remain idiopathic even after full workup, and misdiagnosis leads to delayed treatment with irreversible visual loss. Accurate etiological classification is therefore a critical, unmet clinical need.
Within this context, the PREUVEE project is developing a comprehensive AI-driven platform to enhance both etiological diagnosis and prognostic prediction of uveitis. The platform integrates machine learning, deep learning, image processing, and natural language processing to analyze heterogeneous patient data, including electronic health records, laboratory results, and multi-modal ophthalmic imaging. A core component of this platform is an etiological diagnostic algorithm that currently leverages structured clinical data and expert-interpreted ocular imaging insights. While early results are promising, the algorithm's performance is constrained by its limited exploitation of the rich, high-dimensional information encoded in raw ophthalmic images.
Ocular imaging modalities, principally fundus photography, fluorescein angiography (FA), and optical coherence tomography (OCT), capture structural and pathological details that are pivotal in diagnosing uveitis and identifying its underlying causes. Retinal vasculitis patterns, choroidal infiltrates, macular edema, and disc changes are among the many image-level biomarkers that carry strong etiological signals. Unlocking these signals through advanced AI techniques constitutes the central objective of this PhD project.

This PhD project pursues three interconnected scientific objectives:

• Extract and standardize fundus, angiographic, and OCT images in DICOM format from the hospital information system, establishing a high-quality, curated imaging dataset for model training and evaluation.
• Develop an innovative semi-supervised image feature extraction pipeline that reduces dependency on costly manual annotations while maximizing the diagnostic information harvested from both labelled and unlabeled data.
• Integrate the extracted image-derived features into the multimodal diagnostic algorithm and quantitatively assess the improvement in etiological classification performance relative to the clinical-data-only baseline.