Development of a Deep Learning System for Identification of Neuro-Ophthalmological Conditions on Color Fundus Photographs in Emergency Department

Overview

In recent years, artificial intelligence (AI) has been widely integrated into the medical field, contributing in particular to improved patient diagnosis. The BONSAI study, Brain and Optic Nerve Study with AI, in which our team is participating, has successfully demonstrated the ability of AI to identify individual neuro-ophthalmological or neurological pathologies affecting the optic nerves and/or brain, from a simple fundus image. While this is a promising advance, it remains limited in current clinical practice. Our major challenge is to be able to identify a wider range of optic nerve and/or brain pathologies simultaneously in the same analysis, so as to improve patient management, especially for those referred to emergency departments. Indeed, in the absence of a precise diagnosis, complications can be irreversible and life-threatening. Among the most alarming clinical signs in the emergency department is papilledema of stasis, which, accompanied by acute headaches, may indicate the presence of intracranial hypertension, inflammatory or ischemic pathology. The latter may be a manifestation of Horton's disease. Our team has developed an AI algorithm to diagnose retinal and optic nerve abnormalities based on retinophotographs taken under ideal conditions during scheduled consultations, and not on images of patients presenting to the emergency department. In hospitals without ophthalmology emergency departments, it is essential that emergency physicians (emergency physicians, general practitioners, neurologists) are able to assess the fundus in the absence of an ophthalmology specialist. This assessment, although part of the general examination, often presents challenges for non-ophthalmologists. The aim of our study is to improve the performance of our AI algorithm so that it can discriminate between different retinal and optic nerve pathologies in the emergency department. We therefore plan to build a database of fundus images by prospectively including patients presenting to the ophthalmology and neurology emergency departments of the Fondation Adolphe de Rothschild Hospital. The performance of the algorithm developed will be evaluated according to standard criteria of sensitivity, specificity, area under the curve (AUC) and accuracy.

Primary outcome measures

• Proportion of correct predictions among all positive predictions out of all total predictions of the algorithm (positive + negative) [Time frame: Day 30]

Secondary outcome measures (3)

• Sensitivity of the algorithm for each eye disease [Time frame: Day 30]
• Specificity of the algorithm for each eye disease [Time frame: Day 30]
• Area under the curve (AUC) for each eye disease [Time frame: Day 30]

Eligibility criteria

Inclusion criteria

• Patient aged 18 and over
• Presenting to the emergency department of the Fondation Adolphe de Rothschild hospital
• Express consent to participate in the study
• Member or beneficiary of a social security scheme

Exclusion criteria

• Patient under legal protection
• Pregnant or breast-feeding women

Criteria are shown verbatim from the registry (in English). Final eligibility is always assessed by the study center.

Healthy volunteers: Yes

Study design

Observational model

Cohort

Study locations

France · 1 center

• Hôpital Fondation Adolphe de Rothschild — Paris

Identifiers

NCT: NCT06643338 · DMA_2024_6

Primary sources (government registries)

• ClinicalTrials.gov (US NIH)

View this study on ClinicalTrials.gov ↗