View abstract

Cet abstract a été assigné à session AOB Free Papers 2
Type de présentationOral presentation
TitreArtificial Intelligence (AI) for Community-based Glaucoma Risk Stratification
ButTo validate a deep learning model for automated glaucoma risk stratification from fundus images by comparison with expert grading in a community screening setting.
MéthodesWe conducted a prospective multicenter diagnostic accuracy study, screening 1,378 individuals over the age of 40 years in the foyer of eight Belgian hospitals using questionnaires, rebound tonometry, and fundus photography, with independent image grading by 6 glaucoma experts and an AI tool.
RésultatsParticipants were more often female (63.8%), with a mean age of 60.1 years (SD 10.8); 34 (3.3%) having self-reported glaucoma and the mean intraocular pressure was 14.8 mmHg (SD 3.9). The AI risk score showed 100% sensitivity, 74.8% specificity, and an Area Under the Curve (AUC) of 0.93. The Intraclass Correlation Coefficient values were 0.73 for cup-to-disc ratio, 0.55 for image–only risk scores, and 0.54 for risk scores including metadata.
ConclusionVariability among glaucoma expert assessments illustrates the challenge of consistently grading glaucoma risk from fundus images. Against this backdrop, the AI model demonstrated robust accuracy (AUC 0.93), underscoring its potential to act as a supportive tool for ophthalmologists, enhancing efficiency and consistency in identifying individuals who may benefit from further specialist evaluation.
Conflit d'intérêtOui
Détails conflits d'intérêtIngeborg Stalmans is co-founder, shareholder, and consultant of MONA.health.
Auteurs 1
InitialesT
NomJacobs
InstitutUZL/KUL
VilleLeuven
Auteurs 2
InitialesA
NomEhongo Bidime
InstitutHôpital Erasme
VilleBrussels
Auteurs 3
InitialesN
NomCollignon
InstitutCHU Liège
VilleLuik
Auteurs 4
InitialesP
NomKestelyn
InstitutAZ Sint-Lucas
VilleGhent
Auteurs 5
InitialesS
NomKiekens
InstitutUZA
VilleAntwerp
Auteurs 6
InitialesS
NomPourjavan
InstitutCliniques Universitaires Saint-Luc
VilleBrussels
Auteurs 7
InitialesS
NomVan de Veire
InstitutAZ Sint-Jan
VilleBruges
Auteurs 8
InitialesV
Nom De Groot
InstitutZAS Cadix
VilleAntwerp
Auteurs 9
InitialesI
NomStalmans
InstitutUZL/KUL
VilleLeuven
Auteurs 10
InitialesJ
NomVan Eijgen
InstitutUZL/KUL
VilleLeuven
top ^