Abstract bekijken

Deze abstract is toegekend aan sessie AOB Free Papers 2
Type presentatieOral presentation
TitelArtificial Intelligence Glaucoma Screening: Real-World Implementation and Validation
DoelThis study evaluates whether artificial intelligence (AI) - based screening can detect asymptomatic patients and reduce morbidity.
MethodesParticipants aged between 55-65 years identified from primary care databases were randomly invited to undergo glaucoma screening. Glaucoma risk was determined from the fundus image using an AI algorithm (MONA G-Risk®). Subjects were referred for further glaucoma assessment according to the AI’s outcome or an IOP of ≥24mmHg. Validation of the screening protocol was performed by reviewing the entire fundus photos dataset by six paired glaucoma experts. Referred participants performed visual field test as well as a face-to-face assessment with ophthalmologist.
ResultatenOf 1,038 subjects invited, 65% (671) were screened. The AI algorithm referred 9.8% (66/671) of participants while human-expert fundus review referred 77 more (143/671 or 21.3%). After excluding 42 who missed VF testing, 629 participants remained for analysis. Glaucoma was found in 6.4% (40/629; 95% CI 5–9%) and IOP ≥24 mmHg in 3.2% (20/629; 95% CI 2–5%). The AI algorithm’s sensitivity and specificity for detecting glaucoma were 78% (95% CI 62–89%) and 95% (95% CI 93–97%). The incremental cost-effectiveness ratio was €1,743 for the AI screening program implementation versus standard care scenario over a ten year horizon.
ConclusieGlaucoma screening with AI technology outperformed human detection. AI could boost screening efficiency and case detection, paving the basis for earlier treatment and avoidable blindness prevention.
BelangenverstrengelingNee
Details verstrengelde belangenNone
Auteurs 1
InitialenA
NaamCabrita
InstituutSanta Maria University Hospital, Lisbon
StadLisbon
Auteurs 2
InitialenZ
NaamWehbi
InstituutUZ Leuven, Belgium
StadLeuven
Auteurs 3
InitialenI
NaamStalmans
InstituutUZ Leuven, Belgium
StadLeuven
Auteurs 4
InitialenL
NaamPinto
InstituutSanta Maria University Hospital, Lisbon
StadLisbon
top ^