Key takeaways:
- The mannequin had an space underneath the curve of 0.92 within the detection of extreme glaucoma.
- Further coaching information may increase the mannequin’s efficiency in gentle to reasonable glaucoma.
MINNEAPOLIS — A deep studying mannequin skilled on fundus pictures confirmed promise within the detection of extreme glaucoma, with decrease accuracy in gentle to reasonable instances, in accordance with a poster introduced at Optometry’s Assembly.
“Glaucoma requires energetic screenings and requires a dependable and inexpensive automated system to allow early detection, particularly in high-risk communities,” Mikayla Kaliski, a third-year optometry scholar on the Southern California School of Optometry, stated in a video presentation. “This results in our challenge goal: to leverage synthetic intelligence to develop and consider the deep studying mannequin for detecting the presence or absence of glaucoma from fundus pictures and categorize the stage of glaucoma as gentle, reasonable or extreme.”
Kaliski and colleagues skilled a mannequin utilizing digital well being document “information based mostly on multimodal ophthalmologic investigation” matched with 6,665 fundus pictures from EyePACS, a diabetic retinopathy screening program. They assessed the mannequin’s sensitivity, specificity, accuracy and space underneath the curve.
The mannequin carried out greatest within the detection of extreme glaucoma, with 88% accuracy, 100% sensitivity and 83% specificity, with an space underneath the curve of 0.92.
As glaucoma severity decreased, so did sensitivity and space underneath the curve. Respectively, these had been 48% and 0.62 for reasonable glaucoma and 36% and 0.59 for gentle glaucoma.
“Total, the mannequin reveals promise for detecting extreme glaucoma from retinal pictures,” Kaliski stated. “To enhance the mannequin efficiency for detecting gentle and reasonable glaucoma levels, future work will contain incorporating further supply information to seize a various vary of affected person instances and improve the transferability of the algorithm, along with integrating [electronic medical record] information factors resembling affected person demographic, household historical past and age into coaching the mannequin to enhance accuracy and reliability throughout all levels of glaucoma.”
Sources/Disclosures
Collapse
Kaliski M, et al. Improvement of a deep studying mannequin for detecting gentle, reasonable, and extreme glaucoma. Offered at: Optometry’s Assembly; June 25-28, 2025; Minneapolis.
Disclosures:
Healio couldn’t verify related monetary disclosures on the time of publication.
