DR. SWATI UPADHYAYA

DR. RENGARAJ VENKATESH, DR. KAVITHA S.

Abstract

Aim: To assess sensitivity and specificity of offline AI in diagnosing glaucoma. Methods: Disc images of 303 subjects taken on a fundus on phone camera, at 6 vision centers of a tertiary eye hospital, were analyzed by the offline AI and by VC doctors masked to the diagnosis given by the AI, through teleophthalmology and were graded as normal, glaucoma or disc suspect. Subjects deemed referable by either the AI or the VC doctor were referred to a glaucoma specialist at the base hospital. The AI output was compared against VC doctor and specialist diagnosis. Results: When compared against the VC doctors, the sensitivity and specificity of the AI to detect referral warranted glaucoma (glaucoma and suspects) was 73.17% and 98.6%. For subjects referred to the base hospital and evaluated by the glaucoma specialist, sensitivity of AI to pick up confirmed glaucoma was 96.43% where it was 53.3% for the VC doctor.

Conclusion: AI can prove to be effective in reducing over diagnosis and referral

Full Text