Diagnostic Performance of the Offline Medios Artificial Intelligence for Glaucoma Detection in a Rural...

Area of Science:

• Ophthalmology
• Medical Artificial Intelligence
• Diagnostic Technology

Background:

• Glaucoma diagnosis in primary eye care settings presents challenges due to limited specialist access.
• Developing efficient and accurate diagnostic tools is crucial for early glaucoma detection and management.

Purpose of the Study:

• To evaluate the diagnostic performance of an offline Artificial Intelligence (AI) glaucoma software using nonmydriatic fundus images.
• To compare the AI's diagnostic efficacy against tele-ophthalmologists and a glaucoma specialist in a primary eye care setting.

Main Methods:

• A prospective cross-sectional study involving 303 participants at satellite vision centers.
• Fundus images were analyzed by Medios AI software and compared with diagnoses from tele-ophthalmologists and a glaucoma specialist.
• Sensitivity, specificity, and agreement rates were calculated to assess diagnostic accuracy.

Main Results:

• The AI achieved a sensitivity of 0.91 and specificity of 0.93 in detecting referable glaucoma compared to tele-ophthalmologists.
• Agreement between the AI and glaucoma specialist was 80.3%, exceeding the 55.3% agreement between tele-ophthalmologists and the specialist.
• The AI demonstrated fewer false positive referrals than tele-ophthalmologists.

Conclusions:

• Offline Medios AI software shows promising accuracy in identifying referable glaucoma from remote primary eye care settings.
• The AI's diagnostic agreement with specialists suggests its potential as a valuable tool to support glaucoma screening and diagnosis.