Evaluating the Influence of Clinical Data on Inter-Observer Variability in Optic Disc Analysis for AI-Assisted...

Area of Science:

• Ophthalmology
• Artificial Intelligence in Healthcare
• Medical Imaging Analysis

Background:

• Accurate glaucoma diagnosis relies on optic disc assessment from fundus photographs.
• Establishing reliable ground truth for AI development is challenged by subjective interpretations.
• Inter-observer variability in expert grading can affect diagnostic outcomes.

Purpose of the Study:

• To assess inter-observer variability in optic disc evaluation using fundus photographs.
• To determine the impact of clinical data, such as intraocular pressure (IOP), on diagnostic consistency.
• To evaluate the implications of this variability for AI-driven glaucoma research.

Main Methods:

• Two glaucoma specialists initially classified 70 subjects' fundus photos as normal or abnormal.
• Referrals were made based on classifications, followed by IOP measurements.
• Four specialists independently categorized images in a second stage, with and without additional clinical data (IOP, contralateral eye).

Main Results:

• Moderate agreement was found between senior and junior specialists in the initial assessment.
• Intraocular pressure (IOP) was identified as a key factor influencing referral decisions.
• Agreement among four specialists varied, improving with additional clinical information; significant variability existed in optic disc excavation assessment.

Conclusions:

• Clinical risk factors, including IOP and bilateral data, significantly impact diagnostic accuracy and consistency.
• Sole reliance on fundus photographs for AI training is problematic due to inter-observer variability.
• Multimodal datasets integrating clinical information are crucial for developing robust AI models for glaucoma screening.