A Novel Fundus Image Reading Tool for Efficient Generation of a Multi-dimensional Categorical Image Database for...

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• A novel multi-step retinal fundus image reading system was developed.
• This system aims to provide high-quality, large-scale data for machine learning algorithms.
• Grader variability in a large dataset generated by this system was assessed.

Purpose of the Study:

• To develop and evaluate a novel multi-step retinal fundus image reading system.
• To assess grader variability in a large-scale dataset generated by this system.
• To generate a high-quality dataset for machine learning algorithm development in ophthalmology.

Main Methods:

• A 5-step retinal fundus image reading tool was created.
• The tool rates image quality, abnormality presence, findings with location, diagnoses, and clinical significance.
• Each image was evaluated by three licensed ophthalmologists, and inter-grader agreements were analyzed.

Main Results:

• Over 234,000 readings from 79,000+ images were collected from 55 ophthalmologists.
• Complete agreement among three graders on abnormality was 46.6%, with 69.9% agreement by at least two raters.
• Inter-grader agreement rates for findings and diagnoses varied, with higher agreement on abnormalities and by specialists in their subspecialties.

Conclusions:

• The novel reading tool successfully generated a large-scale, informative dataset for machine learning.
• This dataset can support the development of AI for automated disease identification and clinical decision support.
• Addressing grader variability is crucial for the development of robust machine learning algorithms in ophthalmology.