SEVEN MAIN LIMITATIONS OF AI IN ANTERIOR SEGMENT DISEASES

Published 2025 - 29th ESCRS Winter Meeting

Reference: FP06.09 | Type: Free Paper | DOI: 10.82333/kp2z-ps87

Authors: Dimitri Azar* ¹ , Luis De Sisternes ²

¹Ophthalmology,University of IL,Chicago,United States, ²Verily Life Sciences,San Francisco,United States

Purpose

To identify major technical limitations of the generalizeability of AI in the diagnosis of anterior segment diseases

Setting

Literature review and data synthesis.

Methods

We reviewed relevant AI medical literatutre to identify the limitations in the generalizeability of AI in anterior segment diagnosis

Results

Seven major limitations were identified:

I.Background knowledge bias

II. Data representation bias

III. Data collection bias

IV. Model development / processing

V. Model evaluation

VI. Model deployment / use

VII. Model performance monitoring

Examples of these seven biases were identified in generative, LLM and CNN artifial intelligence applications for anterior segment diagnosis. Methods for reducing these potential limitations were identified. These will be discussed during the presentation.

Conclusion

Big data, cloud connectivity and portable diagnostic devices have led to a tremendous growth of AI applications in anterior segment disease diagnosis. With the increasing use of AI in anterior semnet disease diagnosis, there are seven major limitations to generizeability of the results. We will describe the biases identified in generative, LLM and CNN artifial intelligence applications for anterior segment diagnosis and the methods for reducing these potential limitations.