# Enhancement of Optical Coherence Tomography Images Using Adversarial Neural Networks: Impacts on Ophthalmic Practice

## TL;DR

This study explores how AI can improve OCT images used in eye care, finding that it enhances image clarity but has mixed effects on clinical usefulness.

## Contribution

The novel application of Real-ESRGAN GANs to enhance OCT images and assess their impact on diagnostic accuracy and ophthalmologist perception.

## Key findings

- 69.4% of ophthalmologists reported improved clarity of retinal layers in AI-enhanced OCT images.
- 78.9% acknowledged that AI-enhanced images significantly highlight biomarkers.
- 60.5% of participants did not find AI-enhanced images useful for case management.

## Abstract

Purpose: The integration of artificial intelligence (AI) into ophthalmology has rapidly advanced, particularly in the analysis of optical coherence tomography (OCT) images. OCT is a non-invasive imaging modality widely used to visualize retinal layers for diagnostic and therapeutic purposes. Recent advancements suggest that AI can enhance image resolution, interpret data more effectively, and improve clinical decision-making. This study's purpose is to evaluate the application of an AI-driven filter to enhance OCT images, focusing on its impact on image clarity and diagnostic accuracy.

Methods: We utilized generative adversarial networks (GANs), specifically the real-enhanced super-resolution GAN (Real-ESRGAN), to enhance the resolution and clarity of OCT images. The study included eight anonymized OCT images from the Ophthalmology Department at the Universidade Federal de São Paulo (UNIFESP), featuring various retinal pathologies. A total of 147 ophthalmologists (96 retinal specialists and 51 non-specialists) assessed the images before and after AI enhancement. Descriptive statistics, contingency tables, and chi-squared tests were employed to analyze the data, using the Jamovi software (jamovi (Version 2.3) [Computer Software]. Retrieved from https://www.jamovi.org) and the R software (R Foundation for Statistical Computing, Vienna, Austria) with a significance level set at 5%.

Results: The majority of ophthalmologists (69.4%) found that AI-enhanced images improved the clarity of retinal layers, with external retinal structures being particularly well-evaluated. However, opinions varied regarding the clinical utility of AI-enhanced images in diagnostic and prognostic contexts. While 61.9% of participants agreed that AI facilitated biomarker identification, 60.5% did not believe it provided additional relevant information for case management. Notably, AI-enhanced images were considered to significantly highlight biomarkers, with 78.9% of ophthalmologists acknowledging this benefit.

Conclusions: AI-enhanced OCT images, particularly those processed with GANs, can significantly improve the clarity and detail of retinal images, which may aid in more precise diagnostics. Despite the positive feedback, the variability in perceived clinical utility suggests that AI integration should be approached cautiously. Training and experience in using AI tools are crucial, and further research with larger samples and diverse clinical settings is needed to better understand AI's impact on clinical practice and its cost-benefit ratio. Future studies should explore the comparative efficacy of different AI algorithms and their potential to enhance diagnostic accuracy and patient outcomes.