Carbon-Neutral Approach to Accurately Diagnosing Central Serous Retinopathy Using AI

Abstract

Purpose :
The objective is to utilize the power of artificial intelligence (AI) to improve the diagnostic capabilities for central serous chorioretinopathy (CSCR), the 4th most common potentially blinding retinal disease. CSCR is characterized by localized detachment of the macula due to build up and it presents a significant challenge for clinicians who need to identify it accurately and promptly for optimal patient outcomes. CSCR has an incidence rate of around 10 per 100,000 in men making it one of the most prevalent retinal diseases that can cause visual impairment. This research aims to utilize an AI model developed on Google's Collaboration platform to detect CSCR in Optical Coherence Tomography (OCT) scans.

Methods : To develop the AI model we used a dataset from Kaggle.com that contained a total of 1996 images which consisted of 999 normal OCT scans and 997 CSCR OCT scans. These images were carefully divided into training, validation and testing sets. We leveraged Google's Collaboration platform to ensure model training within just 1 hour and 53 minutes. This approach not only saved costs but also aligned with environmentally sustainable practices. The performance of the model was rigorously evaluated using metrics such as accuracy, precision, recall (sensitivity), specificity and F1 score.

Results : The AI model showcased accuracy and reliability achieving a flawless 100% score across critical performance metrics like accuracy, precision, recall (sensitivity) specificity and F1 score. Furthermore, the model displayed an Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) value of 1. This demonstrates the model’s ability to effectively distinguish between individuals with CSCR and those with normal retinas.

Conclusions : This study not only establishes the application of AI in ophthalmological diagnostics but also addresses a major retinal disease, CSCR – one of the leading causes of visual impairment. This research has the potential to make an impact on patient care and clinical outcomes by reducing the workload of clinicians and making diagnostic processes more efficient. The results highlight how AI technologies can revolutionize the diagnosis of retinal diseases encouraging further investigation into their integration in everyday clinical practice.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.