Automated Early-Stage Glaucoma Detection Using a Robust Concatenated AI Model

Glaucoma is a leading cause of irreversible blindness worldwide; therefore, detection of this disease in its early stage is crucial. However, previous efforts to identify early-stage glaucoma have faced challenges, including insufficient accuracy, sensitivity, and specificity. This study presents a...

Full description

Saved in:
Bibliographic Details
Main Authors: Wheyming Song, Ing-Chou Lai
Format: Article
Language:English
Published: MDPI AG 2025-05-01
Series:Bioengineering
Subjects:
Online Access:https://www.mdpi.com/2306-5354/12/5/516
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Glaucoma is a leading cause of irreversible blindness worldwide; therefore, detection of this disease in its early stage is crucial. However, previous efforts to identify early-stage glaucoma have faced challenges, including insufficient accuracy, sensitivity, and specificity. This study presents a concatenated artificial intelligence model that combines two types of input features: fundus images and quantitative retinal thickness parameters derived from macular and peri-papillary retinal nerve fiber layer (RNFL) thickness measurements. These features undergo an intelligent transformation, referred to as “smart preprocessing”, to enhance their utility. The model employs two classification approaches: a convolutional neural network approach for processing image features and an artificial neural network approach for analyzing quantitative retinal thickness parameters. To maximize performance, hyperparameters were fine-tuned using a robust methodology for the design of experiments. The proposed AI model demonstrated outstanding performance in early-stage glaucoma detection, outperforming existing models; its accuracy, sensitivity, specificity, precision, and <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msub><mi>F</mi><mn>1</mn></msub></semantics></math></inline-formula>-Score all exceeding 0.90.
ISSN:2306-5354