Related Experiment Video
Updated: Jul 2, 2026
06:15
Anterior High-Resolution Optical Coherence Tomography in the Diagnosis and Therapeutic Monitoring of Ocular Surface Squamous Neoplasia
Published on: August 9, 2024
Artificial intelligence for diagnosis of keratoconus using Scheimpflug based corneal tomography
Sadaf Qayyum1, Memoona Arshad1, Humaima Saeed1
1Department of Optometry, The University of Faisalabad, Faisalabad 38000, Punjab, Pakistan.
International Journal of Ophthalmology
|July 1, 2026
Summary
Deep learning models accurately differentiate keratoconus (KC) from normal eyes using corneal topography. These AI tools show high diagnostic potential for precise KC management.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Keratoconus (KC) is a progressive corneal ectasia affecting vision.
- Accurate differentiation from regular astigmatism is crucial for timely management.
- Current diagnostic methods can be subjective and time-consuming.
Purpose of the Study:
- To develop and evaluate deep learning (DL) models for diagnosing KC.
- To assess the diagnostic accuracy of DL models using corneal topography data.
- To compare the performance of different DL architectures in KC detection.
Main Methods:
- A cross-sectional study utilizing Galilei dual Scheimpflug corneal topography.
- Four corneal maps (anterior/posterior curvature, thickness, elevation) were analyzed.
- Four convolutional neural network models (DenseNet-121, ResNet-50, Inception-V3, EfficientNet-B0) were trained and validated.
Main Results:
- DL models achieved high diagnostic accuracy, with DenseNet-121 reaching 99.2% and ResNet-50 reaching 99.0%.
- DenseNet-121 and ResNet-50 demonstrated a perfect Area Under the Curve (AUC) of 1.00.
- External validation confirmed excellent accuracies, with EfficientNet-B0 at 98.1% and DenseNet-121 at 98.3%.
Conclusions:
- Deep learning models exhibit excellent diagnostic accuracy for keratoconus detection.
- These AI-driven approaches hold significant potential for clinical implementation.
- Optimized KC management with enhanced precision is achievable using these DL tools.
