Automated cornea diagnosis using deep convolutional neural networks based on cornea topography maps
Benjamin Fassbind1, Achim Langenbucher2, Andreas Streich3
1Department of Computer Science, Lucerne University of Applied Sciences and Arts, Rotkreuz/Zug, 6343, Switzerland. benjamin.fassbind@hotmail.com.
Scientific Reports
|April 21, 2023
Summary
This study introduces an AI model using Convolutional Neural Networks (CNNs) to automatically detect corneal abnormalities from Optical Coherence Tomography (OCT) scans. The AI accurately identifies conditions like keratoconus, improving diagnostic efficiency.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Corneal topography maps are crucial for diagnosing eye conditions.
- Automated analysis of these maps can enhance diagnostic accuracy and efficiency.
Purpose of the Study:
- To develop and validate a deep learning model for automatic detection of corneal abnormalities, specifically keratoconus.
- To assess the model's performance in distinguishing between healthy and pathological corneas using Optical Coherence Tomography (OCT) scans.
Main Methods:
- Utilized a dataset of 1940 anterior segment OCT scans from Saarland University Hospital.
- Applied a fine-tuned ConvNeXt Convolutional Neural Network (CNN) architecture for image analysis.
- Trained and validated the model on annotated scans to identify corneal pathologies.
Main Results:
- Achieved a sensitivity of 98.46% and a specificity of 91.96% in differentiating healthy from pathological corneas.
- Demonstrated the model's capability for screening corneal pathologies, including keratoconus.
- The approach is scanner-independent and provides visual explanations for its decisions.
Conclusions:
- The developed AI model effectively screens for corneal pathologies using OCT scans.
- This automated approach aids in the early detection and classification of conditions like keratoconus.
- The model's interpretability enhances clinical trust and utility.
