Novel Machine-Learning Based Framework Using Electroretinography Data for the Detection of Early-Stage Glaucoma.

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Computational Neuroscience

Background:

• Early glaucoma diagnosis remains a significant challenge in ophthalmology.
• Current diagnostic methods primarily rely on structural measures (e.g., OCT), often overlooking the potential of functional measures like electroretinograms (ERGs).
• There is a need for advanced techniques to leverage the rich information within ERG signals for early disease detection.

Purpose of the Study:

• To develop a novel, reliable predictive framework for early glaucoma detection.
• To create a machine-learning-based algorithm capable of analyzing electroretinogram (ERG) signals for medically relevant information.
• To establish a foundational step towards improving glaucoma diagnosis by integrating functional measures.

Main Methods:

• ERG signals from 60 DBA/2 mouse eyes were collected and grouped for binary and multiclass classification based on age and intraocular pressure (IOP).
• Statistical and wavelet-based features were engineered and extracted from the ERG signals.
• Five machine learning algorithms were evaluated, with a focus on identifying important predictors and classification performance.

Main Results:

• The random forest (bagged trees) ensemble classifier achieved the highest accuracy, reaching 91.7% for binary classification and 80% for multiclass classification.
• The results indicate that machine learning models can detect subtle changes in ERG signals when trained with advanced features, such as those derived from wavelet analyses.
• The study successfully identified key ERG tests and features crucial for accurate glaucoma classification.

Conclusions:

• A novel machine-learning-based method for analyzing ERG signals was developed, offering additional insights for early glaucoma detection.
• The proposed framework demonstrated promising performance metrics in detecting functional deficits associated with early-stage glaucoma in mice.
• This approach highlights the potential of leveraging functional ERG data with advanced machine learning for improved glaucoma diagnosis.