Deep Learning-Based Classification of Inherited Retinal Diseases Using Fundus Autofluorescence.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning is increasingly used for ophthalmological diseases.
• Inherited retinal diseases (IRDs) are rare genetic conditions with distinct fundus autofluorescence (FAF) imaging phenotypes.
• Automated classification of IRDs using FAF imaging is an emerging area.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for the automatic classification of inherited retinal diseases (IRDs) based on fundus autofluorescence (FAF) images.
• To differentiate between specific IRDs (retinitis pigmentosa, Best disease, Stargardt disease) and healthy controls using FAF imaging.

Main Methods:

• A multilayer deep convolutional neural network (CNN) was trained and validated using 389 FAF images.
• Established augmentation techniques and an Adam optimizer were employed for training.
• Classifiers were tested on 94 independent FAF images.

Main Results:

• The deep learning classifiers achieved a global accuracy of 0.95 for IRD detection.
• High precision-recall area under the curve (PRC-AUC) values were obtained: 0.988 for Best disease, 0.999 for retinitis pigmentosa, 0.996 for Stargardt disease, and 0.989 for healthy controls.

Conclusions:

• A deep learning-based algorithm can effectively detect and classify inherited retinal diseases from FAF images.
• The developed classifiers demonstrated excellent performance, suggesting their potential as a diagnostic tool.
• This AI model may provide valuable insights for future therapeutic strategies in IRDs.