Using deep learning to identify inherited retinal diseases based on wide-field retinal imaging data

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Inherited retinal diseases (IRDs) are a diverse group of genetic disorders affecting vision.
• Accurate and timely diagnosis of IRDs is crucial for patient management and genetic counseling.
• Current diagnostic methods often rely on genetic testing, which can be time-consuming and costly.

Purpose of the Study:

• To evaluate a novel deep learning (DL) method for automated IRD identification using multimodal fundus imaging.
• To assess the feasibility of predicting causative gene groups based on fundus autofluorescence (FAF) and pseudocolor fundus (pCF) images.
• To explore the potential of DL in providing decision support for clinicians prior to genetic testing.

Main Methods:

• A retrospective dataset of 409 patients with and without IRD was analyzed.
• Wide-field FAF and pCF images, along with genetic test results, were utilized.
• Nine EfficientNet-V2-m convolutional neural networks were trained for binary IRD classification and classification into two causative gene groups.

Main Results:

• Multimodal DL models using both FAF and pCF images achieved the highest performance.
• The binary IRD vs. non-IRD classification model demonstrated a mean accuracy of 0.95 ± 0.01.
• Gene group classification models achieved mean accuracies of 0.92 ± 0.03 (Group 1) and 0.85 ± 0.03 (Group 2).

Conclusions:

• Image-based deep learning classifiers can accurately identify IRD from FAF and pCF images.
• This AI approach offers potential decision support for clinicians during initial patient visits.
• Further prospective validation is necessary before widespread clinical implementation.