Deep learning models for screening of high myopia using optical coherence tomography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• High myopia poses a significant risk for serious ocular pathologies.
• Accurate and efficient screening methods are crucial for early detection and management.
• Spectral-domain optical coherence tomography (OCT) provides detailed cross-sectional retinal images.

Purpose of the Study:

• To validate and evaluate deep learning (DL) models for high myopia screening using OCT images.
• To compare the diagnostic performance of DL models with human retina specialists.

Main Methods:

• A retrospective cross-sectional study involving 690 eyes from 492 patients.
• Eyes classified into normal, high myopia, and other retinal disease groups based on axial length.
• Three DL models (ResNet 50, Inception V3, VGG 16) trained and validated on OCT images.

Main Results:

• The ResNet 50 DL model achieved 100.0% absolute agreement and an area under the receiver operating characteristic curve of 0.99.
• DL model performance was comparable to retina specialists, who had 99.11% absolute agreement.
• The ResNet 50 model demonstrated reliable diagnostic performance for identifying high myopia.

Conclusions:

• Deep learning models, particularly ResNet 50, show high accuracy in screening for high myopia using OCT.
• These DL models offer a reliable and potentially efficient tool for ophthalmologists in diagnosing high myopia.