Automated Detection of Keratorefractive Laser Surgeries on Optical Coherence Tomography Using Deep Learning.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Anterior segment optical coherence tomography (AS-OCT) is crucial for eye imaging.
• Accurate identification of prior keratorefractive laser surgeries is vital for subsequent eye care.
• Distinguishing between different types of laser procedures and their refractive targets (myopic vs. hyperopic) can be challenging.

Purpose of the Study:

• To develop and evaluate a deep learning neural network for automated detection of various keratorefractive laser surgeries using AS-OCT.
• To differentiate between myopic and hyperopic treatments within identified surgical categories.
• To assess the algorithm's performance in classifying non-operated eyes as a control group.

Main Methods:

• A deep learning neural network was trained and validated on 14,948 AS-OCT eye scans from 1,166 patients.
• An 80/10/10 distribution was used for training, validation, and testing datasets.
• Performance metrics included accuracy, F1 scores, AUPRC, and AUROC.

Main Results:

• The neural network achieved 96% accuracy and a 96% weighted-average F1 score in detecting different surgical classes.
• It accurately identified hyperopic and myopic subclasses within surgical categories with 90% accuracy and a 90% weighted-average F1 score.
• The algorithm demonstrated high performance across various metrics on the test dataset.

Conclusions:

• Deep learning models can reliably classify keratorefractive laser surgery history from AS-OCT scans.
• This technology can enhance treatment planning, intraocular lens calculations, and ectasia risk assessment.
• AS-OCT, augmented by AI, can evolve into a comprehensive screening tool in refractive clinics.