Deep learning-based automated classification of choroidal layers in en face swept-source optical coherence tomography...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate classification of choroidal layers is crucial for diagnosing various eye conditions.
• Current manual methods for stratifying choroidal layers are time-consuming and subjective.
• En face swept-source optical coherence tomography (SS-OCT) provides detailed cross-sectional views of the choroid.

Purpose of the Study:

• To develop and validate a deep learning algorithm for automated choroidal layer segmentation.
• To improve the efficiency and accuracy of choroidal layer classification in SS-OCT imaging.
• To establish a reliable AI-based tool for ophthalmic research and clinical practice.

Main Methods:

• A cohort of 117 healthy subjects underwent SS-OCT volume scans.
• En face SS-OCT images were acquired at 2.6 μm intervals.
• A deep learning model based on ResNet architecture was developed and trained on 16,025 images.

Main Results:

• The algorithm achieved a balanced accuracy of 84.30% for choroidal layer classification.
• Classification accuracy remained high with an error range of 5.2 μm (92.61%).
• The model incorporated boundary-enhancing undersampling and subclass ensemble techniques for improved performance.

Conclusions:

• Deep learning enables accurate automated stratification of choroidal layers from en face SS-OCT images.
• The developed algorithm offers a promising tool for objective and efficient ocular tissue analysis.
• This AI-driven approach has the potential to advance ophthalmic diagnostics and research.