Lesion detection with fine-grained image categorization for myopic traction maculopathy (MTM) using optical coherence...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Myopic traction maculopathy (MTM) causes vision loss in high myopia due to macular traction.
• Optical coherence tomography (OCT) is crucial for diagnosing and classifying MTM patterns.
• OCT-based classification guides clinical management strategies for MTM.

Purpose of the Study:

• To develop a deep learning model for automated MTM identification in highly myopic (HM) eyes.
• To classify MTM into five distinct patterns using OCT images.

Main Methods:

• A ResNet-34 deep learning architecture was employed.
• The model was trained on 2837 OCT images from 958 HM patients.
• Performance was evaluated on an independent test set of 604 images using AUC, accuracy, sensitivity, and specificity.

Main Results:

• The model achieved high training performance (F1-score 0.953, AUCs 0.961-0.998).
• On the test set, sensitivities ranged from 91.67% to 97.78%, and specificities from 98.33% to 99.17%.
• The model's performance was comparable or superior to experienced clinicians, with visual explanations provided via heatmaps.

Conclusions:

• A deep learning model for MTM classification using OCT images was successfully developed.
• The model demonstrates performance on par with or exceeding retinal specialists.
• This AI tool is suitable for large-scale screening and MTM identification in HM eyes.