ARTIFICIAL INTELLIGENCE-ENHANCED OPTICAL COHERENCE TOMOGRAPHY ANALYSIS FOR DETECTING INTERNAL LIMITING MEMBRANE...

Area of Science:

• Ophthalmology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Epiretinal membrane (ERM) removal is a common ophthalmic surgical procedure.
• Distinguishing between ERM-only removal and combined internal limiting membrane (ILM) and ERM peeling is crucial for surgical assessment.
• Optical coherence tomography (OCT) is a key imaging modality for evaluating retinal structures post-surgery.

Purpose of the Study:

• To evaluate the accuracy of human graders and artificial intelligence (AI) models in identifying surgical techniques for ERM removal using postoperative OCT scans.
• To compare the performance of different AI models against human interpretation in classifying ERM surgery types.

Main Methods:

• Retrospective analysis of 250 eyes undergoing vitrectomy for idiopathic ERM.
• Classification of surgeries into ERM-only removal or ILM+ERM removal groups based on surgical records.
• Training and testing of human graders and AI models (ResNet18, UwU-OrthLatt, UwU-PR-Relax) on labeled and masked OCT scans.

Main Results:

• Human grader accuracy was 50%.
• AI models demonstrated superior performance, with accuracies ranging from 61±3% (ResNet18) to 70±5% (UwU-OrthLatt).
• AI models significantly outperformed human interpretation in identifying ILM removal.

Conclusions:

• AI models show a higher capability than human graders in detecting ILM removal from OCT scans after ERM surgery.
• AI has significant potential to enhance the accuracy and efficiency of ophthalmic imaging analysis in clinical practice.