Deep learning-based prediction of the retinal structural alterations after epiretinal membrane surgery

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Epiretinal membrane (ERM) removal surgery requires careful postoperative monitoring.
• Optical coherence tomography (OCT) is crucial for assessing surgical success.
• Predictive imaging could aid in surgical planning and patient education.

Purpose of the Study:

• To generate synthesized postoperative OCT images of ERM using deep learning.
• To evaluate the accuracy of these synthesized images compared to real postoperative OCT scans.
• To assess the potential of AI in predicting surgical outcomes for ERM.

Main Methods:

• A Pix2Pix Generative Adversarial Network (GAN) model was trained on 500 pairs of preoperative and postoperative OCT images.
• The model generated 60 synthesized postoperative OCT images from preoperative scans.
• Structural Similarity Index Measure (SSIM) was used to quantify image similarity.

Main Results:

• The Pix2Pix GAN model successfully generated 60 synthesized postoperative OCT images.
• The mean SSIM between synthesized and actual postoperative OCT images was 0.913.
• High SSIM indicates strong similarity between predicted and real images.

Conclusions:

• Deep learning, specifically the Pix2Pix GAN model, can generate accurate predictive postoperative OCT images after ERM surgery.
• Synthesized OCT images show high structural similarity to actual postoperative scans.
• This technology holds potential for improving preoperative planning and postoperative assessment in ERM treatment.