Surgical scene generation and adversarial networks for physics-based iOCT synthesis

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Intraoperative optical coherence tomography (iOCT) integration enhances ophthalmic surgery.
• Computer-assisted algorithms can improve iOCT-guided interventions.
• Limited iOCT data availability hinders algorithm development.

Purpose of the Study:

• To introduce a novel framework for generating synthetic iOCT data using a virtual setup and deep learning.
• To overcome the constraints of limited real iOCT data for research and development.
• To demonstrate the utility of synthetic iOCT data for advancing ophthalmic surgical procedures.

Main Methods:

• A virtual setup was created to simulate retinal layer geometry from real data.
• Virtual microsurgical instruments were integrated into the simulation.
• Scene rendering incorporated iOCT imaging artifacts to generate label maps.
• A generative adversarial network (GAN) synthesized iOCT B-scans from label maps.

Main Results:

• The study generated synthetic iOCT B-scans and label maps.
• Experiments investigated the similarity between real and synthetic iOCT images.
• The relevance of synthetic data for image-guided interventions was demonstrated.
• The potential for 3D iOCT data synthesis was shown.

Conclusions:

• The proposed framework successfully generates synthetic iOCT data.
• Synthetic iOCT data can be valuable for developing and testing computer-assisted ophthalmic surgery algorithms.
• This approach offers a scalable solution to iOCT data scarcity, promoting innovation in the field.