Convolutional neural network-based common-path optical coherence tomography A-scan boundary-tracking training and...

Area of Science:

• Medical Imaging
• Computational Ophthalmology
• Artificial Intelligence in Medicine

Background:

• Image-guided procedures require accurate real-time imaging.
• Common-path optical coherence tomography (CP-OCT) offers potential for needle insertion guidance.
• Generating large, diverse datasets for training AI models is challenging.

Purpose of the Study:

• To develop a parallel Monte Carlo (MC) simulation platform for synthetic CP-OCT A-scan image generation.
• To create a large dataset for training an AI model for Descemet's membrane (DM) localization.
• To evaluate the performance of the trained AI model in image-guided needle insertion.

Main Methods:

• A parallel Monte Carlo (MC) simulation platform was developed to generate synthetic CP-OCT A-scan images.
• 100,000 A-scan images were generated using 50 distinct eye models.
• An end-to-end convolutional neural network, Ascan-Net, was trained on the synthetic dataset for DM localization.

Main Results:

• The MC simulation platform demonstrated efficient computation time across various configurations.
• The trained Ascan-Net achieved improved tracking accuracy for DM localization.
• Performance was validated on ex-vivo human and porcine cornea data and simulated datasets, outperforming the Canny-edge detector.

Conclusions:

• The developed MC simulation platform effectively generates synthetic CP-OCT data for AI training.
• Ascan-Net shows significant potential for enhancing accuracy in image-guided needle insertion procedures.
• This approach facilitates the development of robust AI tools for ophthalmic surgery.