Stereo reconstruction from microscopic images for computer-assisted ophthalmic surgery

Area of Science:

• Ophthalmic Surgery
• Computer-Assisted Surgery
• Medical Imaging

Background:

• Ophthalmic surgery requires precise 3D scene understanding, particularly depth perception.
• Corneal distortions and lack of texture in surgical scenes pose challenges for existing stereo reconstruction methods.

Purpose of the Study:

• To present a novel platform for stereo reconstruction in anterior segment ophthalmic surgery.
• To enhance scene understanding and depth perception for computer-assisted eye surgery.
• To address artifacts like lack of texture and corneal distortions.

Main Methods:

• A two-step approach: sparse 3D point cloud generation followed by dense 3D representation using surface fitting and anatomical priors.
• Pre-processing step to rectify corneal distortion artifacts using a 3D cornea geometry model and ray tracing.
• Utilizes microscopic images for 3D reconstruction.

Main Results:

• Accurate 3D reconstruction evaluated on ex vivo porcine eyes, phantom eyes, and synthetic images.
• Demonstrates enhanced scene understanding through an accurate 3D eye representation.
• Enables estimation of instrument-to-layer distances with a mean average error of 190 µm in porcine eyes.

Conclusions:

• Significant advancement in stereo reconstruction for ophthalmic surgery by addressing corneal distortions.
• Potential to improve surgical outcomes through intra-operative computer assistance.
• Enables novel tools like virtual distance sensors for enhanced surgical guidance.