Abstract
Vitreoretinal surgery, requiring precise microscale tissue manipulation, is well-suited for robotic assistance. Image registration enhances surgeons' visual perception by aligning high-resolution preoperative OCT images with the intraoperative environment, improving visibility of anatomical features not seen in microscope images. However, optical distortions from the cornea, lens, eye curvature, and scanning patterns challenge the use of diagnostic data in robotic navigation. This study introduces a novel technique for curvature-corrected retinal registration, integrating diagnostic OCT with instrument-integrated OCT. The pipeline comprises feature extraction, curvature correction, initial alignment, and fine registration. Experiments using an artificial model eye and ex vivo porcine eye validate the method. Curvature correction achieves accuracy comparable to existing methods, with deviations of 17 [Formula: see text]m for the model eye and 460 [Formula: see text]m for the porcine eye. Post-registration, the fiducial marker error reduces to 103 [Formula: see text]m for the model eye and 318 [Formula: see text]m for the porcine eye. Our method provides intraoperative diagnostic context, enabling reliable topological assistance in retinal robotic systems.