Performance Evaluation of a Novel Retinal Proximity Detection System for Ophthalmic Surgical Guidance

Purpose: In vitreoretinal surgery, as surgeons manipulate sharp surgical instruments in close proximity to the retina, they could unintentionally damage it, potentially causing sight loss. This project’s goal is to provide real-time surgical guidance to minimize this complication. Methods: Two tracking methods are used: optical tracking with a stereo camera to locate the retina’s position and an electromagnetic sensor to track the surgical instrument. A rigid registration of the two reference systems allows to determine the distance between the tip of the instrument and the retina. A virtual three-dimensional environment, generated in real time, allows the surgeon to navigate through and see the procedure from different perspectives to that offered by the microscope’s stereo camera. An evaluation of the algorithm has been performed at the macroscale with a tracking volume size of 46x56x30 cm. Results: With an exemplar chosen ground truth distance of 15cm, the optical tracker proved to be able to detect distances with an average accuracy of 97.15%, with a mean absolute error of 4.28mm and a standard deviation of 1.5mm. The registered camera-sensor system showed an average distance detection accuracy of 92.84%, with a mean absolute error of 1.07 cm and a standard deviation of 1.068cm. The frame per seconds (fps) rate drops from 30fps to 11fps in a mid-range performance laptop, generating a negligible delay. The delay generated is expected to be lower on the final testing high-range performance computer due to faster image processing. The algorithm only runs on the Central Processing Unit (CPU), never using the Graphic Processing Unit (GPU). Conclusion: The experimental results showed that this approach has the potential to provide reliable real-time surgical guidance to inform the surgeon of potential risk during surgery. Experiments at the microscale with the surgical microscope will be performed next.