3-D retinal curvature estimation

Real-time fundus reconstruction and intraocular mapping using an ophthalmic endoscope

BACKGROUND

Robotic ophthalmic endoscope holders allow surgeons to execute dual-hand operations in eye surgery. To prevent needle-like endoscopes from invading the retina when moving, surgeons expect visual and real-time information about the relative special relationship between the endoscope and fundus.

METHODS

This study develops a real-time fundus reconstruction method. First, using deep learning, the method estimates the distance between the fundus part corresponding to every pixel of the RGB endoscopic image and the endoscope. Then, by combining the estimated distance with the kinematics of a robotic holder, the point cloud representing the present fundus area is generated, and by which the size and position of the eyeball are estimated.

RESULTS

This method shows a real-time frequency of 10 Hz, which is robust to eyeball movement. The error of fundus reconstruction is about 0.5 mm, and the error of eyeball estimation is about 1 mm.

CONCLUSION

Using this fundus reconstruction method can map the position of the endoscope inside the eyeball when using a robotic endoscope holder in eye surgery. The overall accuracy level meets the ophthalmologists' accuracy requirements of ophthalmologists.

Glare-free retinal imaging using a portable light field fundus camera

We present the retinal plenoptoscope, a novel light field retinal imaging device designed to overcome many of the problems that limit the use of portable non-mydriatic fundus cameras, including image quality and lack of stereopsis. The design and prototype construction of this device is detailed and the ideal relationship between the eye pupil, system aperture stop and micro-image separation is investigated. A comparison of the theoretical entrance pupil size, multi-view baseline and depth resolution indicates that a higher degree of stereopsis is possible than with stereo fundus cameras. We also show that the effects of corneal backscatter on image quality can be removed through a novel method of glare identification and selective image rendering. This method is then extended to produce glare-free depth maps from densely estimated depth fields, creating representations of retinal topography from a single exposure. These methods are demonstrated on physical models and live human eyes using a prototype device based on a Lytro Illum consumer light field camera. The Retinal Plenoptoscope offers a viable, robust modality for non-mydriatic color and 3-D retinal imaging.

Retinal image registration under the assumption of a spherical eye

We propose a method for registering a pair of retinal images. The proposed approach employs point correspondences and assumes that the human eye has a spherical shape. The image registration problem is formulated as a 3D pose estimation problem, solved by estimating the rigid transformation that relates the views from which the two images were acquired. Given this estimate, each image can be warped upon the other so that pixels with the same coordinates image the same retinal point. Extensive experimental evaluation shows improved accuracy over state of the art methods, as well as robustness to noise and spurious keypoint matches. Experiments also indicate the method's applicability to the comparative analysis of images from different examinations that may exhibit changes and its applicability to diagnostic support.