Robotic Optical Coherence Tomography of Human Subjects with Posture-Invariant Head and Eye Alignment in Six Degrees of Freedom

Abstract

Ophthalmic optical coherence tomography (OCT) has achieved remarkable clinical success but remains sequestered in ophthalmology specialty offices. Recently introduced robotic OCT systems seek to expand patient access but fall short of their full potential due to significant imaging workspace and motion planning restrictions. Here, we present a next-generation robotic OCT system capable of imaging in any head orientation or posture that is mechanically reachable. This system overcomes prior restrictions by eliminating fixed-base tracking components, extending robot reach, and planning alignment in six degrees of freedom. With this robotic system, we show repeatable subject imaging independent of posture (standing, seated, reclined, and supine) under widely varying head orientations for multiple human subjects. For each subject, we obtained a consistent view of the retina, including the fovea, retinal vasculature, and edge of the optic nerve head. We believe this robotic approach can extend OCT as an eye disease screening, diagnosis, and monitoring tool to previously unreached patient populations.

Keywords: Medical robotics; motion stabilization; optical coherence tomography.

Fig. 1.

Second-generation robotically-aligned OCT system (a) using a UR5e robot arm for 70% longer reach. Scanner-integrated face tracking cameras (b, c) and the tracking workspace (green/blue shaded) move with the robot to eliminate head orientation requirements.

Fig. 2.

Face tracking using synthetic frontal face view. Point clouds derived from left (a) and right (b) grayscale and depth images are projected onto the synthetic camera image plane to form composite grayscale and depth images (c). Face detection and landmarking (d) is then performed on the composite grayscale image.

Fig. 3.

Illustration of face alignment planning from the current (a) to the target (b) scanner position. The path moves the retinal scanner’s pupil pivot in a straight line (c) towards the estimated pupil position (d) while the robot end-effector smoothly re-orients (e) towards the front of the face.

Fig. 4.

Robotic alignment of OCT scanner with diverse postures (standing, seated, supine) and head orientations (roll, pitch, yaw). OCT imaging is possible for any reachable and collision-free alignment pose.

Fig. 5.

Illustration of left, include, and right pupil camera view with tracking results overlaid. The pupil position is triangulated whereas eye gaze is derived from Purkinje reflections of the eye illumination LEDs.

Fig. 6.

Registered summed voxel projections, non-averaged B-scans through the fovea, registered volumes, and retinal SNR of the eyes of two subjects (a and b) with five varied postures and head poses. From left to right, imaging configurations progressively add head yaw, pitch, and roll. Retinal anatomy is imaged consistently despite large variations in head pose and the presence of significant subject motion.

Fig. 7.

Registered summed voxel projections, averaged B-scans through the fovea, registered volumes, and retinal SNR of the eyes of two subjects (a and b) with four varied postures. Retinal anatomy is imaged consistently despite large variations in posture and, for the standing case, the presence of significant subject motion. Note that these imaging sessions occur before (a) and after (b) installation of the vertical translation stage for the robot.