Towards Motion Compensation in Autonomous Robotic Subretinal Injections

TL;DR

This work tackles the challenge of accurate subretinal injections under ongoing retinal motion by introducing an OCT-based motion compensation pipeline that uses real-time $B^{5}$-scans to track the $Z$-axis position of retinal layers. The pipeline segments the ILM, RPE, and needle, constructs a surface point cloud, and commands end-effector motion along the $Z$-axis to keep the needle at a fixed retinal depth, implemented on the SHER 2.0 platform. Validation on ex vivo porcine eyes with sinusoidal retinal motion demonstrates the approach can track motion for larger amplitudes but reveals latency and drift as major limitations, with four of seven injections achieving bleb formation. The findings highlight the potential of depth-based robotic stabilization for safe automated subretinal interventions and point to predictive modeling as a key path to improved accuracy and safety in future work.

Abstract

Exudative (wet) age-related macular degeneration (AMD) is a leading cause of vision loss in older adults, typically treated with intravitreal injections. Emerging therapies, such as subretinal injections of stem cells, gene therapy, small molecules and RPE cells require precise delivery to avoid damaging delicate retinal structures. Robotic systems can potentially offer the necessary precision for these procedures. This paper presents a novel approach for motion compensation in robotic subretinal injections, utilizing real time Optical Coherence Tomography (OCT). The proposed method leverages B$^5$-scans, a rapid acquisition of small-volume OCT data, for dynamic tracking of retinal motion along the Z-axis, compensating for physiological movements such as breathing and heartbeat. Validation experiments on ex vivo porcine eyes revealed challenges in maintaining a consistent tool-to-retina distance, with deviations of up to 200 $μm$ for 100 $μm$ amplitude motions and over 80 $μm$ for 25 $μm$ amplitude motions over one minute. Subretinal injections faced additional difficulties, with phase shifts causing the needle to move off-target and inject into the vitreous. These results highlight the need for improved motion prediction and horizontal stability to enhance the accuracy and safety of robotic subretinal procedures.

Figures (7)

• Figure 1: OCT B-scan images showing the simulated breathing motion transmitted to the retina with the needle following the same motion while trying to keep a constant distance from it. Red and green lines represent the starting positions of the needle tip and retina respectively. Our motion compensation is based on comparing the current with the previously acquired volumes, so we see a latency in the needle's motion (scans 3 and 5). The needle continues to move in the direction opposite the retina. The pink and cyan lines are the internal limiting membrane (ILM) and retinal pigment epithelium (RPE) layers of the retina respectively.
• Figure 2: Our pipeline starts by acquiring B$^{5}$-scans (a), segmenting them using a deep neural network (b) and generating a surface point cloud based on the segmentation results (c). The median Z-axis position of the ILM points (green) in the point cloud is calculated and compared to the previous scan. Based on their difference, a velocity is applied to the robot end-effector and the process is repeated until it is terminated.
• Figure 3: The experimental setup includes, (a) the SHER 2.0, an OCT microscope, a piezo-actuated linear stage to generate a sinusoidal motion along the $Z$ direction (up and down) simulating effects of human breathing on the eye, a syringe pump, a syringe with a 42 gauge needle, and (b) a closer view of the open sky porcine eye in the 3D printed holder.
• Figure 4: Examples of the retinal detachment caused by the bleb (yellow) creation after successful subretinal injection. The needle is shown in red.
• Figure 5: Z-axis positions of the needle tip (blue) and linear stage controller (orange) over time.