Real-time Deformation-aware Control for Autonomous Robotic Subretinal Injection under iOCT Guidance

TL;DR

The paper addresses tissue deformation challenges in autonomous iOCT-guided subretinal injections by developing a real-time deformation-aware control framework that uses B$^{5}$-scans to continuously update a dynamic virtual target layer between the ILM and RPE. The method integrates a U-Net OCT segmentation network, 3D point-cloud reconstruction, RANSAC-based outlier removal, and retinal inpainting to accurately track the needle relative to the virtual target, with adaptive velocity control defined by $v_c$ and the distance to $L_{ ext{target}}$, enabling real-time operation at approximately 9 Hz. Validated on ex vivo open-sky porcine eyes, the approach achieved a final axial error of about $17\,\mathrm{\mu m}$ and a subretinal bleb generation success rate of around $90\%$, significantly improving over fixed-point targeting methods that yielded ~35% bleb success. The results demonstrate improved anatomical targeting and reliability for autonomous subretinal injections, with potential impact for safer, more effective retinal gene and cell therapies.

Abstract

Robotic platforms provide consistent and precise tool positioning that significantly enhances retinal microsurgery. Integrating such systems with intraoperative optical coherence tomography (iOCT) enables image-guided robotic interventions, allowing autonomous performance of advanced treatments, such as injecting therapeutic agents into the subretinal space. However, tissue deformations due to tool-tissue interactions constitute a significant challenge in autonomous iOCT-guided robotic subretinal injections. Such interactions impact correct needle positioning and procedure outcomes. This paper presents a novel method for autonomous subretinal injection under iOCT guidance that considers tissue deformations during the insertion procedure. The technique is achieved through real-time segmentation and 3D reconstruction of the surgical scene from densely sampled iOCT B-scans, which we refer to as B${^5}$-scans. Using B${^5}$-scans we monitor the position of the instrument relative to a virtual target layer between the ILM and RPE. Our experiments on ex vivo porcine eyes demonstrate dynamic adjustment of the insertion depth and overall improved accuracy in needle positioning compared to prior autonomous insertion approaches. Compared to a 35% success rate in subretinal bleb generation with previous approaches, our method reliably created subretinal blebs in 90% our experiments.

Figures (8)

• Figure 1: Tissue deformation during subretinal injection with a target (pink dashed line), defined as a virtual layer between ILM (green) and RPE (blue). In comparison, the fixed target point (yellow) is outside of the retina at the end of the insertion.
• Figure 2: Overview of our real-time needle insertion pipeline. (a) B$^{5}$-scan, consisting of five 2D B-scans, showing different cross sections of the needle and retina. (b) Segmented B-scans using our segmentation network, needle top surface (red), ILM (green), RPE (blue). (c) Point cloud generated from segmentation results without processing. (d) Point cloud generated from segmentation results after processing with inpainted ILM and RPE layers, removed outlier needle points and visualized virtual target layer (pink). (e) Needle tip area of the processed point cloud. Pink points are the virtual target layer, gray line represents the A-scan going through the needle tip. (f) Robot control adjustments based on needle position.
• Figure 3: Example input and ground truth pairs from our dataset. Needle top surface (red), ILM (green), RPE (blue).
• Figure 4: Front view diagram of the needle (gray), scan lines (blue) and their intersections corresponding to points in the point cloud (red and green). (a) Perfect alignment of scan lines and needle, parts of the needle visible in each scan line. (b) Average alignment, few scan lines do not contain needle. (c) Worst case alignment. In each case the point closest to the center of the needle higher than others (green). Worst case needle tip depth error is half needle thickness.
• Figure 5: Overview of our experimental setup. (a) Steady Hand Eye Robot He2012, (b) OCT imaging system (Proveo 8 with EnFocus, Leica, Germany), (c) Syringe pump (PHD2000, Harvard Apparatus, USA), (d) Syringe with 42 gauge needle (INCYTO Co., Ltd., South Korea), (e) Ex-vivo open-sky porcine eye on 3D printed container.