Autonomous Needle Navigation in Subretinal Injections via iOCT

Abstract

Subretinal injection is an effective method for direct delivery of therapeutic agents to treat prevalent subretinal diseases. Among the challenges for surgeons are physiological hand tremor, difficulty resolving single-micron scale depth perception, and lack of tactile feedback. The recent introduction of intraoperative Optical Coherence Tomography (iOCT) enables precise depth information during subretinal surgery. However, even when relying on iOCT, achieving the required micron-scale precision remains a significant surgical challenge. This work presents a robot-assisted workflow for high-precision autonomous needle navigation for subretinal injection. The workflow includes online registration between robot and iOCT coordinates; tool-tip localization in iOCT coordinates using a Convolutional Neural Network (CNN); and tool-tip planning and tracking system using real-time Model Predictive Control (MPC). The proposed workflow is validated using a silicone eye phantom and ex vivo porcine eyes. The experimental results demonstrate that the mean error to reach the user-defined target and the mean procedure duration are within an acceptable precision range. The proposed workflow achieves a 100% success rate for subretinal injection, while maintaining scleral forces at the scleral insertion point below 15mN throughout the navigation procedures.

Keywords: Computer Vision for Medical Robotics; Medical Robots and Systems; Surgical Robotics: Planning.

Fig. 1.

Experimental setup of iOCT guided autonomous needle navigation for subretinal injection: (a) left, general view, (a) right, close view of the surgical area, (b) a silicone eye phantom (left), an ex vivo open-sky eye (middle), and an ex vivo intact eye (right).

Fig. 2.

6-step workflow , which includes experimental setup, Jacobian matrix calculation, target definition, needle navigation, subretinal injection, and tool retraction.

Fig. 3.

(a) Original top-down view of the ROI without the BIOM lens (see Fig. 1). (b) The top-down view of the silicone eye phantom with the BIOM lens. (c) The top-down view of an open-sky eye with the BIOM lens. (d) Illustration of the B-scan. (e) The top-down view of an intact eye with the BIOM lens. (f) An extreme case of the distortion caused by the BIOM lens when the needle is positioned at the edge of the field of view.

Fig. 4.

Basic structure of proposed convolutional neural network. (a)-(b) Input of the network. (a) Top-down image of the ROI. (b) B-scan image at the tooltip position, indicated by the red dashed line in (a). (c)-(d) Output coordinates of the network.

Fig. 5.

Examples of data augmentation techniques, which includes dropout, cropping, rotation, resizing, change of brightness, and other artifacts.

Fig. 6.

Overview of the framework: (a) ROS node communications for DDP-based MPC. (b) Illustration of RCM constraint (6). (c) Implementation of RCM constraint during the navigation process.

Fig. 7.

(a) The FBG-integrated tool. (b) The top-down image of the tool in an intact eye with lens removed. (c) The corresponding B-scan of the tool. The red points are the predicted tool-tip positions by the network.

Fig. 8.

Box plot analysis of errors for reaching the target in different eye models along XYZ axes.

Fig. 9.

Distributions of errors for different eye models along XYZ axes.

Fig. 10.

Six examples of successful subretinal injections. The first row is derived from open-sky eye experiments, while the second row is derived from intact eye experiments. Clear blebs are formed in all examples.

Fig. 11.

Experimental results of scleral force measurement: (a) The norm of scleral fore measured in step 2, 3, 4, and 6 for 10 trials. (b) The trial that defines the target right above the RPE layer. (c) The trial that defines the target around 100μm below the RPE layer.