Robotics and optical coherence tomography: current works and future perspectives [Invited]

Abstract

Optical coherence tomography (OCT) is an interferometric technique for micron-level imaging in biological and non-biological contexts. As a non-invasive, non-ionizing, and video-rate imaging modality, OCT is widely used in biomedical and clinical applications, especially ophthalmology, where it functions in many roles, including tissue mapping, disease diagnosis, and intrasurgical visualization. In recent years, the rapid growth of medical robotics has led to new applications for OCT, primarily for 3D free-space scanning, volumetric perception, and novel optical designs for specialized medical applications. This review paper surveys these recent developments at the intersection of OCT and robotics and organizes them by degree of integration and application, with a focus on biomedical and clinical topics. We conclude with perspectives on how these recent innovations may lead to further advances in imaging and medical technology.

Fig. 1.

Overview of the four Robot-OCT configurations. (A) Robot-adjacent OCT setup with a 1060 nm SS-OCT sample arm in the microscope and an adjacent robot-controlled needle system with a 6-axis robot arm. Adapted from [14]. (B) Seven-axis robot manipulator with an OCT-guided optical system mounted to its end-effector. Adapted from [57]. (C) An OCT-forceps probe integrated with and controlled by a 6-DOF parallel robot system. Adapted from [58]. (D) An endoscopic robot with an OCT probe for optimization-based visual servoing control. Adapted from [15].

Fig. 2.

Overview of robot-adjacent configurations for robot-guided medical applications. (A) A tabletop OCT system with the intraocular robotic interventional surgical system (IRISS) in cataract surgery. Adapted from [74]. The objective lens exhibits a focal length of $54$ mm and a $10\times 10$ mm field of view. (B) B-1 shows the dual-modality system with a tabletop OCT for robotic laser surgery with applications for tumor resection in neurosurgery. The tumor was imaged in a 5 mm $\times$ 5 mm region of interest with an imaging depth of approximately 2 mm. B-2 depicts an OCT image of the tumor upper surface. B-1 and B-2 are adapted from [64]. (C) Tabletop OCT system to collect data from a deformable phantom for a surgical pushing task performed by a Hexapod robot. Adapted from [62]. The OCT volume covers a field of view of 3 mm $\times$ 3 mm $\times$ 3.5 mm. (D) D-1 shows the tabletop OCT system platform for robot-controlled automated suturing. D-2 and D-3 depict OCT B-scan images before and after the wound edge closure from OCT-guided suturing. Adapted from [50]. (E) E-1 shows the tabletop OCT system to estimate tissue motion produced by a 6-DOF robot arm (OCT volume with a field of view of 5 mm $\times$ 5 mm $\times$ 3.5 mm). E-2 depicts the sequence of OCT images for motion estimation. E-1 and E-2 are adapted from [60]. (F) F-1 shows a 3D printed handle holding the surgical needle to an artificial anterior chamber based on a robot-OCT system with an IRB 120 industrial robot arm manipulator (ABB, Zurich, Switzerland). F-2 depicts an en face OCT view of a needle in cornea. F-3 describes the OCT B-scan image of the cross-sectional view along the needle’s axis. (The scalebar is adjusted for F-2 and F-3. F-1, F-2 and F-3 are adapted from [56]).

Fig. 3.

Overview of robot-mounted OCT systems with OCT scanner attached to 6-DOF or 7-DOF robot manipulator. (A) An integrated galvanometer-based OCT system attached to a 6-DOF robot arm with complementary sensors for face and pupil tracking. Adapted from [37]. (B) A 6-DOF robotic OCT system for inspection of monolithic storage devices and laser-based microsurgery tasks. Adapted from [35]. (C) A large-area robotically assisted optical coherence tomography (LARA-OCT) system with a 7-DOF robot arm and a 3.3MHz swept-source OCT sensor. Adapted from [43]. (D) Large field of view scanning with robot-mounted OCT system. Adapted from [38]. (E) An integrated OCT system with 6-DOF manipulator for large-area kidney scanning. Adapted from [10]. (F) A 6-DOF robot system equipped with a portable OCT sensor and a structured light camera for automated large-field scanning. Adapted from [20].

Fig. 4.

Roboticized OCT sensing tools and associated robotic systems. (A) A handheld 6-DOF micromanipulator with an integrated OCT scanner [114,115]. Adapted from [115]. (B) A Stewart-Gough robot-OCT platform for precise oblique injection. Adapted from [116]. (C) An 800 nm OCT microneedle for ultrahigh-resolution deep-brain imaging and laser ablation. Adapted from [24]. (D) Linearly actuated bi-manual surgical tools with OCT guided depth correction. Adapted from [113].

Fig. 5.

Endoscopic robotic-OCT systems and scanners. (A) The distal end of an OCT neuroendoscope continuum robot. Adapted from [125]. (B) An endoscopic probe controlled by a quartered piezoelectric tube and Lissajous scans generated from sinusoidal voltage control. Adapted from [54]. (C) A motor-free telerobotic OCT endoscope for high-resolution intraluminal imaging (Figure and annotation size adapted from [17]). (D) A pneumatic OCT endoscope for tortuous and narrow internal lumens. Adapted from [26]. (E) The distal end of a robot steerable OCT-guided catheter. Adapted from [126]. (F) System design of endoscopic OCT system and high resolution imaging of a bioresorbable vascular scaffold implanted in swine coronary artery. Adapted from [25].

Fig. 6.

Machine learning models for fundamental robotics problems with robotic OCT systems. (A) The flow diagram of the reinforcement learning framework for robotic needle insertion system (ex vivo human cornea target). Adapted from [56]. (B) An example of the OCT B-scan image segmentation by using the U-net deep learning architecture. Adapted from [65]. (C) A hexapod robot for pose estimation using deep learning models and volumetric OCT data. Adapted from [146]. (D) A multi-input deep learning framework for force estimation by using the shear wave elastography and volumetric OCT data. Adapted from [63]. (E) OCT image motion estimation with a deep learning architecture adjusted from the convolutional neural network modules via the temporal OCT volumes. Adapted from [60].