Unsupervised out-of-distribution detection for safer robotically guided retinal microsurgery.

PURPOSE: A fundamental problem in designing safe machine learning systems is identifying when samples presented to a deployed model differ from those observed at training time. Detecting so-called out-of-distribution (OoD) samples is crucial in safety-critical applications such as robotically guided retinal microsurgery, where distances between the instrument and the retina are derived from sequences of 1D images that are acquired by an instrument-integrated optical coherence tomography (iiOCT) probe. METHODS: This work investigates the feasibility of using an OoD detector to identify when images from the iiOCT probe are inappropriate for subsequent machine learning-based distance estimation. We show how a simple OoD detector based on the Mahalanobis distance can successfully reject corrupted samples coming from real-world ex vivo porcine eyes. RESULTS: Our results demonstrate that the proposed approach can successfully detect OoD samples and help maintain the performance of the downstream task within reasonable levels. MahaAD outperformed a supervised approach trained on the same kind of corruptions and achieved the best performance in detecting OoD cases from a collection of iiOCT samples with real-world corruptions. CONCLUSION: The results indicate that detecting corrupted iiOCT data through OoD detection is feasible and does not need prior knowledge of possible corruptions. Consequently, MahaAD could aid in ensuring patient safety during robotically guided microsurgery by preventing deployed prediction models from estimating distances that put the patient at risk.

Randomised controlled trial on robot-assisted versus manual surgery for pucker peeling.

BACKGROUND: The aim was to explore the feasibility and safety of performing common surgical steps in epiretinal membrane (ERM) peeling using the Preceyes Surgical System (PSS). METHODS: In a tertiary centre, 15 pseudophakic patients with an idiopathic ERM were randomised to robot-assistance or manual surgery in a 2:1 ratio. In the robot-assisted group, the following steps were performed using PSS: (1) staining the internal limiting membrane (ILM), (2) removal of the dye, (3) creating an ILM flap, (4) completing the peeling, (5) holding a light pipe and (6) fluid-air exchange. Primary outcome measures were feasibility and safety. Secondary outcome measures were duration, best-corrected visual acuity (BCVA) and central retinal thickness (CRT). Moreover, the distance travelled by the instrument during peeling was assessed using motion tracking software. RESULTS: All steps performed with PSS were feasible with no clinical adverse events or complications. The surgical time was longer in the robot-assisted group (mean 56 min, SD = 12 vs. 24 min, SD = 5). During the study, the duration of robot-assisted surgeries decreased from 72 to 46 min. The distance travelled by the forceps was shorter in the robot-assisted group (mean 403 mm, SD = 186 vs. 550 mm, SD = 134). BCVA and CRT improved equally in both groups. CONCLUSIONS: This is the world's first randomised controlled trial on robotic surgery for ERM. Although more time-consuming, we found that several surgical steps were feasible with assistance of the PSS.

Synthesis of Tetrapeptides Containing Dehydroalanine, Dehydrophenylalanine and Oxazole as Building Blocks for Construction of Foldamers and Bioinspired Catalysts.

The incorporation of dehydroamino acid or fragments of oxazole into peptide chain is accompanied by a distorted three-dimensional structure and additionally enables the introduction of non-typical side-chain substituents. Thus, such compounds could be building blocks for obtaining novel foldamers and/or artificial enzymes (artzymes). In this paper, effective synthetic procedures leading to such building blocks-tetrapeptides containing glycyldehydroalanine, glycyldehydrophenylalanine, and glycyloxazole subunits-are described. Peptides containing serine were used as substrates for their conversion into peptides containing dehydroalanine and aminomethyloxazole-4-carboxylic acid while considering possible requirements for the introduction of these fragments into long-chain peptides at the last steps of synthesis.

First-in-Human Robot-Assisted Subretinal Drug Delivery Under Local Anesthesia.

PURPOSE: To report the results of a first-in-human study using a robotic device to assist subretinal drug delivery in patients undergoing vitreoretinal surgery for macular hemorrhage. DESIGN: Double-armed, randomized controlled surgical trial (ClinicalTrials.gov identifier: NCT03052881). METHODS: The study was performed at the Oxford Eye Hospital, Oxford University Hospitals NHS Foundation Trust, Oxford, United Kingdom. In total, 12 participants were recruited-6 in the robot-assisted and 6 in the control manual surgery arm according to the prespecified inclusion and exclusion criteria. All subjects presented with acute loss of vision owing to a subfoveal hemorrhage secondary to neovascular age-related macular degeneration. After standard vitrectomy, intraoperative optical coherence tomography-guided subretinal injection of tissue plasminogen activator (TPA) was performed by either robot-assisted or conventional manual technique under local anesthesia. The robotic part of the procedure involved advancement of a cannula through the retina and stabilizing it during foot-controlled injection of up to 100 µL of TPA solution. We assessed surgical success, duration of surgery, adverse events, and tolerability of surgery under local anesthesia. RESULTS: The procedure was well tolerated by all participants and safely performed in all cases. Total duration of surgery, time taken to complete the injection, and retinal microtrauma were similar between the groups and not clinically significant. Subretinal hemorrhage was successfully displaced at 1 month postintervention, except for 1 control subject, and the median gain in visual acuity was similar in both arms. CONCLUSIONS: This first-in-human study demonstrates the feasibility and safety of high-precision robot-assisted subretinal drug delivery as part of the surgical management of submacular hemorrhage, simulating its potential future application in gene or cell therapy.

Clinical Evaluation of an Instrument-Integrated OCT-Based Distance Sensor for Robotic Vitreoretinal Surgery.

Purpose: To assess the efficacy of an instrument-integrated OCT (iiOCT)-based distance sensor during robotic vitreoretinal surgery using the Preceyes Surgical System (PSS; Preceyes B.V.). Design: Single-center interventional study. Participants: Patients requiring vitreoretinal surgery. Methods: Five patients were enrolled. Standard preoperative OCT images were obtained. After vitrectomy, a predefined set of actions was performed using the iiOCT-based sensor. Images then were processed to assess the signal-to-noise ratio (SNR) at various angles to the retina and at different distances between the instrument tip and the retinal surface. Preoperative and intraoperative OCT images were compared qualitatively and quantitatively. Main Outcomes Measures: The feasibility in performing surgical tasks using the iiOCT-based sensor during vitreoretinal surgery, the SNR when imaging the retina, differences among intraoperative and preoperative OCT images, and characteristics of intraoperative retinal movements detected with the iiOCT-based probe. Results: Surgeons were able to perform all the tasks but one. The PSS was able to maintain a fixed distance. The SNR of the iiOCT-based sensor signal was adequate to determine the distance to the retina and to control the PSS. Analysis of iiOCT-based sensor A-scans identified 3 clearly distinguishable retinal layers, including the inner retinal boundary and the interface at the retinal pigment epithelium-Bruch's membrane. Thickness values differed by less than 5% from that measured by preoperative OCT, indicating its accuracy. The Fourier analysis of iiOCT-based sensor recordings identified anteroposterior retinal movements attributed to heartbeat and respiration. Conclusions: This iiOCT-based sensor was tested successfully and promises reliable use during robot-assisted surgery. An iiOCT-based sensor is a promising step toward OCT-guided robotic retinal surgery.

A comparison of robotic and manual surgery for internal limiting membrane peeling.

PURPOSE: To compare the Preceyes Surgical Robotic System (Eindhoven, Netherlands) to manual internal limiting membrane (ILM) peeling using the Eyesi surgical simulator (VRmagic, Mannheim, Germany) as the operative platform. METHODS: A comparative study was carried out with surgeons initially performing ILM peeling manually and then with the robot. Twenty-three vitreoretinal surgeons agreed to participate and all consented to the use of their surgical data from the Eyesi surgical simulator. Surgeons were given a 5-min demonstration of the devices and were allowed to practice for 10 min before attempting the membrane peel. Initially, the peel was performed manually and afterwards, this was repeated using the robot-controlled forceps. Surgical simulator outcome measures were compared between approaches. RESULTS: The average time required for the procedure was 5 min for the manual approach and 9 min with the robot (paired t test, p = 0.002). Intraocular instrument movement was reduced by half with the robot. On average 344 mm was required to complete the ILM peeling with the robot compared with 600 mm using the manual approach (paired t test, p = 0.002). There were fewer macular retinal hemorrhages with the robot: 53 with manual surgery, 32 with the robot (Mann-Whitney U test, p = 0.035). Retinal injuries were eliminated with the robot. CONCLUSIONS: Intraocular robotic surgery is still in its infancy and validation work is needed to understand the potential benefits and limitations of emerging technologies. Safety enhancements over current techniques may be possible and could lead to the broader adoption of robotic intraocular surgery in the future.

Human/robotic interaction: vision limits performance in simulated vitreoretinal surgery.

PURPOSE: Compare accuracy and precision in XYZ of stationary and dynamic tasks performed by surgeons with and without the use of a tele-operated robotic micromanipulator in a simulated vitreoretinal environment. The tasks were performed using a surgical microscope or while observing a video monitor. METHOD: Two experienced and two novice surgeons performed tracking and static tasks at a fixed depth with hand-held instruments on a Preceyes Surgical System R0.4. Visualization was through a standard microscope or a video display. The distances between the instrument tip and the targets (in µm) determined tracking errors in accuracy and precision. RESULTS: Using a microscope, dynamic or static accuracy and precision in XY (planar) movements were similar among test subjects. In Z (depth) movements, experience lead to more precision in both dynamic and static tasks (dynamic 35 ± 14 versus 60 ± 37 µm; static 27 ± 8 versus 36 ± 10 µm), and more accuracy in dynamic tasks (58 ± 35 versus 109 ± 79 µm). Robotic assistance improved both precision and accuracy in Z (1-3 ± 1 µm) in both groups. Using a video screen in combination with robotic assistance improved all performance measurements and reduced any differences due to experience. CONCLUSIONS: Robotics increases precision and accuracy, with greater benefit observed in less experienced surgeons. However, human control was a limiting factor in the achieved improvement. A major limitation was visualization of the target surface, in particular in depth. To maximize the benefit of robotic assistance, visualization must be optimized.

Release of experimental retinal vein occlusions by direct intraluminal injection of ocriplasmin.

PURPOSE: Retinal vein occlusions (RVO) are a major cause of vision loss in people aged 50 years and older. Current therapeutic options limit the consequences of RVO but do not eliminate the cause. Cannulation of the involved vessel and removal of the clot may provide a more permanent solution with a less demanding follow-up. However, cannulation of smaller retinal veins remains challenging. This paper explores the use of ocriplasmin (recombinant plasmin without its kringles) to clear RVO, using a robotic micromanipulator. METHODS: Branch RVO were induced in a porcine model with rose bengal followed by 532 nm endolaser to the superior venous branch of the optic nerve. The vein was cannulated proximal to the occlusion or beyond the first branching vessel from the obstruction. The vein was infused with a physiologic citric acid buffer solution (CAM) or CAM/ocriplasmin. The time of cannulation, number of attempts, and the ability to release the thrombus were recorded. RESULTS: Cannulation and infusion was possible in all the cases. The use of a micromanipulator allowed for a consistent cannulation of the retinal vein and positional stability allowed the vein to remain cannulated for up to 20 min. In none of the attempts (5/5) with CAM did the thrombus dissolve, despite repeat infusion/relaxation cycles. In 7/7 injections of CAM/ocriplasmin near to the point of obstruction, the clot started to dissolve within a few minutes of injection. An infusion, attempted beyond the first venous branch point proximal to the clot, was unsuccessful in 2/3 attempts. CONCLUSIONS: Ocriplasmin is effective in resolving RVO if injected close to the site of occlusion with the use of a micromanipulator.

Robotic Assisted Cannulation of Occluded Retinal Veins.

PURPOSE: To develop a methodology for cannulating porcine retinal venules using a robotic assistive arm after inducing a retinal vein occlusion using the photosensitizer rose bengal. METHODOLOGY: Retinal vein occlusions proximal to the first vascular branch point were induced following intravenous injection of rose bengal by exposure to 532nm laser light delivered by slit-lamp or endolaser probe. Retinal veins were cannulated by positioning a glass catheter tip using a robotically controlled micromanipulator above venules with an outer diameter of 80µm or more and performing a preset piercing maneuver, controlled robotically. The ability of a balanced salt (BSS) solution to remove an occlusion by repeat distention of the retinal vein was also assessed. RESULTS: Cannulation using the preset piercing program was successful in 9 of 9 eyes. Piercing using the micromanipulator under manual control was successful in only 24 of 52 attempts, with several attempts leading to double piercing. The best location for cannulation was directly proximal to the occlusion. Infusion of BSS did not result in the resolution of the occlusion. CONCLUSION: Cannulation of venules using a robotic microassistive arm can be achieved with consistency, provided the piercing is robotically driven. The model appears robust enough to allow testing of therapeutic strategies aimed at eliminating a retinal vein thrombus and its evolution over time.