Robotically Aligned and Automatically Controlled Systems for Retinal Optical Coherence Tomography

Abstract

Optical Coherence Tomography (OCT) has ushered in a new era in ophthalmology, offering a non-invasive imaging technique that has revolutionized the structural examination of the human eye. OCT achieves this transformative capability by providing highly detailed images of the retina, boasting remarkable micrometer-scale resolution. As a result, OCT has emerged as a cornerstone technology in the diagnosis and management of a wide range of ocular diseases. However, despite its numerous advantages, notable limitations persist in conventional OCT systems. These systems are typically bulky, immobile tabletop devices confined to specialized photography suites within eye care facilities. Moreover, their operation and alignment require the expertise of skilled ophthalmic technicians. Patients, too, must meet specific criteria, including the ability to maintain an upright seated position and follow precise instructions for using a chinrest and directing their gaze toward a fixation target. These collective requirements severely limit the accessibility of OCT in urgent and routine care settings, such as primary care clinics. This dissertation explores the potential of Robotically-Aligned OCT (RAOCT) as a groundbreaking solution to address the aforementioned limitations of conventional OCT systems. RAOCT represents a novel technology that aims to provide an automated imaging alternative, reducing the reliance on tabletop or operator-driven OCT systems. First, we tackle the crucial need for angular control in RAOCT. Angular control refers to the ability to adjust the angle of incidence of the OCT beam into the eye, which we do by tracking the gaze orientation of the subject's eye. This feature holds paramount significance in retinal OCT, as the angle of beam incidence directly determines the imaged region of interest in the retina. Traditional OCT relies on subjects to precisely position their eyes and direct their gaze toward a pre-calibrated fixation target. To capture a different region of interest in the eye, operators need to provide additional fixation targets and instruct subjects to shift their gaze towards them each time the region of interest changes. With automated angular control, RAOCT eliminates this requirement by aligning the scanner based on the tracked gaze wherever it may be, allowing for imaging without fixation targets and offering precise control over the region of interest. We leverage angular control in RAOCT to introduce retinal Simultaneous Localization and Mapping (SLAM). By controlling the region of interest through adjustments in the angle of beam incidence onto the eye, we successfully localize the acquired volumes relative to other retinal data. This capability enables real-time mapping of the retina while also providing a robust initial estimate required for high-quality montaging during post-processing. Secondly, we introduce an automatic focusing mechanism to compensate for refractive errors. To make RAOCT a versatile imaging technology capable of operating autonomously and accommodating a wide range of subjects, it needs the ability to automatically correct for defocus introduced by the subject's eyes. Traditional OCT systems rely on human operators to either manually adjust for defocus or trigger auto-focusing when supported by their system. Our development includes a system that not only automates the focusing process for OCT, surpassing existing methods in speed, but also autonomously triggers focusing by determining when the system is aligned and ready for retinal imaging. Finally, we address residual alignment errors in RAOCT stemming from image processing delays, I/O latency, and hardware imperfections. We introduce a novel sensor-driven Digital Motion Correction (DMC) algorithm, a significant departure from traditional motion-correction methods known for their fragility and reliance on precise initial estimates. DMC leverages high-resolution telemetry regarding eye position and hardware encoder data to remap individual A-Scans to their predicted locations within the volume.DMC achieves motion correction down to the accuracy of the eye tracking system, eliminating the need for fragile optimization algorithms. In summary, these contributions to RAOCT hold the promise of mitigating the limitations of the technology's initial prototype. By automating its operation and addressing current system shortcomings, RAOCT is poised to fulfill its potential in ophthalmic diagnostics, thus benefiting patients and clinicians alike.