In vivo retinal imaging for fixational eye motion detection using a high-speed digital micromirror device (DMD)-based ophthalmoscope

High-resolution, ultrafast, wide-field retinal eye-tracking for enhanced quantification of fixational and saccadic motion

We introduce a novel, noninvasive retinal eye-tracking system capable of detecting eye displacements with an angular resolution of 0.039 arcmin and a maximum velocity of 300°/s across an 8° span. Our system is designed based on a confocal retinal imaging module similar to a scanning laser ophthalmoscope. It utilizes a 2D MEMS scanner ensuring high image frame acquisition frequencies up to 1.24 kHz. In contrast with leading eye-tracking technology, we measure the eye displacements via the collection of the observed spatial excursions for all the times corresponding a full acquisition cycle, thus obviating the need for both a baseline reference frame and absolute spatial calibration. Using this approach, we demonstrate the precise measurement of eye movements with magnitudes exceeding the spatial extent of a single frame, which is not possible using existing image-based retinal trackers. We describe our retinal tracker, tracking algorithms and assess the performance of our system by using programmed artificial eye movements. We also demonstrate the clinical capabilities of our system with in vivo subjects by detecting microsaccades with angular extents as small as 0.028°. The rich kinematic ocular data provided by our system with its exquisite degree of accuracy and extended dynamic range opens new and exciting avenues in retinal imaging and clinical neuroscience. Several subtle features of ocular motion such as saccadic dysfunction, fixation instability and abnormal smooth pursuit can be readily extracted and inferred from the measured retinal trajectories thus offering a promising tool for identifying biomarkers of neurodegenerative diseases associated with these ocular symptoms.

Modeling and optimization of closed-loop retinal motion tracking in scanning light ophthalmoscopy

A model of closed-loop retinal motion tracking in an adaptive optics scanning light ophthalmoscope (AOSLO) was built, and the tracking performance was optimized by minimizing the root-mean-square of residual motion. We started with an evaluation of the fidelity of the retinal motion measurement, and then analyzed the transfer function of the system and power spectral density of retinal motion from human eyes, to achieve optimal control gain and sampling frequency. The performance was further enhanced by incorporating retinal motion prediction during the period in which the slow scanner was retracing. After optimization, residual image motion performance was improved by 33% with a nearly 50% reduction in computational cost in comparison to our previous setup, reaching a 3 dB bandwidth of 15-17 Hz, which is close to the frame rate (∼21  fps) of this AOSLO system.

Embedded deep learning in ophthalmology: making ophthalmic imaging smarter

Deep learning has recently gained high interest in ophthalmology due to its ability to detect clinically significant features for diagnosis and prognosis. Despite these significant advances, little is known about the ability of various deep learning systems to be embedded within ophthalmic imaging devices, allowing automated image acquisition. In this work, we will review the existing and future directions for 'active acquisition'-embedded deep learning, leading to as high-quality images with little intervention by the human operator. In clinical practice, the improved image quality should translate into more robust deep learning-based clinical diagnostics. Embedded deep learning will be enabled by the constantly improving hardware performance with low cost. We will briefly review possible computation methods in larger clinical systems. Briefly, they can be included in a three-layer framework composed of edge, fog, and cloud layers, the former being performed at a device level. Improved egde-layer performance via 'active acquisition' serves as an automatic data curation operator translating to better quality data in electronic health records, as well as on the cloud layer, for improved deep learning-based clinical data mining.