Evaluation of Motion Artifact Associated with Use of a Custom, Handheld Imaging Device for Non-Invasive Retinal Blood Flow (RBF) Measurements in Telehealth Settings

Abstract

 

Purpose

 

To evaluate motion artifact during the use of a custom, non-mydriatic, handheld imaging device to non-invasively capture vessel-specific RBF using laser speckle contrast imaging (LSCI).

 

Methods

 

Five adult Wistar rats were anesthetized and laid on their side for retinal imaging. The custom device was held by the operator while stabilizing an elbow on the table during imaging. In each eye, two separate sets of 40 consecutive images of the same region using red laser (650 nm) illumination were acquired in rapid succession (at 100 frames/sec) and analyzed to calculate the translational shift across the image stack. Three points corresponding to prominent vascular features (branch points and vessel edges or termini) were selected on each image. For each point, the standard deviation from the mean of the magnitude of the motion across the image stack were calculated and compared to identify intra- and inter-session variability in the raw laser speckle images. An LSCI image was developed from each stack; three prominent features were compared to further evaluate inter-session translational motion.

 

Results

 

Intra- and inter-session variability due to motion artifact during handheld retinal imaging was low. Fig. 1A shows an image obtained from a raw laser speckle image stack. Each cluster (red, yellow, and pink) represents the three feature points across the full image stack, with the median (green) shown. Fig. 1B shows the amount of motion artifact based on the mean and standard deviation of the distance from the median value for each feature point for each set of image stacks across all imaging sessions. In 60% of the imaging sessions, 90% of the motion artifact was less than 8 pixels, with an overall motion artifact (mean of medians) of 5.43 ± 2.03 pixels. Motion artifact in the LSCI images (Fig. 2A), determined by the distance from the median of the mean difference between feature points, was 3.63 ± 2.21 pixels.

 

Conclusions

 

Mechanisms that compensate for translational motion of up to ±10 pixels will adequately address motion artifact and enable the custom, non-mydriatic, handheld device to obtain high-fidelity LSCI images for RBF measurements. Teleretinal screening of RBF has promise for the early diagnosis and longitudinal management of various ocular and systemic diseases.