Advanced image processing for visible-light OCT oximetry in rodents

Abstract

Purpose : Visible-light optical coherence tomography (vis-OCT) can perform retinal oximetry, but currently relies upon manual vessel segmentation. We present an automatic image processing technique to extract the oxygen saturation of hemoglobin (sO₂) from circumpapillary vis-OCT scans in rodents.

Methods : Rodent imaging procedures were approved by the Northwestern University IACUC and conformed to the ARVO Statement on Animal Research. Brown Norway rats were anesthetized with 3% isoflurane for 3 minutes and intramuscular injection of a ketamine (0.37 mg/kg) and xylazine (0.07 mg/kg) cocktail. Drops of 1% tropicamide hydrochloride ophthalmic solution and commercial artificial tears were applied for pupil dilation and prevention of corneal dehydration, respectively. Animals were imaged using a custom-built vis-OCT system, designed for rodent imaging.
For the imaging protocol, a raster image is first acquired to confirm proper positioning of the optic nerve head. We next performed repeated circumpapillary scans (up to 100 scans). To determine vessel locations automatically, a peak-finding algorithm was applied to the one-dimensional shadowgram that was created for each B-scan. A graph-search algorithm was applied to determine the location of the posterior vessel wall for oxygen saturation calculation. Root mean square error (RMSE), mean error (ME), and standard deviation (SD) were used as metrics of the accuracy, bias, and precision, respectively.

Results : We compared our automatic segmentation technique with and without graph-search segmentation and found that the automatic technique had better sO₂ accuracy (0.02 vs 0.06 for 20 B-scans), reduced bias (-0.02 vs. -0.06), but similar overall precision (0.18 vs. 0.18). Average arterial and venous sO₂ values were 0.96 ± 0.01 and 0.74 ± 0.03, respectively. All the R² values for the sO₂ fitting were above 0.8 indicating that our sO₂ fitting model predicted the data well.

Conclusions : We developed an automatic image processing method to extract the retinal oxygen saturation from circumpapillary vis-OCT scans and demonstrated its improved performance over conventional techniques.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.