Assessing adaptive optics scanning light ophthalmoscopy (AOSLO) image quality using a model eye phantom

Abstract

Purpose : To assess image quality across four different AOSLO systems and longitudinally within one system using a model eye phantom.

Methods : Six model eyes (17.5mm focal length, 305.5µm/deg) were constructed to house a 3D-printed retinal phantom. All phantoms were imaged in two separate sessions on two AOSLO systems (UCSF and MCW). One phantom was imaged on two additional systems (UW Madison and Marquette) and was further imaged at MCW weekly for six consecutive weeks. Using ImageJ, intensity profiles were generated of the central row of “cone” structures for two reflective patches on the phantom with spacings of either 4µm or 5µm. The Michelson contrast of these structures was measured and compared between all phantoms (MCW and UCSF), for one phantom longitudinally (MCW), and for one phantom across four systems. A third patch of “cone” structures on the phantom (3µm spacing) was assessed for overall resolution via counts of the “cones” across all images collected.

Results : For the six model eyes, contrast measures were not significantly different for the 4µm or 5µm patches using the MCW (p=0.85, p=0.47) or the UCSF (p=0.07, p=0.15) system (repeated measures one-way ANOVA). Likewise, no differences in contrast were observed in the longitudinal data for the one model eye imaged at MCW for either patch (p=0.11, p=0.13; repeated measures one-way ANOVA). Across all four systems, an ordinary one-way ANOVA revealed significantly different contrast measures for the 4µm patch (p=0.0002) but not the 5µm patch (p=0.07). Additionally, a Tukey’s post-hoc test revealed significant differences between UCSF and the other three systems (MCW p=0.0003; UW Madison p=0.0001; Marquette p=0.01). Of the 137 images in which the 3µm patch was analyzed, all “cones” could be resolved in 87 images (63.5%).

Conclusions : Image contrast was stable when assessed on one AOSLO system longitudinally, though there were significant differences in image contrast across AOSLO systems. This indicates the model eye phantom could be useful in monitoring system performance within a longitudinal trial as well as comparing system performance across sites. Images with poorer focus or lower signal-to-noise were more likely to have lower contrast and fewer “cones” resolved than expected across systems, suggesting that such images could be used to aid in site certification prior to patient imaging.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.