Abstract
Purpose:
Vital to expanding the clinical utility of adaptive optics (AO) ophthalmic imaging is the development of robust geometrical metrics for describing the photoreceptor mosaic. This requires a firm theoretical understanding of the strengths and limitations of various metrics as well as empirical data on the sensitivity of individual metrics and the relationships between metrics. Here we explore the relationship between individual metrics as a potential space to identify abnormalities across subjects. In particular, we focus on the effects of diffuse and focal photoreceptor loss.
Methods:
Between 10 and 90 regions of interest were extracted from 14 normal confocal and split-detector AO scanning light ophthalmoscope photoreceptor montages, and cone locations were identified using a previously described algorithm. Density, nearest neighbor distance (NND), inter-cell distance (ICD), furthest distance (FD), percent six-neighbor cells, nearest neighbor regularity (NNR), number of neighbors regularity (NoNR), and Voronoi cell area regularity (VCAR) were extracted from coordinates with bound Voronoi regions. Confidence intervals (CIs) with 95% significance were calculated for each of the metric relationships. Between 5-80% of cones were removed to simulate diffuse and focal cell loss.
Results:
NND, ICD, and FD fell outside their CIs at 50, 80 and 30% diffuse cell loss, respectively. Percent six-neighbor cells, NNR, NoNR, and VCAR fell outside their CIs at 10, 30, 60 and 10% cone loss, respectively. All regularity metrics were more sensitive to cell loss at lower densities. VCAR was most sensitive to focal cell loss, with its values falling outside the CI at 5% loss. ICD was sensitive to focal changes in high-density mosaics above 15% focal loss, whereas FD fell outside its CI at 31% focal loss in only low-density mosaics. NND, NoNR, NNRI, and percent six-neighbor cells were insensitive to any focal cell loss.
Conclusions:
Spacing metrics are insensitive to cell undersampling, which enables estimating mosaic spacing in cases where every cell cannot be reliably identified. However, these same metrics are unable to detect early pathology. Conversely, regularity metrics are highly sensitive to both focal and diffuse cell loss, but require accurate cell identification. Thus, both types of metrics may be needed in combination to provide complete and accurate assessments of mosaic integrity.