Purpose
Previously, researchers looking to quantitatively assess RGC health were often required to manually define regions of interest (ROIs) for assessment of Brn3a or fluorogold -labelled retinal wholemounts, due to the impractibility and workload of performing whole retinal counts. In the present study, a segmentation algorithm is presented for the fully automatic determination of spatial distribution of retinal ganglion cells (RGCs) in retinal wholemounts, which enables the assessment of changes in RGC density and nearest neighbour distance (NND). This is applied to a model of rat ocular hypertension, where we test the assumption that initial RGC loss occurs in the retinal periphery.
Methods
A segmentation algorithm was designed to divide wholemount retinal images into 15 concentric rings or circles of 0.6mm increasing diameter. Each retina was further segmented into pre-defined quadrants. The area of each segment was automatically determined for each region, and RGC counts made using a previouly described algorithm. Mean RGC density and NND was next determined for each retinal area. This novel algorithm was applied to wholemounts from a rodent OHT model at 3 and 8 weeks (minimum n=3/time point, and compared to controls).
Results
The algorithm was easy to use and quick - the whole process taking less than 5 min per retina. Global mean NND increased significantly by 8.2 % between the control and 3 week OHT model (p= 0.0052), with a significant reduction in global RGC density (p= 0.0009). The whole circle and regional mean NND and density graphs indicate initial loss of RGCs occurring in the peripheral regions. No significant difference was found between the densities of the 3 week post OHT and the control eyes in the central 2 circles. However, the outer 2 circles of the retina displayed a significant decrease in density (p= 0.0043).
Conclusions
The lack of significant difference between control and 3 week mean NND in the central rings indicates that in the initial stages of OHT, RGC loss occurs predominantly in the periphery of the retina. Central loss of RGCs in our OHT model occurred therafter. This result supports the findings of existing studies on OHT, and suggests that our segmentation algorithm is an effective tool that can be applied to many other models involving retinal wholemount image analysis.