Classification of retinopathic injury using image cytometry and vasculature complexity

Retinopathic injuries are a common symptom of many diseases. However, if detected early, much of the damage caused by these injuries can be prevented, or in some cases reversed. In this study, images of retinas were classified as normal or injured using the vascular cell count, vasculature coverage, and vessel caliber. To model retinal vasculopathies, retinal vasculature from mice with the BCL-2 gene either partially or completely knocked out were compared. The bcl-2 gene is a critical regulator of apoptosis and angiogenesis, and therefore its absence has a significant impact on the number of vascular cells and vasculature complexity. When the aforementioned features were extracted from the images, classification was performed using a majority vote between a linear classifier, k-nearest-neighbors classification, and a support vector machine. This resulted in a classification accuracy of 81% using the “leave one out” error determination method.