Diabetic retinopathy (DR) is a leading cause of blindness amongst the working-age populace and is expected to rise in the coming years. DR is exceptionally difficult to detect in its early stages: by the time most patients seek medical consultation for vision loss, DR is already in its advanced stages. Catching DR early has the potential to prevent blindness in 90% of patients. ETDRS and ICDR are current systems in place to assess early DR. They rely on detecting microvascular changes in the retina, lack objective measurability (which leads to interobserver bias), and also are limited to the technologies and knowledge of possible clinical indicators of DR at the time of their creation. Sun et. al. attempted to use AI to quantify several potential indicators of DR - beyond those used in ETDRS and ICDR - and used receiver operating characteristic (ROC) analysis to determine the strength of the correlation. Individuals from different provinces in China were recruited provided that they fit the inclusion criteria with regards to age, Type 2 diabetes diagnostic criteria, and cognitive capabilities; any who had a history of mental illnesses, severe diseases, or eye surgery were excluded. The final sample size was 516 participants. These participants were taken to a completely dark room where a team trained by certified opthamologists took pictures focused on the macula of their eyes. The pictures were then processed and normalized for parameters such as color, brightness, and contrast in order to more extensively extract parameters such as microaneurysms and hemorrhaging. The EVision AI system was especially helpful due to its ability to extract local features of an image while also being able to map and capture spaces and longer distances, providing richer analysis of the pictures taken. The model works by first identifying a region of interest to extract data from, enhancing the image for more details, creating a probability map of predicted lesions, and then calculating the overall area and the position and number of predicted lesions. Of the 1012 eyes that were photographed, 90.0% of them had no DR, and the remaining 10% had varying levels of severity of DR. When comparing broadly across the different severities of DR, the differences in paramters such as microaneurysms and hemorrhage area (among other parameters) were statistically significant; even after adjustment for severity level and factors such as age, weight, height, BMI, smoking habits, hypertension, diabetes duration etc. the differences remained statistically signficant. Area under curve (AUC) analysis of the accuracy of DR prediction from different clinical parameters was also performed, and the AUC values of all parameters tested were quite high, all above 0.8. Even after adjustment for the aforementioned factors, the AUC values still all remained above 0.8, indicating that these parameters can be reliable predictors of DR onset. The clinical parameters hemorrhage area over retinal area and hemorrhage area over microaneurysm were surprisingly good at predicting DR onset, and maybe the parameters to monitor extensively for in the future.