Over 200 Images of retina, obtained by Fundus imaging were acquired and analyzed. 100 of them were normal and tagged as 0, and 100 were affected by diabetic retinopathy and tagged as 1.
The images were preprocessed to highlight the sites of microaneurysm and suppress all other features.
These images were then used to train machine learning models based on SVM, Random Forest, Decision Tree and K Nearest Neighbors. The train and test data were split into 60-40, 70-30 and 80-20 combinations of train and test ratios. The best model was the random forest model obtained by training over 70% of data.
Will be shared soon.