Microaneurysm Detection Using Principal Component Analysis and Machine Learning Methods

Diabetic retinopathy (DR) is an eye abnormality caused by long-term diabetes and it is the most common cause of blindness before the age of 50. Microaneurysms (MAs), resulting from leakage from retinal blood vessels, are early indicators of DR. In this paper, we analyzed MA detectability using small 25 by 25 pixel patches extracted from fundus images in the DIAbetic RETinopathy DataBase - Calibration Level 1 (DIARETDB1). Raw pixel intensities of extracted patches served directly as inputs into the following classifiers: random forest (RF), neural network, and support vector machine. We also explored the use of two techniques (principal component analysis and RF feature importance) for reducing input dimensionality. With traditional machine learning methods and leave-10-patients-out cross validation, our method outperformed a deep learning-based MA detection method, with AUC performance improved from 0.962 to 0.985 and F-measure improved from 0.913 to 0.926, using the same DIARETDB1 database. Furthermore, we validated our method on a different dataset-retinopathy online challenge (ROC) data set. The performance of the three classifiers and the pattern with different percentage of principal components are consistent on the two data sets. Especially, we trained the RF on DIARETDB1 and applied it to ROC; the performance is very similar to that of the RF trained and tested using cross validation on ROC data set. This result indicates that our method has the potential to generalize to different datasets.